fair.c 269.1 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
19 20
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
21
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
22
 */
23
#include "sched.h"
24 25 26

#include <trace/events/sched.h>

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

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

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

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

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

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

87
const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
88

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

99 100 101 102 103 104 105 106 107
#ifdef CONFIG_CFS_BANDWIDTH
/*
 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
 * each time a cfs_rq requests quota.
 *
 * Note: in the case that the slice exceeds the runtime remaining (either due
 * to consumption or the quota being specified to be smaller than the slice)
 * we will always only issue the remaining available time.
 *
108 109 110
 * (default: 5 msec, units: microseconds)
 */
unsigned int sysctl_sched_cfs_bandwidth_slice		= 5000UL;
111 112
#endif

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

121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
static inline void update_load_add(struct load_weight *lw, unsigned long inc)
{
	lw->weight += inc;
	lw->inv_weight = 0;
}

static inline void update_load_sub(struct load_weight *lw, unsigned long dec)
{
	lw->weight -= dec;
	lw->inv_weight = 0;
}

static inline void update_load_set(struct load_weight *lw, unsigned long w)
{
	lw->weight = w;
	lw->inv_weight = 0;
}

139 140 141 142 143 144 145 146 147
/*
 * Increase the granularity value when there are more CPUs,
 * because with more CPUs the 'effective latency' as visible
 * to users decreases. But the relationship is not linear,
 * so pick a second-best guess by going with the log2 of the
 * number of CPUs.
 *
 * This idea comes from the SD scheduler of Con Kolivas:
 */
148
static unsigned int get_update_sysctl_factor(void)
149
{
150
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
	unsigned int factor;

	switch (sysctl_sched_tunable_scaling) {
	case SCHED_TUNABLESCALING_NONE:
		factor = 1;
		break;
	case SCHED_TUNABLESCALING_LINEAR:
		factor = cpus;
		break;
	case SCHED_TUNABLESCALING_LOG:
	default:
		factor = 1 + ilog2(cpus);
		break;
	}

	return factor;
}

static void update_sysctl(void)
{
	unsigned int factor = get_update_sysctl_factor();

#define SET_SYSCTL(name) \
	(sysctl_##name = (factor) * normalized_sysctl_##name)
	SET_SYSCTL(sched_min_granularity);
	SET_SYSCTL(sched_latency);
	SET_SYSCTL(sched_wakeup_granularity);
#undef SET_SYSCTL
}

void sched_init_granularity(void)
{
	update_sysctl();
}

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

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204
static void __update_inv_weight(struct load_weight *lw)
{
	unsigned long w;

	if (likely(lw->inv_weight))
		return;

	w = scale_load_down(lw->weight);

	if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
		lw->inv_weight = 1;
	else if (unlikely(!w))
		lw->inv_weight = WMULT_CONST;
	else
		lw->inv_weight = WMULT_CONST / w;
}
205 206

/*
207 208 209 210
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
211
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
212 213 214 215 216
 * we're guaranteed shift stays positive because inv_weight is guaranteed to
 * fit 32 bits, and NICE_0_LOAD gives another 10 bits; therefore shift >= 22.
 *
 * Or, weight =< lw.weight (because lw.weight is the runqueue weight), thus
 * weight/lw.weight <= 1, and therefore our shift will also be positive.
217
 */
218
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
219
{
220 221
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
222

223
	__update_inv_weight(lw);
224

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

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

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

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


const struct sched_class fair_sched_class;
245

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

250
#ifdef CONFIG_FAIR_GROUP_SCHED
251

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

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

P
Peter Zijlstra 已提交
264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
/* Walk up scheduling entities hierarchy */
#define for_each_sched_entity(se) \
		for (; se; se = se->parent)

static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
{
	return p->se.cfs_rq;
}

/* runqueue on which this entity is (to be) queued */
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	return se->cfs_rq;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return grp->my_q;
}

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

		cfs_rq->on_list = 1;
	}
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->on_list) {
		list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
		cfs_rq->on_list = 0;
	}
}

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

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

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

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

375 376 377 378 379 380 381 382 383 384 385 386 387
static void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
	int se_depth, pse_depth;

	/*
	 * preemption test can be made between sibling entities who are in the
	 * same cfs_rq i.e who have a common parent. Walk up the hierarchy of
	 * both tasks until we find their ancestors who are siblings of common
	 * parent.
	 */

	/* First walk up until both entities are at same depth */
P
Peter Zijlstra 已提交
388 389
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

	while (se_depth > pse_depth) {
		se_depth--;
		*se = parent_entity(*se);
	}

	while (pse_depth > se_depth) {
		pse_depth--;
		*pse = parent_entity(*pse);
	}

	while (!is_same_group(*se, *pse)) {
		*se = parent_entity(*se);
		*pse = parent_entity(*pse);
	}
}

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

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

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


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

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

P
Peter Zijlstra 已提交
428 429 430 431 432 433 434 435 436 437 438 439 440 441
static inline struct cfs_rq *cfs_rq_of(struct sched_entity *se)
{
	struct task_struct *p = task_of(se);
	struct rq *rq = task_rq(p);

	return &rq->cfs;
}

/* runqueue "owned" by this group */
static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
{
	return NULL;
}

442 443 444 445 446 447 448 449
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

static inline void list_del_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

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

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

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

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

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

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

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

478
	return max_vruntime;
479 480
}

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

	return min_vruntime;
}

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

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

501 502
	u64 vruntime = cfs_rq->min_vruntime;

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

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

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

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

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

	/*
	 * Find the right place in the rbtree:
	 */
	while (*link) {
		parent = *link;
		entry = rb_entry(parent, struct sched_entity, run_node);
		/*
		 * We dont care about collisions. Nodes with
		 * the same key stay together.
		 */
548
		if (entity_before(se, entry)) {
549 550 551
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
552
			leftmost = false;
553 554 555 556
		}
	}

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

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

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

	if (!left)
		return NULL;

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

576 577 578 579 580 581 582 583 584 585 586
static struct sched_entity *__pick_next_entity(struct sched_entity *se)
{
	struct rb_node *next = rb_next(&se->run_node);

	if (!next)
		return NULL;

	return rb_entry(next, struct sched_entity, run_node);
}

#ifdef CONFIG_SCHED_DEBUG
587
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
588
{
589
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root);
590

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

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

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

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

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

614 615 616 617 618 619 620
#define WRT_SYSCTL(name) \
	(normalized_sysctl_##name = sysctl_##name / (factor))
	WRT_SYSCTL(sched_min_granularity);
	WRT_SYSCTL(sched_latency);
	WRT_SYSCTL(sched_wakeup_granularity);
#undef WRT_SYSCTL

621 622 623
	return 0;
}
#endif
624

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

	return delta;
}

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

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

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

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

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

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

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

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

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

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

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

704 705 706 707 708 709 710
	/*
	 * Tasks are intialized with full load to be seen as heavy tasks until
	 * they get a chance to stabilize to their real load level.
	 * Group entities are intialized with zero load to reflect the fact that
	 * nothing has been attached to the task group yet.
	 */
	if (entity_is_task(se))
711 712
		sa->runnable_load_avg = sa->load_avg = scale_load_down(se->load.weight);

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

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

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

721 722 723 724 725 726 727 728 729 730 731 732 733
/*
 * With new tasks being created, their initial util_avgs are extrapolated
 * based on the cfs_rq's current util_avg:
 *
 *   util_avg = cfs_rq->util_avg / (cfs_rq->load_avg + 1) * se.load.weight
 *
 * However, in many cases, the above util_avg does not give a desired
 * value. Moreover, the sum of the util_avgs may be divergent, such
 * as when the series is a harmonic series.
 *
 * To solve this problem, we also cap the util_avg of successive tasks to
 * only 1/2 of the left utilization budget:
 *
734
 *   util_avg_cap = (cpu_scale - cfs_rq->avg.util_avg) / 2^n
735
 *
736
 * where n denotes the nth task and cpu_scale the CPU capacity.
737
 *
738 739
 * For example, for a CPU with 1024 of capacity, a simplest series from
 * the beginning would be like:
740 741 742 743 744 745 746 747 748 749 750
 *
 *  task  util_avg: 512, 256, 128,  64,  32,   16,    8, ...
 * cfs_rq util_avg: 512, 768, 896, 960, 992, 1008, 1016, ...
 *
 * Finally, that extrapolated util_avg is clamped to the cap (util_avg_cap)
 * if util_avg > util_avg_cap.
 */
void post_init_entity_util_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct sched_avg *sa = &se->avg;
751 752
	long cpu_scale = arch_scale_cpu_capacity(NULL, cpu_of(rq_of(cfs_rq)));
	long cap = (long)(cpu_scale - cfs_rq->avg.util_avg) / 2;
753 754 755 756 757 758 759 760 761 762 763 764

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

			if (sa->util_avg > cap)
				sa->util_avg = cap;
		} else {
			sa->util_avg = cap;
		}
	}
765 766 767 768 769 770 771

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

784
	attach_entity_cfs_rq(se);
785 786
}

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

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

	if (unlikely(!curr))
		return;

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

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

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

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

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

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

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

	account_cfs_rq_runtime(cfs_rq, delta_exec);
835 836
}

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

842
static inline void
843
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
844
{
845 846 847 848 849 850 851
	u64 wait_start, prev_wait_start;

	if (!schedstat_enabled())
		return;

	wait_start = rq_clock(rq_of(cfs_rq));
	prev_wait_start = schedstat_val(se->statistics.wait_start);
852 853

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

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

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

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

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

	if (entity_is_task(se)) {
		p = task_of(se);
		if (task_on_rq_migrating(p)) {
			/*
			 * Preserve migrating task's wait time so wait_start
			 * time stamp can be adjusted to accumulate wait time
			 * prior to migration.
			 */
879
			__schedstat_set(se->statistics.wait_start, delta);
880 881 882 883 884
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

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

892
static inline void
893 894 895
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
896 897 898 899 900 901 902
	u64 sleep_start, block_start;

	if (!schedstat_enabled())
		return;

	sleep_start = schedstat_val(se->statistics.sleep_start);
	block_start = schedstat_val(se->statistics.block_start);
903 904 905 906

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

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

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

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

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

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

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

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

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

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

			trace_sched_stat_blocked(tsk, delta);

			/*
			 * Blocking time is in units of nanosecs, so shift by
			 * 20 to get a milliseconds-range estimation of the
			 * amount of time that the task spent sleeping:
			 */
			if (unlikely(prof_on == SLEEP_PROFILING)) {
				profile_hits(SLEEP_PROFILING,
						(void *)get_wchan(tsk),
						delta >> 20);
			}
			account_scheduler_latency(tsk, delta >> 10, 0);
		}
	}
958 959
}

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

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

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

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

	if (!schedstat_enabled())
		return;

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

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

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

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

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

1022 1023
#ifdef CONFIG_NUMA_BALANCING
/*
1024 1025 1026
 * Approximate time to scan a full NUMA task in ms. The task scan period is
 * calculated based on the tasks virtual memory size and
 * numa_balancing_scan_size.
1027
 */
1028 1029
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1030 1031 1032

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

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

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

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

	struct rcu_head rcu;
	unsigned long total_faults;
	unsigned long max_faults_cpu;
	/*
	 * Faults_cpu is used to decide whether memory should move
	 * towards the CPU. As a consequence, these stats are weighted
	 * more by CPU use than by memory faults.
	 */
	unsigned long *faults_cpu;
	unsigned long faults[0];
};

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

1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083
static unsigned int task_nr_scan_windows(struct task_struct *p)
{
	unsigned long rss = 0;
	unsigned long nr_scan_pages;

	/*
	 * Calculations based on RSS as non-present and empty pages are skipped
	 * by the PTE scanner and NUMA hinting faults should be trapped based
	 * on resident pages
	 */
	nr_scan_pages = sysctl_numa_balancing_scan_size << (20 - PAGE_SHIFT);
	rss = get_mm_rss(p->mm);
	if (!rss)
		rss = nr_scan_pages;

	rss = round_up(rss, nr_scan_pages);
	return rss / nr_scan_pages;
}

/* For sanitys sake, never scan more PTEs than MAX_SCAN_WINDOW MB/sec. */
#define MAX_SCAN_WINDOW 2560

static unsigned int task_scan_min(struct task_struct *p)
{
1084
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1085 1086 1087
	unsigned int scan, floor;
	unsigned int windows = 1;

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

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

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

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

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

	return max(smin, period);
}

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

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

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

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

		smax = max(smax, period);
	}

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

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

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

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

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

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

1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192
static void account_numa_enqueue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running += (p->numa_preferred_nid != -1);
	rq->nr_preferred_running += (p->numa_preferred_nid == task_node(p));
}

static void account_numa_dequeue(struct rq *rq, struct task_struct *p)
{
	rq->nr_numa_running -= (p->numa_preferred_nid != -1);
	rq->nr_preferred_running -= (p->numa_preferred_nid == task_node(p));
}

1193 1194 1195 1196 1197 1198 1199 1200 1201
/* Shared or private faults. */
#define NR_NUMA_HINT_FAULT_TYPES 2

/* Memory and CPU locality */
#define NR_NUMA_HINT_FAULT_STATS (NR_NUMA_HINT_FAULT_TYPES * 2)

/* Averaged statistics, and temporary buffers. */
#define NR_NUMA_HINT_FAULT_BUCKETS (NR_NUMA_HINT_FAULT_STATS * 2)

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

1207
/*
1208
 * The averaged statistics, shared & private, memory & CPU,
1209 1210 1211 1212 1213
 * occupy the first half of the array. The second half of the
 * array is for current counters, which are averaged into the
 * first set by task_numa_placement.
 */
static inline int task_faults_idx(enum numa_faults_stats s, int nid, int priv)
1214
{
1215
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1216 1217 1218 1219
}

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

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

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

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

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

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

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

	return faults;
}

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

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

	return faults;
}

1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277
/*
 * A node triggering more than 1/3 as many NUMA faults as the maximum is
 * considered part of a numa group's pseudo-interleaving set. Migrations
 * between these nodes are slowed down, to allow things to settle down.
 */
#define ACTIVE_NODE_FRACTION 3

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

1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

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

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

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

		/*
		 * On systems with a backplane NUMA topology, compare groups
		 * of nodes, and move tasks towards the group with the most
		 * memory accesses. When comparing two nodes at distance
		 * "hoplimit", only nodes closer by than "hoplimit" are part
		 * of each group. Skip other nodes.
		 */
		if (sched_numa_topology_type == NUMA_BACKPLANE &&
1315
					dist >= maxdist)
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342
			continue;

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

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

		score += faults;
	}

	return score;
}

1343 1344 1345 1346 1347 1348
/*
 * These return the fraction of accesses done by a particular task, or
 * task group, on a particular numa node.  The group weight is given a
 * larger multiplier, in order to group tasks together that are almost
 * evenly spread out between numa nodes.
 */
1349 1350
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1351
{
1352
	unsigned long faults, total_faults;
1353

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

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

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

1365
	return 1000 * faults / total_faults;
1366 1367
}

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

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1379 1380
		return 0;

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

1384
	return 1000 * faults / total_faults;
1385 1386
}

1387 1388 1389 1390 1391 1392 1393 1394
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
	struct numa_group *ng = p->numa_group;
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);
1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405
	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;
1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436

	/*
	 * 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;

	/*
1437 1438
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1439
	 */
1440 1441
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1442 1443 1444
		return true;

	/*
1445 1446 1447 1448 1449 1450
	 * 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)
1451
	 */
1452 1453
	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;
1454 1455
}

1456
static unsigned long weighted_cpuload(struct rq *rq);
1457 1458
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1459
static unsigned long capacity_of(int cpu);
1460

1461
/* Cached statistics for all CPUs within a node */
1462 1463
struct numa_stats {
	unsigned long load;
1464 1465

	/* Total compute capacity of CPUs on a node */
1466
	unsigned long compute_capacity;
1467

1468
	unsigned int nr_running;
1469
};
1470

1471 1472 1473 1474 1475
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1476 1477
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1478 1479 1480 1481 1482 1483

	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;
1484
		ns->load += weighted_cpuload(rq);
1485
		ns->compute_capacity += capacity_of(cpu);
1486 1487

		cpus++;
1488 1489
	}

1490 1491 1492 1493 1494
	/*
	 * 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.
	 *
1495
	 * We'll detect a huge imbalance and bail there.
1496 1497 1498 1499
	 */
	if (!cpus)
		return;

1500 1501 1502 1503
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

1504
	capacity = min_t(unsigned, capacity,
1505
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1506 1507
}

1508 1509
struct task_numa_env {
	struct task_struct *p;
1510

1511 1512
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1513

1514
	struct numa_stats src_stats, dst_stats;
1515

1516
	int imbalance_pct;
1517
	int dist;
1518 1519 1520

	struct task_struct *best_task;
	long best_imp;
1521 1522 1523
	int best_cpu;
};

1524 1525 1526
static void task_numa_assign(struct task_numa_env *env,
			     struct task_struct *p, long imp)
{
1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541
	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);
	}

1542 1543
	if (env->best_task)
		put_task_struct(env->best_task);
1544 1545
	if (p)
		get_task_struct(p);
1546 1547 1548 1549 1550 1551

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

1552
static bool load_too_imbalanced(long src_load, long dst_load,
1553 1554
				struct task_numa_env *env)
{
1555 1556
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567
	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;
1568

1569
	imb = abs(dst_load * src_capacity - src_load * dst_capacity);
1570

1571
	orig_src_load = env->src_stats.load;
1572
	orig_dst_load = env->dst_stats.load;
1573

1574
	old_imb = abs(orig_dst_load * src_capacity - orig_src_load * dst_capacity);
1575 1576 1577

	/* Would this change make things worse? */
	return (imb > old_imb);
1578 1579
}

1580 1581 1582 1583 1584 1585 1586
/*
 * Maximum NUMA importance can be 1998 (2*999);
 * SMALLIMP @ 30 would be close to 1998/64.
 * Used to deter task migration.
 */
#define SMALLIMP	30

1587 1588 1589 1590 1591 1592
/*
 * 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
 */
1593
static void task_numa_compare(struct task_numa_env *env,
1594
			      long taskimp, long groupimp, bool maymove)
1595 1596 1597
{
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1598
	long src_load, dst_load;
1599
	long load;
1600
	long imp = env->p->numa_group ? groupimp : taskimp;
1601
	long moveimp = imp;
1602
	int dist = env->dist;
1603

1604 1605 1606
	if (READ_ONCE(dst_rq->numa_migrate_on))
		return;

1607
	rcu_read_lock();
1608 1609
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1610 1611
		cur = NULL;

1612 1613 1614 1615 1616 1617 1618
	/*
	 * 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;

1619
	if (!cur) {
1620
		if (maymove && moveimp >= env->best_imp)
1621 1622 1623 1624 1625
			goto assign;
		else
			goto unlock;
	}

1626 1627 1628 1629
	/*
	 * "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
1630
	 * the value is, the more remote accesses that would be expected to
1631 1632
	 * be incurred if the tasks were swapped.
	 */
1633 1634 1635
	/* Skip this swap candidate if cannot move to the source cpu */
	if (!cpumask_test_cpu(env->src_cpu, &cur->cpus_allowed))
		goto unlock;
1636

1637 1638 1639 1640 1641 1642 1643
	/*
	 * If dst and source tasks are in the same NUMA group, or not
	 * in any group then look only at task weights.
	 */
	if (cur->numa_group == env->p->numa_group) {
		imp = taskimp + task_weight(cur, env->src_nid, dist) -
		      task_weight(cur, env->dst_nid, dist);
1644
		/*
1645 1646
		 * Add some hysteresis to prevent swapping the
		 * tasks within a group over tiny differences.
1647
		 */
1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
		if (cur->numa_group)
			imp -= imp / 16;
	} else {
		/*
		 * Compare the group weights. If a task is all by itself
		 * (not part of a group), use the task weight instead.
		 */
		if (cur->numa_group && env->p->numa_group)
			imp += group_weight(cur, env->src_nid, dist) -
			       group_weight(cur, env->dst_nid, dist);
		else
			imp += task_weight(cur, env->src_nid, dist) -
			       task_weight(cur, env->dst_nid, dist);
1661 1662
	}

1663
	if (maymove && moveimp > imp && moveimp > env->best_imp) {
1664
		imp = moveimp;
1665
		cur = NULL;
1666
		goto assign;
1667
	}
1668

1669 1670 1671 1672 1673 1674 1675 1676 1677
	/*
	 * 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;

1678 1679 1680
	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
1681 1682 1683 1684
	load = task_h_load(env->p) - task_h_load(cur);
	if (!load)
		goto assign;

1685 1686
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1687

1688
	if (load_too_imbalanced(src_load, dst_load, env))
1689 1690
		goto unlock;

1691
assign:
1692 1693 1694 1695
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1696 1697
	if (!cur) {
		/*
1698
		 * select_idle_siblings() uses an per-CPU cpumask that
1699 1700 1701
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1702 1703
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1704 1705
		local_irq_enable();
	}
1706

1707 1708 1709 1710 1711
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1712 1713
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1714
{
1715 1716
	long src_load, dst_load, load;
	bool maymove = false;
1717 1718
	int cpu;

1719 1720 1721 1722 1723 1724 1725 1726 1727 1728
	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);

1729 1730
	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
1731
		if (!cpumask_test_cpu(cpu, &env->p->cpus_allowed))
1732 1733 1734
			continue;

		env->dst_cpu = cpu;
1735
		task_numa_compare(env, taskimp, groupimp, maymove);
1736 1737 1738
	}
}

1739 1740 1741 1742
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1743

1744
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1745
		.src_nid = task_node(p),
1746 1747 1748 1749 1750

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1751
		.best_cpu = -1,
1752 1753
	};
	struct sched_domain *sd;
1754
	struct rq *best_rq;
1755
	unsigned long taskweight, groupweight;
1756
	int nid, ret, dist;
1757
	long taskimp, groupimp;
1758

1759
	/*
1760 1761 1762 1763 1764 1765
	 * 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.
1766 1767
	 */
	rcu_read_lock();
1768
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1769 1770
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1771 1772
	rcu_read_unlock();

1773 1774 1775 1776 1777 1778 1779
	/*
	 * 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)) {
1780
		sched_setnuma(p, task_node(p));
1781 1782 1783
		return -EINVAL;
	}

1784
	env.dst_nid = p->numa_preferred_nid;
1785 1786 1787 1788 1789 1790
	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;
1791
	update_numa_stats(&env.dst_stats, env.dst_nid);
1792

1793
	/* Try to find a spot on the preferred nid. */
1794
	task_numa_find_cpu(&env, taskimp, groupimp);
1795

1796 1797 1798 1799 1800 1801 1802
	/*
	 * 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.
	 */
1803
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1804 1805 1806
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1807

1808
			dist = node_distance(env.src_nid, env.dst_nid);
1809 1810 1811 1812 1813
			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);
			}
1814

1815
			/* Only consider nodes where both task and groups benefit */
1816 1817
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1818
			if (taskimp < 0 && groupimp < 0)
1819 1820
				continue;

1821
			env.dist = dist;
1822 1823
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1824
			task_numa_find_cpu(&env, taskimp, groupimp);
1825 1826 1827
		}
	}

1828 1829 1830 1831 1832 1833 1834 1835
	/*
	 * 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.
	 */
1836 1837 1838 1839
	if (p->numa_group) {
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
1840
			nid = cpu_to_node(env.best_cpu);
1841

1842 1843
		if (nid != p->numa_preferred_nid)
			sched_setnuma(p, nid);
1844 1845 1846 1847 1848
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1849

1850
	best_rq = cpu_rq(env.best_cpu);
1851
	if (env.best_task == NULL) {
1852
		ret = migrate_task_to(p, env.best_cpu);
1853
		WRITE_ONCE(best_rq->numa_migrate_on, 0);
1854 1855
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1856 1857 1858
		return ret;
	}

1859
	ret = migrate_swap(p, env.best_task, env.best_cpu, env.src_cpu);
1860
	WRITE_ONCE(best_rq->numa_migrate_on, 0);
1861

1862 1863
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1864 1865
	put_task_struct(env.best_task);
	return ret;
1866 1867
}

1868 1869 1870
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1871 1872
	unsigned long interval = HZ;

1873
	/* This task has no NUMA fault statistics yet */
1874
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1875 1876
		return;

1877
	/* Periodically retry migrating the task to the preferred node */
1878
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
1879
	p->numa_migrate_retry = jiffies + interval;
1880 1881

	/* Success if task is already running on preferred CPU */
1882
	if (task_node(p) == p->numa_preferred_nid)
1883 1884 1885
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1886
	task_numa_migrate(p);
1887 1888
}

1889
/*
1890
 * Find out how many nodes on the workload is actively running on. Do this by
1891 1892 1893 1894
 * 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.
 */
1895
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1896 1897
{
	unsigned long faults, max_faults = 0;
1898
	int nid, active_nodes = 0;
1899 1900 1901 1902 1903 1904 1905 1906 1907

	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);
1908 1909
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1910
	}
1911 1912 1913

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1914 1915
}

1916 1917 1918
/*
 * 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
1919 1920 1921
 * 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.
1922 1923
 */
#define NUMA_PERIOD_SLOTS 10
1924
#define NUMA_PERIOD_THRESHOLD 7
1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935

/*
 * 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;
1936
	int lr_ratio, ps_ratio;
1937 1938 1939 1940 1941 1942 1943 1944
	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
1945 1946 1947
	 * 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
1948
	 */
1949
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965
		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);
1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984
	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;
1985 1986 1987 1988 1989
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		/*
1990 1991 1992
		 * 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.
1993
		 */
1994 1995
		int ratio = max(lr_ratio, ps_ratio);
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;
1996 1997 1998 1999 2000 2001 2002
	}

	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));
}

2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
/*
 * Get the fraction of time the task has been running since the last
 * NUMA placement cycle. The scheduler keeps similar statistics, but
 * decays those on a 32ms period, which is orders of magnitude off
 * from the dozens-of-seconds NUMA balancing period. Use the scheduler
 * stats only if the task is so new there are no NUMA statistics yet.
 */
static u64 numa_get_avg_runtime(struct task_struct *p, u64 *period)
{
	u64 runtime, delta, now;
	/* Use the start of this time slice to avoid calculations. */
	now = p->se.exec_start;
	runtime = p->se.sum_exec_runtime;

	if (p->last_task_numa_placement) {
		delta = runtime - p->last_sum_exec_runtime;
		*period = now - p->last_task_numa_placement;
	} else {
2021
		delta = p->se.avg.load_sum;
2022
		*period = LOAD_AVG_MAX;
2023 2024 2025 2026 2027 2028 2029 2030
	}

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

	return delta;
}

2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077
/*
 * 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;
2078
		nodemask_t max_group = NODE_MASK_NONE;
2079 2080 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
		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. */
2112 2113
		if (!max_faults)
			break;
2114 2115 2116 2117 2118
		nodes = max_group;
	}
	return nid;
}

2119 2120
static void task_numa_placement(struct task_struct *p)
{
2121 2122
	int seq, nid, max_nid = -1;
	unsigned long max_faults = 0;
2123
	unsigned long fault_types[2] = { 0, 0 };
2124 2125
	unsigned long total_faults;
	u64 runtime, period;
2126
	spinlock_t *group_lock = NULL;
2127

2128 2129 2130 2131 2132
	/*
	 * 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:
	 */
2133
	seq = READ_ONCE(p->mm->numa_scan_seq);
2134 2135 2136
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2137
	p->numa_scan_period_max = task_scan_max(p);
2138

2139 2140 2141 2142
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2143 2144 2145
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2146
		spin_lock_irq(group_lock);
2147 2148
	}

2149 2150
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2151 2152
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2153
		unsigned long faults = 0, group_faults = 0;
2154
		int priv;
2155

2156
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2157
			long diff, f_diff, f_weight;
2158

2159 2160 2161 2162
			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);
2163

2164
			/* Decay existing window, copy faults since last scan */
2165 2166 2167
			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;
2168

2169 2170 2171 2172 2173 2174 2175 2176
			/*
			 * 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);
2177
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2178
				   (total_faults + 1);
2179 2180
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2181

2182 2183 2184
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2185
			p->total_numa_faults += diff;
2186
			if (p->numa_group) {
2187 2188 2189 2190 2191 2192 2193 2194 2195
				/*
				 * safe because we can only change our own group
				 *
				 * mem_idx represents the offset for a given
				 * nid and priv in a specific region because it
				 * is at the beginning of the numa_faults array.
				 */
				p->numa_group->faults[mem_idx] += diff;
				p->numa_group->faults_cpu[mem_idx] += f_diff;
2196
				p->numa_group->total_faults += diff;
2197
				group_faults += p->numa_group->faults[mem_idx];
2198
			}
2199 2200
		}

2201 2202 2203 2204 2205 2206 2207
		if (!p->numa_group) {
			if (faults > max_faults) {
				max_faults = faults;
				max_nid = nid;
			}
		} else if (group_faults > max_faults) {
			max_faults = group_faults;
2208 2209
			max_nid = nid;
		}
2210 2211
	}

2212
	if (p->numa_group) {
2213
		numa_group_count_active_nodes(p->numa_group);
2214
		spin_unlock_irq(group_lock);
2215
		max_nid = preferred_group_nid(p, max_nid);
2216 2217
	}

2218 2219 2220 2221
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);
2222
	}
2223 2224

	update_task_scan_period(p, fault_types[0], fault_types[1]);
2225 2226
}

2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237
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);
}

2238 2239
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2240 2241 2242 2243 2244 2245 2246 2247 2248
{
	struct numa_group *grp, *my_grp;
	struct task_struct *tsk;
	bool join = false;
	int cpu = cpupid_to_cpu(cpupid);
	int i;

	if (unlikely(!p->numa_group)) {
		unsigned int size = sizeof(struct numa_group) +
2249
				    4*nr_node_ids*sizeof(unsigned long);
2250 2251 2252 2253 2254 2255

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

		atomic_set(&grp->refcount, 1);
2256 2257
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2258
		spin_lock_init(&grp->lock);
2259
		grp->gid = p->pid;
2260
		/* Second half of the array tracks nids where faults happen */
2261 2262
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2263

2264
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2265
			grp->faults[i] = p->numa_faults[i];
2266

2267
		grp->total_faults = p->total_numa_faults;
2268

2269 2270 2271 2272 2273
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2274
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2275 2276

	if (!cpupid_match_pid(tsk, cpupid))
2277
		goto no_join;
2278 2279 2280

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2281
		goto no_join;
2282 2283 2284

	my_grp = p->numa_group;
	if (grp == my_grp)
2285
		goto no_join;
2286 2287 2288 2289 2290 2291

	/*
	 * 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)
2292
		goto no_join;
2293 2294 2295 2296 2297

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2298
		goto no_join;
2299

2300 2301 2302 2303 2304 2305 2306
	/* 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;
2307

2308 2309 2310
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2311
	if (join && !get_numa_group(grp))
2312
		goto no_join;
2313 2314 2315 2316 2317 2318

	rcu_read_unlock();

	if (!join)
		return;

2319 2320
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2321

2322
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2323 2324
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2325
	}
2326 2327
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2328 2329 2330 2331 2332

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

	spin_unlock(&my_grp->lock);
2333
	spin_unlock_irq(&grp->lock);
2334 2335 2336 2337

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2338 2339 2340 2341 2342
	return;

no_join:
	rcu_read_unlock();
	return;
2343 2344 2345 2346 2347
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2348
	void *numa_faults = p->numa_faults;
2349 2350
	unsigned long flags;
	int i;
2351 2352

	if (grp) {
2353
		spin_lock_irqsave(&grp->lock, flags);
2354
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2355
			grp->faults[i] -= p->numa_faults[i];
2356
		grp->total_faults -= p->total_numa_faults;
2357

2358
		grp->nr_tasks--;
2359
		spin_unlock_irqrestore(&grp->lock, flags);
2360
		RCU_INIT_POINTER(p->numa_group, NULL);
2361 2362 2363
		put_numa_group(grp);
	}

2364
	p->numa_faults = NULL;
2365
	kfree(numa_faults);
2366 2367
}

2368 2369 2370
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2371
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2372 2373
{
	struct task_struct *p = current;
2374
	bool migrated = flags & TNF_MIGRATED;
2375
	int cpu_node = task_node(current);
2376
	int local = !!(flags & TNF_FAULT_LOCAL);
2377
	struct numa_group *ng;
2378
	int priv;
2379

2380
	if (!static_branch_likely(&sched_numa_balancing))
2381 2382
		return;

2383 2384 2385 2386
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2387
	/* Allocate buffer to track faults on a per-node basis */
2388 2389
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2390
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2391

2392 2393
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2394
			return;
2395

2396
		p->total_numa_faults = 0;
2397
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2398
	}
2399

2400 2401 2402 2403 2404 2405 2406 2407
	/*
	 * 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);
2408
		if (!priv && !(flags & TNF_NO_GROUP))
2409
			task_numa_group(p, last_cpupid, flags, &priv);
2410 2411
	}

2412 2413 2414 2415 2416 2417
	/*
	 * 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.
	 */
2418 2419 2420 2421
	ng = p->numa_group;
	if (!priv && !local && ng && ng->active_nodes > 1 &&
				numa_is_active_node(cpu_node, ng) &&
				numa_is_active_node(mem_node, ng))
2422 2423
		local = 1;

2424 2425 2426 2427
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
2428 2429
	if (time_after(jiffies, p->numa_migrate_retry)) {
		task_numa_placement(p);
2430
		numa_migrate_preferred(p);
2431
	}
2432

I
Ingo Molnar 已提交
2433 2434
	if (migrated)
		p->numa_pages_migrated += pages;
2435 2436
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2437

2438 2439
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2440
	p->numa_faults_locality[local] += pages;
2441 2442
}

2443 2444
static void reset_ptenuma_scan(struct task_struct *p)
{
2445 2446 2447 2448 2449 2450 2451 2452
	/*
	 * 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:
	 */
2453
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2454 2455 2456
	p->mm->numa_scan_offset = 0;
}

2457 2458 2459 2460 2461 2462 2463 2464 2465
/*
 * 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;
2466
	u64 runtime = p->se.sum_exec_runtime;
2467
	struct vm_area_struct *vma;
2468
	unsigned long start, end;
2469
	unsigned long nr_pte_updates = 0;
2470
	long pages, virtpages;
2471

2472
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485

	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;

2486
	if (!mm->numa_next_scan) {
2487 2488
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2489 2490
	}

2491 2492 2493 2494 2495 2496 2497
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2498 2499
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
2500
		p->numa_scan_period = task_scan_start(p);
2501
	}
2502

2503
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2504 2505 2506
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2507 2508 2509 2510 2511 2512
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2513 2514 2515
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2516
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2517 2518
	if (!pages)
		return;
2519

2520

2521 2522
	if (!down_read_trylock(&mm->mmap_sem))
		return;
2523
	vma = find_vma(mm, start);
2524 2525
	if (!vma) {
		reset_ptenuma_scan(p);
2526
		start = 0;
2527 2528
		vma = mm->mmap;
	}
2529
	for (; vma; vma = vma->vm_next) {
2530
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2531
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2532
			continue;
2533
		}
2534

2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
		/*
		 * 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
2545 2546 2547 2548 2549 2550
		/*
		 * 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;
2551

2552 2553 2554 2555
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2556
			nr_pte_updates = change_prot_numa(vma, start, end);
2557 2558

			/*
2559 2560 2561 2562 2563 2564
			 * 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.
2565 2566 2567
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2568
			virtpages -= (end - start) >> PAGE_SHIFT;
2569

2570
			start = end;
2571
			if (pages <= 0 || virtpages <= 0)
2572
				goto out;
2573 2574

			cond_resched();
2575
		} while (end != vma->vm_end);
2576
	}
2577

2578
out:
2579
	/*
P
Peter Zijlstra 已提交
2580 2581 2582 2583
	 * 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.
2584 2585
	 */
	if (vma)
2586
		mm->numa_scan_offset = start;
2587 2588 2589
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600

	/*
	 * 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;
	}
2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625
}

/*
 * 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;

2626
	if (now > curr->node_stamp + period) {
2627
		if (!curr->node_stamp)
2628
			curr->numa_scan_period = task_scan_start(curr);
2629
		curr->node_stamp += period;
2630 2631 2632 2633 2634 2635 2636

		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);
		}
	}
}
2637

2638 2639 2640 2641 2642
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);

2643 2644 2645
	if (!static_branch_likely(&sched_numa_balancing))
		return;

2646 2647 2648
	if (!p->mm || !p->numa_faults || (p->flags & PF_EXITING))
		return;

2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668
	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);
2669 2670
}

2671 2672 2673 2674
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2675 2676 2677 2678 2679 2680 2681 2682

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)
{
}
2683

2684 2685 2686 2687
static inline void update_scan_period(struct task_struct *p, int new_cpu)
{
}

2688 2689
#endif /* CONFIG_NUMA_BALANCING */

2690 2691 2692 2693
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2694
	if (!parent_entity(se))
2695
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2696
#ifdef CONFIG_SMP
2697 2698 2699 2700 2701 2702
	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);
	}
2703
#endif
2704 2705 2706 2707 2708 2709 2710
	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);
2711
	if (!parent_entity(se))
2712
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2713
#ifdef CONFIG_SMP
2714 2715
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2716
		list_del_init(&se->group_node);
2717
	}
2718
#endif
2719 2720 2721
	cfs_rq->nr_running--;
}

2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762
/*
 * 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)
{
2763 2764 2765 2766
	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;
2767 2768 2769 2770 2771
}

static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2772 2773 2774 2775 2776
	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);
2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802
}

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

2803
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
2804
			    unsigned long weight, unsigned long runnable)
2805 2806 2807 2808 2809 2810 2811 2812 2813 2814
{
	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);

2815
	se->runnable_weight = runnable;
2816 2817 2818
	update_load_set(&se->load, weight);

#ifdef CONFIG_SMP
2819 2820 2821 2822 2823 2824 2825
	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);
2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841
#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]);

2842
	reweight_entity(cfs_rq, se, weight, weight);
2843 2844 2845
	load->inv_weight = sched_prio_to_wmult[prio];
}

2846
#ifdef CONFIG_FAIR_GROUP_SCHED
2847
#ifdef CONFIG_SMP
2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885
/*
 * 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
2886
 *   ge->load.weight = ----------------------------- = tg->weight   (4)
2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899
 *			    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
 *
2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911
 * 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)
2912 2913 2914 2915 2916 2917 2918 2919 2920
 *
 * 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!
 */
2921
static long calc_group_shares(struct cfs_rq *cfs_rq)
2922
{
2923 2924 2925 2926
	long tg_weight, tg_shares, load, shares;
	struct task_group *tg = cfs_rq->tg;

	tg_shares = READ_ONCE(tg->shares);
2927

2928
	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
2929

2930
	tg_weight = atomic_long_read(&tg->load_avg);
2931

2932 2933 2934
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2935

2936
	shares = (tg_shares * load);
2937 2938
	if (tg_weight)
		shares /= tg_weight;
2939

2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951
	/*
	 * 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.
	 */
2952
	return clamp_t(long, shares, MIN_SHARES, tg_shares);
2953
}
2954 2955

/*
2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980
 * 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).
2981 2982 2983
 */
static long calc_group_runnable(struct cfs_rq *cfs_rq, long shares)
{
2984 2985 2986 2987 2988 2989 2990
	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));
2991 2992 2993 2994

	runnable *= shares;
	if (load_avg)
		runnable /= load_avg;
2995

2996 2997
	return clamp_t(long, runnable, MIN_SHARES, shares);
}
2998
#endif /* CONFIG_SMP */
2999

3000 3001
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

3002 3003 3004 3005 3006
/*
 * 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
3007
{
3008 3009
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
	long shares, runnable;
P
Peter Zijlstra 已提交
3010

3011
	if (!gcfs_rq)
3012 3013
		return;

3014
	if (throttled_hierarchy(gcfs_rq))
P
Peter Zijlstra 已提交
3015
		return;
3016

3017
#ifndef CONFIG_SMP
3018
	runnable = shares = READ_ONCE(gcfs_rq->tg->shares);
3019 3020

	if (likely(se->load.weight == shares))
3021
		return;
3022
#else
3023 3024
	shares   = calc_group_shares(gcfs_rq);
	runnable = calc_group_runnable(gcfs_rq, shares);
3025
#endif
P
Peter Zijlstra 已提交
3026

3027
	reweight_entity(cfs_rq_of(se), se, shares, runnable);
P
Peter Zijlstra 已提交
3028
}
3029

P
Peter Zijlstra 已提交
3030
#else /* CONFIG_FAIR_GROUP_SCHED */
3031
static inline void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
3032 3033 3034 3035
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

3036
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq, int flags)
3037
{
3038 3039
	struct rq *rq = rq_of(cfs_rq);

3040
	if (&rq->cfs == cfs_rq || (flags & SCHED_CPUFREQ_MIGRATION)) {
3041 3042 3043
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
3044
		 * a real problem.
3045 3046 3047 3048 3049 3050 3051 3052 3053 3054
		 *
		 * 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().
		 */
3055
		cpufreq_update_util(rq, flags);
3056 3057 3058
	}
}

3059
#ifdef CONFIG_SMP
3060
#ifdef CONFIG_FAIR_GROUP_SCHED
3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073
/**
 * 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'.
 *
3074
 * Updating tg's load_avg is necessary before update_cfs_share().
3075
 */
3076
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
3077
{
3078
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
3079

3080 3081 3082 3083 3084 3085
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

3086 3087 3088
	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;
3089
	}
3090
}
3091

3092
/*
3093
 * Called within set_task_rq() right before setting a task's CPU. The
3094 3095 3096 3097 3098 3099
 * 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)
{
3100 3101 3102
	u64 p_last_update_time;
	u64 n_last_update_time;

3103 3104 3105 3106 3107 3108 3109 3110 3111 3112
	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.
	 */
3113 3114
	if (!(se->avg.last_update_time && prev))
		return;
3115 3116

#ifndef CONFIG_64BIT
3117
	{
3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131
		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);
3132
	}
3133
#else
3134 3135
	p_last_update_time = prev->avg.last_update_time;
	n_last_update_time = next->avg.last_update_time;
3136
#endif
3137 3138
	__update_load_avg_blocked_se(p_last_update_time, cpu_of(rq_of(prev)), se);
	se->avg.last_update_time = n_last_update_time;
3139
}
3140

3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151

/*
 * 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.
 *
3152 3153 3154
 * 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).
3155 3156 3157 3158 3159 3160 3161 3162
 *
 * 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:
 *
3163
 *   grq->avg.load_avg = grq->load.weight * grq->avg.runnable_avg	(3)
3164 3165 3166
 *
 * And per (1) we have:
 *
3167
 *   ge->avg.runnable_avg == grq->avg.runnable_avg
3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185
 *
 * 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).
 *
3186 3187 3188 3189 3190 3191
 * 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.
3192
 *
3193
 * So we'll have to approximate.. :/
3194
 *
3195
 * Given the constraint:
3196
 *
3197
 *   ge->avg.running_sum <= ge->avg.runnable_sum <= LOAD_AVG_MAX
3198
 *
3199 3200
 * We can construct a rule that adds runnable to a rq by assuming minimal
 * overlap.
3201
 *
3202
 * On removal, we'll assume each task is equally runnable; which yields:
3203
 *
3204
 *   grq->avg.runnable_sum = grq->avg.load_sum / grq->load.weight
3205
 *
3206
 * XXX: only do this for the part of runnable > running ?
3207 3208 3209
 *
 */

3210
static inline void
3211
update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3212 3213 3214 3215 3216 3217 3218
{
	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;

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

3219 3220 3221 3222 3223 3224 3225 3226
	/*
	 * 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.
	 */

3227 3228 3229 3230 3231 3232 3233 3234 3235 3236
	/* 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
3237
update_tg_cfs_runnable(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3238
{
3239 3240 3241 3242
	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;
3243

3244 3245
	if (!runnable_sum)
		return;
3246

3247
	gcfs_rq->prop_runnable_sum = 0;
3248

3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271
	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
3272
	 * As running sum is scale with CPU capacity wehreas the runnable sum
3273 3274 3275 3276 3277 3278
	 * 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);

3279 3280
	load_sum = (s64)se_weight(se) * runnable_sum;
	load_avg = div_s64(load_sum, LOAD_AVG_MAX);
3281

3282 3283
	delta_sum = load_sum - (s64)se_weight(se) * se->avg.load_sum;
	delta_avg = load_avg - se->avg.load_avg;
3284

3285 3286 3287 3288
	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);
3289

3290 3291
	runnable_load_sum = (s64)se_runnable(se) * runnable_sum;
	runnable_load_avg = div_s64(runnable_load_sum, LOAD_AVG_MAX);
3292 3293
	delta_sum = runnable_load_sum - se_weight(se) * se->avg.runnable_load_sum;
	delta_avg = runnable_load_avg - se->avg.runnable_load_avg;
3294

3295 3296
	se->avg.runnable_load_sum = runnable_sum;
	se->avg.runnable_load_avg = runnable_load_avg;
3297

3298
	if (se->on_rq) {
3299 3300
		add_positive(&cfs_rq->avg.runnable_load_avg, delta_avg);
		add_positive(&cfs_rq->avg.runnable_load_sum, delta_sum);
3301 3302 3303
	}
}

3304
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum)
3305
{
3306 3307
	cfs_rq->propagate = 1;
	cfs_rq->prop_runnable_sum += runnable_sum;
3308 3309 3310 3311 3312
}

/* Update task and its cfs_rq load average */
static inline int propagate_entity_load_avg(struct sched_entity *se)
{
3313
	struct cfs_rq *cfs_rq, *gcfs_rq;
3314 3315 3316 3317

	if (entity_is_task(se))
		return 0;

3318 3319
	gcfs_rq = group_cfs_rq(se);
	if (!gcfs_rq->propagate)
3320 3321
		return 0;

3322 3323
	gcfs_rq->propagate = 0;

3324 3325
	cfs_rq = cfs_rq_of(se);

3326
	add_tg_cfs_propagate(cfs_rq, gcfs_rq->prop_runnable_sum);
3327

3328 3329
	update_tg_cfs_util(cfs_rq, se, gcfs_rq);
	update_tg_cfs_runnable(cfs_rq, se, gcfs_rq);
3330 3331 3332 3333

	return 1;
}

3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352
/*
 * 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:
	 */
3353
	if (gcfs_rq->propagate)
3354 3355 3356 3357 3358 3359 3360 3361 3362 3363
		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;
}

3364
#else /* CONFIG_FAIR_GROUP_SCHED */
3365

3366
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
3367 3368 3369 3370 3371 3372

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

3373
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum) {}
3374

3375
#endif /* CONFIG_FAIR_GROUP_SCHED */
3376

3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387
/**
 * 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.
 *
3388 3389 3390 3391
 * 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.
3392
 */
3393
static inline int
3394
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3395
{
3396
	unsigned long removed_load = 0, removed_util = 0, removed_runnable_sum = 0;
3397
	struct sched_avg *sa = &cfs_rq->avg;
3398
	int decayed = 0;
3399

3400 3401
	if (cfs_rq->removed.nr) {
		unsigned long r;
3402
		u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
3403 3404 3405 3406

		raw_spin_lock(&cfs_rq->removed.lock);
		swap(cfs_rq->removed.util_avg, removed_util);
		swap(cfs_rq->removed.load_avg, removed_load);
3407
		swap(cfs_rq->removed.runnable_sum, removed_runnable_sum);
3408 3409 3410 3411
		cfs_rq->removed.nr = 0;
		raw_spin_unlock(&cfs_rq->removed.lock);

		r = removed_load;
3412
		sub_positive(&sa->load_avg, r);
3413
		sub_positive(&sa->load_sum, r * divider);
3414

3415
		r = removed_util;
3416
		sub_positive(&sa->util_avg, r);
3417
		sub_positive(&sa->util_sum, r * divider);
3418

3419
		add_tg_cfs_propagate(cfs_rq, -(long)removed_runnable_sum);
3420 3421

		decayed = 1;
3422
	}
3423

3424
	decayed |= __update_load_avg_cfs_rq(now, cpu_of(rq_of(cfs_rq)), cfs_rq);
3425

3426 3427 3428 3429
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3430

3431
	if (decayed)
3432
		cfs_rq_util_change(cfs_rq, 0);
3433

3434
	return decayed;
3435 3436
}

3437 3438 3439 3440
/**
 * 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
3441
 * @flags: migration hints
3442 3443 3444 3445
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3446
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3447
{
3448 3449 3450 3451 3452 3453 3454 3455 3456
	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
	 */
3457
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475
	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;

3476
	enqueue_load_avg(cfs_rq, se);
3477 3478
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3479 3480

	add_tg_cfs_propagate(cfs_rq, se->avg.load_sum);
3481

3482
	cfs_rq_util_change(cfs_rq, flags);
3483 3484
}

3485 3486 3487 3488 3489 3490 3491 3492
/**
 * 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.
 */
3493 3494
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3495
	dequeue_load_avg(cfs_rq, se);
3496 3497
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3498 3499

	add_tg_cfs_propagate(cfs_rq, -se->avg.load_sum);
3500

3501
	cfs_rq_util_change(cfs_rq, 0);
3502 3503
}

3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530
/*
 * 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)) {

3531 3532 3533 3534 3535 3536 3537 3538
		/*
		 * 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);
3539 3540 3541 3542 3543 3544
		update_tg_load_avg(cfs_rq, 0);

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

3545
#ifndef CONFIG_64BIT
3546 3547
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3548
	u64 last_update_time_copy;
3549
	u64 last_update_time;
3550

3551 3552 3553 3554 3555
	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);
3556 3557 3558

	return last_update_time;
}
3559
#else
3560 3561 3562 3563
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3564 3565
#endif

3566 3567 3568 3569 3570 3571 3572 3573 3574 3575
/*
 * 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);
3576
	__update_load_avg_blocked_se(last_update_time, cpu_of(rq_of(cfs_rq)), se);
3577 3578
}

3579 3580 3581 3582 3583 3584 3585
/*
 * 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);
3586
	unsigned long flags;
3587 3588

	/*
3589 3590 3591 3592 3593 3594 3595
	 * 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.
3596 3597
	 */

3598
	sync_entity_load_avg(se);
3599 3600 3601 3602 3603

	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;
3604
	cfs_rq->removed.runnable_sum	+= se->avg.load_sum; /* == runnable_sum */
3605
	raw_spin_unlock_irqrestore(&cfs_rq->removed.lock, flags);
3606
}
3607

3608 3609
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
3610
	return cfs_rq->avg.runnable_load_avg;
3611 3612 3613 3614 3615 3616 3617
}

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

3618
static int idle_balance(struct rq *this_rq, struct rq_flags *rf);
3619

3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646
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;
3647
	enqueued += (_task_util_est(p) | UTIL_AVG_UNCHANGED);
3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672
	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;

3673 3674 3675 3676
	/* 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));
3677 3678 3679 3680 3681 3682 3683 3684 3685
	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;

3686 3687 3688 3689 3690 3691 3692 3693
	/*
	 * 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;

3694 3695 3696 3697
	/*
	 * Skip update of task's estimated utilization when its EWMA is
	 * already ~1% close to its last activation value.
	 */
3698
	ue.enqueued = (task_util(p) | UTIL_AVG_UNCHANGED);
3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725
	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);
}

3726 3727
#else /* CONFIG_SMP */

3728 3729
#define UPDATE_TG	0x0
#define SKIP_AGE_LOAD	0x0
3730
#define DO_ATTACH	0x0
3731

3732
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int not_used1)
3733
{
3734
	cfs_rq_util_change(cfs_rq, 0);
3735 3736
}

3737
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3738

3739
static inline void
3740
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags) {}
3741 3742 3743
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3744
static inline int idle_balance(struct rq *rq, struct rq_flags *rf)
3745 3746 3747 3748
{
	return 0;
}

3749 3750 3751 3752 3753 3754 3755
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) {}

3756
#endif /* CONFIG_SMP */
3757

P
Peter Zijlstra 已提交
3758 3759 3760 3761 3762 3763 3764 3765 3766
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)
3767
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3768 3769 3770
#endif
}

3771 3772 3773
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3774
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3775

3776 3777 3778 3779 3780 3781
	/*
	 * 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
3782
	if (initial && sched_feat(START_DEBIT))
3783
		vruntime += sched_vslice(cfs_rq, se);
3784

3785
	/* sleeps up to a single latency don't count. */
3786
	if (!initial) {
3787
		unsigned long thresh = sysctl_sched_latency;
3788

3789 3790 3791 3792 3793 3794
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3795

3796
		vruntime -= thresh;
3797 3798
	}

3799
	/* ensure we never gain time by being placed backwards. */
3800
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3801 3802
}

3803 3804
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816
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())  {
3817
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3818
			     "stat_blocked and stat_runtime require the "
3819
			     "kernel parameter schedstats=enable or "
3820 3821 3822 3823 3824
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843

/*
 * 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)
 *
3844
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855
 *	  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.
 */

3856
static void
3857
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3858
{
3859 3860 3861
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3862
	/*
3863 3864
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3865
	 */
3866
	if (renorm && curr)
3867 3868
		se->vruntime += cfs_rq->min_vruntime;

3869 3870
	update_curr(cfs_rq);

3871
	/*
3872 3873 3874 3875
	 * 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.
3876
	 */
3877 3878 3879
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3880 3881 3882 3883 3884 3885 3886 3887
	/*
	 * 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
	 */
3888
	update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
3889
	update_cfs_group(se);
3890
	enqueue_runnable_load_avg(cfs_rq, se);
3891
	account_entity_enqueue(cfs_rq, se);
3892

3893
	if (flags & ENQUEUE_WAKEUP)
3894
		place_entity(cfs_rq, se, 0);
3895

3896
	check_schedstat_required();
3897 3898
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3899
	if (!curr)
3900
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3901
	se->on_rq = 1;
3902

3903
	if (cfs_rq->nr_running == 1) {
3904
		list_add_leaf_cfs_rq(cfs_rq);
3905 3906
		check_enqueue_throttle(cfs_rq);
	}
3907 3908
}

3909
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3910
{
3911 3912
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3913
		if (cfs_rq->last != se)
3914
			break;
3915 3916

		cfs_rq->last = NULL;
3917 3918
	}
}
P
Peter Zijlstra 已提交
3919

3920 3921 3922 3923
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3924
		if (cfs_rq->next != se)
3925
			break;
3926 3927

		cfs_rq->next = NULL;
3928
	}
P
Peter Zijlstra 已提交
3929 3930
}

3931 3932 3933 3934
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3935
		if (cfs_rq->skip != se)
3936
			break;
3937 3938

		cfs_rq->skip = NULL;
3939 3940 3941
	}
}

P
Peter Zijlstra 已提交
3942 3943
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3944 3945 3946 3947 3948
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3949 3950 3951

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3952 3953
}

3954
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3955

3956
static void
3957
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3958
{
3959 3960 3961 3962
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3963 3964 3965 3966 3967 3968 3969 3970 3971

	/*
	 * 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.
	 */
3972
	update_load_avg(cfs_rq, se, UPDATE_TG);
3973
	dequeue_runnable_load_avg(cfs_rq, se);
3974

3975
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
3976

P
Peter Zijlstra 已提交
3977
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3978

3979
	if (se != cfs_rq->curr)
3980
		__dequeue_entity(cfs_rq, se);
3981
	se->on_rq = 0;
3982
	account_entity_dequeue(cfs_rq, se);
3983 3984

	/*
3985 3986 3987 3988
	 * 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.
3989
	 */
3990
	if (!(flags & DEQUEUE_SLEEP))
3991
		se->vruntime -= cfs_rq->min_vruntime;
3992

3993 3994 3995
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3996
	update_cfs_group(se);
3997 3998 3999 4000 4001 4002 4003

	/*
	 * 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.
	 */
4004
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) != DEQUEUE_SAVE)
4005
		update_min_vruntime(cfs_rq);
4006 4007 4008 4009 4010
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
4011
static void
I
Ingo Molnar 已提交
4012
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4013
{
4014
	unsigned long ideal_runtime, delta_exec;
4015 4016
	struct sched_entity *se;
	s64 delta;
4017

P
Peter Zijlstra 已提交
4018
	ideal_runtime = sched_slice(cfs_rq, curr);
4019
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
4020
	if (delta_exec > ideal_runtime) {
4021
		resched_curr(rq_of(cfs_rq));
4022 4023 4024 4025 4026
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037
		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;

4038 4039
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
4040

4041 4042
	if (delta < 0)
		return;
4043

4044
	if (delta > ideal_runtime)
4045
		resched_curr(rq_of(cfs_rq));
4046 4047
}

4048
static void
4049
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
4050
{
4051 4052 4053 4054 4055 4056 4057
	/* '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.
		 */
4058
		update_stats_wait_end(cfs_rq, se);
4059
		__dequeue_entity(cfs_rq, se);
4060
		update_load_avg(cfs_rq, se, UPDATE_TG);
4061 4062
	}

4063
	update_stats_curr_start(cfs_rq, se);
4064
	cfs_rq->curr = se;
4065

I
Ingo Molnar 已提交
4066 4067 4068 4069 4070
	/*
	 * 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):
	 */
4071
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
4072 4073 4074
		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
4075
	}
4076

4077
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
4078 4079
}

4080 4081 4082
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

4083 4084 4085 4086 4087 4088 4089
/*
 * 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
 */
4090 4091
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4092
{
4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103
	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 */
4104

4105 4106 4107 4108 4109
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
4110 4111 4112 4113 4114 4115 4116 4117 4118 4119
		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;
		}

4120 4121 4122
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
4123

4124 4125 4126 4127 4128 4129
	/*
	 * 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;

4130 4131 4132 4133 4134 4135
	/*
	 * 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;

4136
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4137 4138

	return se;
4139 4140
}

4141
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4142

4143
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
4144 4145 4146 4147 4148 4149
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
4150
		update_curr(cfs_rq);
4151

4152 4153 4154
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

4155
	check_spread(cfs_rq, prev);
4156

4157
	if (prev->on_rq) {
4158
		update_stats_wait_start(cfs_rq, prev);
4159 4160
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
4161
		/* in !on_rq case, update occurred at dequeue */
4162
		update_load_avg(cfs_rq, prev, 0);
4163
	}
4164
	cfs_rq->curr = NULL;
4165 4166
}

P
Peter Zijlstra 已提交
4167 4168
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
4169 4170
{
	/*
4171
	 * Update run-time statistics of the 'current'.
4172
	 */
4173
	update_curr(cfs_rq);
4174

4175 4176 4177
	/*
	 * Ensure that runnable average is periodically updated.
	 */
4178
	update_load_avg(cfs_rq, curr, UPDATE_TG);
4179
	update_cfs_group(curr);
4180

P
Peter Zijlstra 已提交
4181 4182 4183 4184 4185
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
4186
	if (queued) {
4187
		resched_curr(rq_of(cfs_rq));
4188 4189
		return;
	}
P
Peter Zijlstra 已提交
4190 4191 4192 4193 4194 4195 4196 4197
	/*
	 * 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
4198
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
4199
		check_preempt_tick(cfs_rq, curr);
4200 4201
}

4202 4203 4204 4205 4206 4207

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

#ifdef CONFIG_CFS_BANDWIDTH
4208 4209

#ifdef HAVE_JUMP_LABEL
4210
static struct static_key __cfs_bandwidth_used;
4211 4212 4213

static inline bool cfs_bandwidth_used(void)
{
4214
	return static_key_false(&__cfs_bandwidth_used);
4215 4216
}

4217
void cfs_bandwidth_usage_inc(void)
4218
{
4219
	static_key_slow_inc_cpuslocked(&__cfs_bandwidth_used);
4220 4221 4222 4223
}

void cfs_bandwidth_usage_dec(void)
{
4224
	static_key_slow_dec_cpuslocked(&__cfs_bandwidth_used);
4225 4226 4227 4228 4229 4230 4231
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

4232 4233
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
4234 4235
#endif /* HAVE_JUMP_LABEL */

4236 4237 4238 4239 4240 4241 4242 4243
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
4244 4245 4246 4247 4248 4249

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

P
Paul Turner 已提交
4250 4251 4252 4253 4254 4255 4256
/*
 * Replenish runtime according to assigned quota and update expiration time.
 * We use sched_clock_cpu directly instead of rq->clock to avoid adding
 * additional synchronization around rq->lock.
 *
 * requires cfs_b->lock
 */
4257
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
4258 4259 4260 4261 4262 4263 4264 4265 4266
{
	u64 now;

	if (cfs_b->quota == RUNTIME_INF)
		return;

	now = sched_clock_cpu(smp_processor_id());
	cfs_b->runtime = cfs_b->quota;
	cfs_b->runtime_expires = now + ktime_to_ns(cfs_b->period);
4267
	cfs_b->expires_seq++;
P
Paul Turner 已提交
4268 4269
}

4270 4271 4272 4273 4274
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

4275 4276 4277 4278
/* 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))
4279
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
4280

4281
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
4282 4283
}

4284 4285
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4286 4287 4288
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
4289
	u64 amount = 0, min_amount, expires;
4290
	int expires_seq;
4291 4292 4293 4294 4295 4296 4297

	/* 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;
4298
	else {
P
Peter Zijlstra 已提交
4299
		start_cfs_bandwidth(cfs_b);
4300 4301 4302 4303 4304 4305

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
4306
	}
4307
	expires_seq = cfs_b->expires_seq;
P
Paul Turner 已提交
4308
	expires = cfs_b->runtime_expires;
4309 4310 4311
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
4312 4313 4314 4315 4316
	/*
	 * we may have advanced our local expiration to account for allowed
	 * spread between our sched_clock and the one on which runtime was
	 * issued.
	 */
4317 4318
	if (cfs_rq->expires_seq != expires_seq) {
		cfs_rq->expires_seq = expires_seq;
P
Paul Turner 已提交
4319
		cfs_rq->runtime_expires = expires;
4320
	}
4321 4322

	return cfs_rq->runtime_remaining > 0;
4323 4324
}

P
Paul Turner 已提交
4325 4326 4327 4328 4329
/*
 * Note: This depends on the synchronization provided by sched_clock and the
 * fact that rq->clock snapshots this value.
 */
static void expire_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4330
{
P
Paul Turner 已提交
4331 4332 4333
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
4334
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
4335 4336
		return;

P
Paul Turner 已提交
4337 4338 4339 4340 4341 4342 4343 4344 4345
	if (cfs_rq->runtime_remaining < 0)
		return;

	/*
	 * If the local deadline has passed we have to consider the
	 * possibility that our sched_clock is 'fast' and the global deadline
	 * has not truly expired.
	 *
	 * Fortunately we can check determine whether this the case by checking
4346
	 * whether the global deadline(cfs_b->expires_seq) has advanced.
P
Paul Turner 已提交
4347
	 */
4348
	if (cfs_rq->expires_seq == cfs_b->expires_seq) {
P
Paul Turner 已提交
4349 4350 4351 4352 4353 4354 4355 4356
		/* extend local deadline, drift is bounded above by 2 ticks */
		cfs_rq->runtime_expires += TICK_NSEC;
	} else {
		/* global deadline is ahead, expiration has passed */
		cfs_rq->runtime_remaining = 0;
	}
}

4357
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
4358 4359
{
	/* dock delta_exec before expiring quota (as it could span periods) */
4360
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
4361 4362 4363
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
4364 4365
		return;

4366 4367 4368 4369 4370
	/*
	 * 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))
4371
		resched_curr(rq_of(cfs_rq));
4372 4373
}

4374
static __always_inline
4375
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
4376
{
4377
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
4378 4379 4380 4381 4382
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

4383 4384
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
4385
	return cfs_bandwidth_used() && cfs_rq->throttled;
4386 4387
}

4388 4389 4390
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
4391
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417
}

/*
 * 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) {
4418
		/* adjust cfs_rq_clock_task() */
4419
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
4420
					     cfs_rq->throttled_clock_task;
4421 4422 4423 4424 4425 4426 4427 4428 4429 4430
	}

	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)];

4431 4432
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
4433
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
4434 4435 4436 4437 4438
	cfs_rq->throttle_count++;

	return 0;
}

4439
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
4440 4441 4442 4443 4444
{
	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
4445
	bool empty;
4446 4447 4448

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

4449
	/* freeze hierarchy runnable averages while throttled */
4450 4451 4452
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		struct cfs_rq *qcfs_rq = cfs_rq_of(se);
		/* throttled entity or throttle-on-deactivate */
		if (!se->on_rq)
			break;

		if (dequeue)
			dequeue_entity(qcfs_rq, se, DEQUEUE_SLEEP);
		qcfs_rq->h_nr_running -= task_delta;

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

	if (!se)
4470
		sub_nr_running(rq, task_delta);
4471 4472

	cfs_rq->throttled = 1;
4473
	cfs_rq->throttled_clock = rq_clock(rq);
4474
	raw_spin_lock(&cfs_b->lock);
4475
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4476

4477 4478
	/*
	 * Add to the _head_ of the list, so that an already-started
4479 4480
	 * 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.
4481
	 */
4482 4483 4484 4485
	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
4486 4487 4488 4489 4490 4491 4492 4493

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

4494 4495 4496
	raw_spin_unlock(&cfs_b->lock);
}

4497
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
4498 4499 4500 4501 4502 4503 4504
{
	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;

4505
	se = cfs_rq->tg->se[cpu_of(rq)];
4506 4507

	cfs_rq->throttled = 0;
4508 4509 4510

	update_rq_clock(rq);

4511
	raw_spin_lock(&cfs_b->lock);
4512
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4513 4514 4515
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4516 4517 4518
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536
	if (!cfs_rq->load.weight)
		return;

	task_delta = cfs_rq->h_nr_running;
	for_each_sched_entity(se) {
		if (se->on_rq)
			enqueue = 0;

		cfs_rq = cfs_rq_of(se);
		if (enqueue)
			enqueue_entity(cfs_rq, se, ENQUEUE_WAKEUP);
		cfs_rq->h_nr_running += task_delta;

		if (cfs_rq_throttled(cfs_rq))
			break;
	}

	if (!se)
4537
		add_nr_running(rq, task_delta);
4538

4539
	/* Determine whether we need to wake up potentially idle CPU: */
4540
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4541
		resched_curr(rq);
4542 4543 4544 4545 4546 4547
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4548 4549
	u64 runtime;
	u64 starting_runtime = remaining;
4550 4551 4552 4553 4554

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);
4555
		struct rq_flags rf;
4556

4557
		rq_lock(rq, &rf);
4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573
		if (!cfs_rq_throttled(cfs_rq))
			goto next;

		runtime = -cfs_rq->runtime_remaining + 1;
		if (runtime > remaining)
			runtime = remaining;
		remaining -= runtime;

		cfs_rq->runtime_remaining += runtime;
		cfs_rq->runtime_expires = expires;

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

next:
4574
		rq_unlock(rq, &rf);
4575 4576 4577 4578 4579 4580

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

4581
	return starting_runtime - remaining;
4582 4583
}

4584 4585 4586 4587 4588 4589 4590 4591
/*
 * 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)
{
4592
	u64 runtime, runtime_expires;
4593
	int throttled;
4594 4595 4596

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4597
		goto out_deactivate;
4598

4599
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4600
	cfs_b->nr_periods += overrun;
4601

4602 4603 4604 4605 4606 4607
	/*
	 * 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
4608 4609 4610

	__refill_cfs_bandwidth_runtime(cfs_b);

4611 4612 4613
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4614
		return 0;
4615 4616
	}

4617 4618 4619
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4620 4621 4622
	runtime_expires = cfs_b->runtime_expires;

	/*
4623 4624 4625 4626 4627
	 * This check is repeated as we are holding onto the new bandwidth while
	 * we unthrottle. This can potentially race with an unthrottled group
	 * trying to acquire new bandwidth from the global pool. This can result
	 * in us over-using our runtime if it is all used during this loop, but
	 * only by limited amounts in that extreme case.
4628
	 */
4629
	while (throttled && cfs_b->runtime > 0 && !cfs_b->distribute_running) {
4630
		runtime = cfs_b->runtime;
4631
		cfs_b->distribute_running = 1;
4632 4633 4634 4635 4636 4637
		raw_spin_unlock(&cfs_b->lock);
		/* we can't nest cfs_b->lock while distributing bandwidth */
		runtime = distribute_cfs_runtime(cfs_b, runtime,
						 runtime_expires);
		raw_spin_lock(&cfs_b->lock);

4638
		cfs_b->distribute_running = 0;
4639
		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4640 4641

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4642
	}
4643

4644 4645 4646 4647 4648 4649 4650
	/*
	 * 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;
4651

4652 4653 4654 4655
	return 0;

out_deactivate:
	return 1;
4656
}
4657

4658 4659 4660 4661 4662 4663 4664
/* 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;

4665 4666 4667 4668
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4669
 * hrtimer base being cleared by hrtimer_start. In the case of
4670 4671
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696
static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
{
	struct hrtimer *refresh_timer = &cfs_b->period_timer;
	u64 remaining;

	/* if the call-back is running a quota refresh is already occurring */
	if (hrtimer_callback_running(refresh_timer))
		return 1;

	/* is a quota refresh about to occur? */
	remaining = ktime_to_ns(hrtimer_expires_remaining(refresh_timer));
	if (remaining < min_expire)
		return 1;

	return 0;
}

static void start_cfs_slack_bandwidth(struct cfs_bandwidth *cfs_b)
{
	u64 min_left = cfs_bandwidth_slack_period + min_bandwidth_expiration;

	/* if there's a quota refresh soon don't bother with slack */
	if (runtime_refresh_within(cfs_b, min_left))
		return;

P
Peter Zijlstra 已提交
4697 4698 4699
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728
}

/* we know any runtime found here is valid as update_curr() precedes return */
static void __return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	s64 slack_runtime = cfs_rq->runtime_remaining - min_cfs_rq_runtime;

	if (slack_runtime <= 0)
		return;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota != RUNTIME_INF &&
	    cfs_rq->runtime_expires == cfs_b->runtime_expires) {
		cfs_b->runtime += slack_runtime;

		/* we are under rq->lock, defer unthrottling using a timer */
		if (cfs_b->runtime > sched_cfs_bandwidth_slice() &&
		    !list_empty(&cfs_b->throttled_cfs_rq))
			start_cfs_slack_bandwidth(cfs_b);
	}
	raw_spin_unlock(&cfs_b->lock);

	/* even if it's not valid for return we don't want to try again */
	cfs_rq->runtime_remaining -= slack_runtime;
}

static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
4729 4730 4731
	if (!cfs_bandwidth_used())
		return;

4732
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747
		return;

	__return_cfs_rq_runtime(cfs_rq);
}

/*
 * This is done with a timer (instead of inline with bandwidth return) since
 * it's necessary to juggle rq->locks to unthrottle their respective cfs_rqs.
 */
static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
{
	u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
	u64 expires;

	/* confirm we're still not at a refresh boundary */
4748
	raw_spin_lock(&cfs_b->lock);
4749 4750 4751 4752 4753
	if (cfs_b->distribute_running) {
		raw_spin_unlock(&cfs_b->lock);
		return;
	}

4754 4755
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4756
		return;
4757
	}
4758

4759
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4760
		runtime = cfs_b->runtime;
4761

4762
	expires = cfs_b->runtime_expires;
4763 4764 4765
	if (runtime)
		cfs_b->distribute_running = 1;

4766 4767 4768 4769 4770 4771 4772 4773 4774
	raw_spin_unlock(&cfs_b->lock);

	if (!runtime)
		return;

	runtime = distribute_cfs_runtime(cfs_b, runtime, expires);

	raw_spin_lock(&cfs_b->lock);
	if (expires == cfs_b->runtime_expires)
4775
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4776
	cfs_b->distribute_running = 0;
4777 4778 4779
	raw_spin_unlock(&cfs_b->lock);
}

4780 4781 4782 4783 4784 4785 4786
/*
 * 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)
{
4787 4788 4789
	if (!cfs_bandwidth_used())
		return;

4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803
	/* 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);
}

4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817
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;
4818
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4819 4820
}

4821
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4822
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4823
{
4824
	if (!cfs_bandwidth_used())
4825
		return false;
4826

4827
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4828
		return false;
4829 4830 4831 4832 4833 4834

	/*
	 * 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))
4835
		return true;
4836 4837

	throttle_cfs_rq(cfs_rq);
4838
	return true;
4839
}
4840 4841 4842 4843 4844

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
4845

4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857
	do_sched_cfs_slack_timer(cfs_b);

	return HRTIMER_NORESTART;
}

static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, period_timer);
	int overrun;
	int idle = 0;

4858
	raw_spin_lock(&cfs_b->lock);
4859
	for (;;) {
P
Peter Zijlstra 已提交
4860
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4861 4862 4863 4864 4865
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4866 4867
	if (idle)
		cfs_b->period_active = 0;
4868
	raw_spin_unlock(&cfs_b->lock);
4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880

	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
4881
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4882 4883 4884
	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;
4885
	cfs_b->distribute_running = 0;
4886 4887 4888 4889 4890 4891 4892 4893
}

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
4894
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4895
{
4896 4897
	u64 overrun;

P
Peter Zijlstra 已提交
4898
	lockdep_assert_held(&cfs_b->lock);
4899

4900 4901 4902 4903 4904 4905 4906 4907
	if (cfs_b->period_active)
		return;

	cfs_b->period_active = 1;
	overrun = hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
	cfs_b->runtime_expires += (overrun + 1) * ktime_to_ns(cfs_b->period);
	cfs_b->expires_seq++;
	hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
4908 4909 4910 4911
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4912 4913 4914 4915
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4916 4917 4918 4919
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4920
/*
4921
 * Both these CPU hotplug callbacks race against unregister_fair_sched_group()
4922 4923 4924 4925 4926 4927
 *
 * The race is harmless, since modifying bandwidth settings of unhooked group
 * bits doesn't do much.
 */

/* cpu online calback */
4928 4929
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
4930
	struct task_group *tg;
4931

4932 4933 4934 4935 4936 4937
	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)];
4938 4939 4940 4941 4942

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
4943
	rcu_read_unlock();
4944 4945
}

4946
/* cpu offline callback */
4947
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4948
{
4949 4950 4951 4952 4953 4954 4955
	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)];
4956 4957 4958 4959 4960 4961 4962 4963

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
4964
		cfs_rq->runtime_remaining = 1;
4965
		/*
4966
		 * Offline rq is schedulable till CPU is completely disabled
4967 4968 4969 4970
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

4971 4972 4973
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
4974
	rcu_read_unlock();
4975 4976 4977
}

#else /* CONFIG_CFS_BANDWIDTH */
4978 4979
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4980
	return rq_clock_task(rq_of(cfs_rq));
4981 4982
}

4983
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4984
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4985
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4986
static inline void sync_throttle(struct task_group *tg, int cpu) {}
4987
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4988 4989 4990 4991 4992

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003

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;
}
5004 5005 5006 5007 5008

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) {}
5009 5010
#endif

5011 5012 5013 5014 5015
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) {}
5016
static inline void update_runtime_enabled(struct rq *rq) {}
5017
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
5018 5019 5020

#endif /* CONFIG_CFS_BANDWIDTH */

5021 5022 5023 5024
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
5025 5026 5027 5028 5029 5030
#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);

5031
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
5032

5033
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
5034 5035 5036 5037 5038 5039
		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)
5040
				resched_curr(rq);
P
Peter Zijlstra 已提交
5041 5042
			return;
		}
5043
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
5044 5045
	}
}
5046 5047 5048 5049 5050 5051 5052 5053 5054 5055

/*
 * 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;

5056
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
5057 5058 5059 5060 5061
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
5062
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
5063 5064 5065 5066
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
5067 5068 5069 5070

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
5071 5072
#endif

5073 5074 5075 5076 5077
/*
 * 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:
 */
5078
static void
5079
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5080 5081
{
	struct cfs_rq *cfs_rq;
5082
	struct sched_entity *se = &p->se;
5083

5084 5085 5086 5087 5088 5089 5090 5091
	/*
	 * 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);

5092 5093 5094 5095 5096 5097
	/*
	 * 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)
5098
		cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
5099

5100
	for_each_sched_entity(se) {
5101
		if (se->on_rq)
5102 5103
			break;
		cfs_rq = cfs_rq_of(se);
5104
		enqueue_entity(cfs_rq, se, flags);
5105 5106 5107 5108 5109 5110

		/*
		 * 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.
5111
		 */
5112 5113
		if (cfs_rq_throttled(cfs_rq))
			break;
5114
		cfs_rq->h_nr_running++;
5115

5116
		flags = ENQUEUE_WAKEUP;
5117
	}
P
Peter Zijlstra 已提交
5118

P
Peter Zijlstra 已提交
5119
	for_each_sched_entity(se) {
5120
		cfs_rq = cfs_rq_of(se);
5121
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
5122

5123 5124 5125
		if (cfs_rq_throttled(cfs_rq))
			break;

5126
		update_load_avg(cfs_rq, se, UPDATE_TG);
5127
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5128 5129
	}

Y
Yuyang Du 已提交
5130
	if (!se)
5131
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
5132

5133
	hrtick_update(rq);
5134 5135
}

5136 5137
static void set_next_buddy(struct sched_entity *se);

5138 5139 5140 5141 5142
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
5143
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5144 5145
{
	struct cfs_rq *cfs_rq;
5146
	struct sched_entity *se = &p->se;
5147
	int task_sleep = flags & DEQUEUE_SLEEP;
5148 5149 5150

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5151
		dequeue_entity(cfs_rq, se, flags);
5152 5153 5154 5155 5156 5157 5158 5159 5160

		/*
		 * end evaluation on encountering a throttled cfs_rq
		 *
		 * note: in the case of encountering a throttled cfs_rq we will
		 * post the final h_nr_running decrement below.
		*/
		if (cfs_rq_throttled(cfs_rq))
			break;
5161
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5162

5163
		/* Don't dequeue parent if it has other entities besides us */
5164
		if (cfs_rq->load.weight) {
5165 5166
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
5167 5168 5169 5170
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
5171 5172
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
5173
			break;
5174
		}
5175
		flags |= DEQUEUE_SLEEP;
5176
	}
P
Peter Zijlstra 已提交
5177

P
Peter Zijlstra 已提交
5178
	for_each_sched_entity(se) {
5179
		cfs_rq = cfs_rq_of(se);
5180
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5181

5182 5183 5184
		if (cfs_rq_throttled(cfs_rq))
			break;

5185
		update_load_avg(cfs_rq, se, UPDATE_TG);
5186
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5187 5188
	}

Y
Yuyang Du 已提交
5189
	if (!se)
5190
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
5191

5192
	util_est_dequeue(&rq->cfs, p, task_sleep);
5193
	hrtick_update(rq);
5194 5195
}

5196
#ifdef CONFIG_SMP
5197 5198 5199 5200 5201

/* 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);

5202
#ifdef CONFIG_NO_HZ_COMMON
5203 5204 5205 5206 5207
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
5208
 * The exact cpuload calculated at every tick would be:
5209
 *
5210 5211
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
5212 5213
 * 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:
5214 5215 5216
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
5217 5218 5219
 *
 * decay_load_missed() below does efficient calculation of
 *
5220 5221 5222 5223 5224 5225
 *   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())
5226
 *
5227
 * The calculation is approximated on a 128 point scale.
5228 5229
 */
#define DEGRADE_SHIFT		7
5230 5231 5232 5233 5234 5235 5236 5237 5238

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 }
};
5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267

/*
 * 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;
}
5268 5269 5270 5271

static struct {
	cpumask_var_t idle_cpus_mask;
	atomic_t nr_cpus;
5272
	int has_blocked;		/* Idle CPUS has blocked load */
5273
	unsigned long next_balance;     /* in jiffy units */
5274
	unsigned long next_blocked;	/* Next update of blocked load in jiffies */
5275 5276
} nohz ____cacheline_aligned;

5277
#endif /* CONFIG_NO_HZ_COMMON */
5278

5279
/**
5280
 * __cpu_load_update - update the rq->cpu_load[] statistics
5281 5282 5283 5284
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
5285
 * Update rq->cpu_load[] statistics. This function is usually called every
5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311
 * 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
5312
 * term.
5313
 */
5314 5315
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
5316
{
5317
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328
	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 */

5329
		old_load = this_rq->cpu_load[i];
5330
#ifdef CONFIG_NO_HZ_COMMON
5331
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
5332 5333 5334 5335 5336 5337 5338 5339 5340
		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;
		}
5341
#endif
5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354
		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;
	}
}

5355
/* Used instead of source_load when we know the type == 0 */
5356
static unsigned long weighted_cpuload(struct rq *rq)
5357
{
5358
	return cfs_rq_runnable_load_avg(&rq->cfs);
5359 5360
}

5361
#ifdef CONFIG_NO_HZ_COMMON
5362 5363
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
5364
 * CPU doing the jiffies update might drift wrt the CPU doing the jiffy reading
5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378
 * 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)
5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389
{
	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.
		 */
5390
		cpu_load_update(this_rq, load, pending_updates);
5391 5392 5393
	}
}

5394 5395 5396 5397
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
5398
static void cpu_load_update_idle(struct rq *this_rq)
5399 5400 5401 5402
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
5403
	if (weighted_cpuload(this_rq))
5404 5405
		return;

5406
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
5407 5408 5409
}

/*
5410 5411 5412 5413
 * 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.
5414
 */
5415
void cpu_load_update_nohz_start(void)
5416 5417
{
	struct rq *this_rq = this_rq();
5418 5419 5420 5421 5422 5423

	/*
	 * 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.
	 */
5424
	this_rq->cpu_load[0] = weighted_cpuload(this_rq);
5425 5426 5427 5428 5429 5430 5431
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
5432
	unsigned long curr_jiffies = READ_ONCE(jiffies);
5433 5434
	struct rq *this_rq = this_rq();
	unsigned long load;
5435
	struct rq_flags rf;
5436 5437 5438 5439

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

5440
	load = weighted_cpuload(this_rq);
5441
	rq_lock(this_rq, &rf);
5442
	update_rq_clock(this_rq);
5443
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
5444
	rq_unlock(this_rq, &rf);
5445
}
5446 5447 5448 5449 5450 5451 5452 5453
#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)
{
5454
#ifdef CONFIG_NO_HZ_COMMON
5455 5456
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
5457
#endif
5458 5459
	cpu_load_update(this_rq, load, 1);
}
5460 5461 5462 5463

/*
 * Called from scheduler_tick()
 */
5464
void cpu_load_update_active(struct rq *this_rq)
5465
{
5466
	unsigned long load = weighted_cpuload(this_rq);
5467 5468 5469 5470 5471

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
5472 5473
}

5474
/*
5475
 * Return a low guess at the load of a migration-source CPU weighted
5476 5477 5478 5479 5480 5481 5482 5483
 * 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);
5484
	unsigned long total = weighted_cpuload(rq);
5485 5486 5487 5488 5489 5490 5491 5492

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

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

/*
5493
 * Return a high guess at the load of a migration-target CPU weighted
5494 5495 5496 5497 5498
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
5499
	unsigned long total = weighted_cpuload(rq);
5500 5501 5502 5503 5504 5505 5506

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

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

5507
static unsigned long capacity_of(int cpu)
5508
{
5509
	return cpu_rq(cpu)->cpu_capacity;
5510 5511
}

5512 5513 5514 5515 5516
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

5517 5518 5519
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5520
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
5521
	unsigned long load_avg = weighted_cpuload(rq);
5522 5523

	if (nr_running)
5524
		return load_avg / nr_running;
5525 5526 5527 5528

	return 0;
}

P
Peter Zijlstra 已提交
5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545
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
5546 5547
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5548
 *
M
Mike Galbraith 已提交
5549
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561
 * 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
5562
 */
5563 5564
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5565 5566
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5567
	int factor = this_cpu_read(sd_llc_size);
5568

M
Mike Galbraith 已提交
5569 5570 5571 5572 5573
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5574 5575
}

5576
/*
5577 5578 5579
 * 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.
5580
 *
5581 5582
 * wake_affine_idle() - only considers 'now', it check if the waking CPU is
 *			cache-affine and is (or	will be) idle.
5583 5584 5585 5586
 *
 * wake_affine_weight() - considers the weight to reflect the average
 *			  scheduling latency of the CPUs. This seems to work
 *			  for the overloaded case.
5587
 */
5588
static int
5589
wake_affine_idle(int this_cpu, int prev_cpu, int sync)
5590
{
5591 5592 5593 5594 5595
	/*
	 * 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.
5596 5597 5598 5599 5600 5601
	 *
	 * 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.
5602
	 */
5603 5604
	if (available_idle_cpu(this_cpu) && cpus_share_cache(this_cpu, prev_cpu))
		return available_idle_cpu(prev_cpu) ? prev_cpu : this_cpu;
5605

5606
	if (sync && cpu_rq(this_cpu)->nr_running == 1)
5607
		return this_cpu;
5608

5609
	return nr_cpumask_bits;
5610 5611
}

5612
static int
5613 5614
wake_affine_weight(struct sched_domain *sd, struct task_struct *p,
		   int this_cpu, int prev_cpu, int sync)
5615 5616 5617 5618
{
	s64 this_eff_load, prev_eff_load;
	unsigned long task_load;

5619
	this_eff_load = target_load(this_cpu, sd->wake_idx);
5620 5621 5622 5623

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

5624
		if (current_load > this_eff_load)
5625
			return this_cpu;
5626

5627
		this_eff_load -= current_load;
5628 5629 5630 5631
	}

	task_load = task_h_load(p);

5632 5633 5634 5635
	this_eff_load += task_load;
	if (sched_feat(WA_BIAS))
		this_eff_load *= 100;
	this_eff_load *= capacity_of(prev_cpu);
5636

5637
	prev_eff_load = source_load(prev_cpu, sd->wake_idx);
5638 5639 5640 5641
	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);
5642

5643 5644 5645 5646 5647 5648 5649 5650 5651 5652
	/*
	 * 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;
5653 5654
}

5655
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
5656
		       int this_cpu, int prev_cpu, int sync)
5657
{
5658
	int target = nr_cpumask_bits;
5659

5660
	if (sched_feat(WA_IDLE))
5661
		target = wake_affine_idle(this_cpu, prev_cpu, sync);
5662

5663 5664
	if (sched_feat(WA_WEIGHT) && target == nr_cpumask_bits)
		target = wake_affine_weight(sd, p, this_cpu, prev_cpu, sync);
5665

5666
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5667 5668
	if (target == nr_cpumask_bits)
		return prev_cpu;
5669

5670 5671 5672
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
	return target;
5673 5674
}

5675
static unsigned long cpu_util_without(int cpu, struct task_struct *p);
5676

5677
static unsigned long capacity_spare_without(int cpu, struct task_struct *p)
5678
{
5679
	return max_t(long, capacity_of(cpu) - cpu_util_without(cpu, p), 0);
5680 5681
}

5682 5683 5684
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
5685 5686
 *
 * Assumes p is allowed on at least one CPU in sd.
5687 5688
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5689
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5690
		  int this_cpu, int sd_flag)
5691
{
5692
	struct sched_group *idlest = NULL, *group = sd->groups;
5693
	struct sched_group *most_spare_sg = NULL;
5694 5695 5696
	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;
5697
	unsigned long most_spare = 0, this_spare = 0;
5698
	int load_idx = sd->forkexec_idx;
5699 5700 5701
	int imbalance_scale = 100 + (sd->imbalance_pct-100)/2;
	unsigned long imbalance = scale_load_down(NICE_0_LOAD) *
				(sd->imbalance_pct-100) / 100;
5702

5703 5704 5705
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5706
	do {
5707 5708
		unsigned long load, avg_load, runnable_load;
		unsigned long spare_cap, max_spare_cap;
5709 5710
		int local_group;
		int i;
5711

5712
		/* Skip over this group if it has no CPUs allowed */
5713
		if (!cpumask_intersects(sched_group_span(group),
5714
					&p->cpus_allowed))
5715 5716 5717
			continue;

		local_group = cpumask_test_cpu(this_cpu,
5718
					       sched_group_span(group));
5719

5720 5721 5722 5723
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
5724
		avg_load = 0;
5725
		runnable_load = 0;
5726
		max_spare_cap = 0;
5727

5728
		for_each_cpu(i, sched_group_span(group)) {
5729
			/* Bias balancing toward CPUs of our domain */
5730 5731 5732 5733 5734
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

5735 5736 5737
			runnable_load += load;

			avg_load += cfs_rq_load_avg(&cpu_rq(i)->cfs);
5738

5739
			spare_cap = capacity_spare_without(i, p);
5740 5741 5742

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
5743 5744
		}

5745
		/* Adjust by relative CPU capacity of the group */
5746 5747 5748 5749
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
		runnable_load = (runnable_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
5750 5751

		if (local_group) {
5752 5753
			this_runnable_load = runnable_load;
			this_avg_load = avg_load;
5754 5755
			this_spare = max_spare_cap;
		} else {
5756 5757 5758
			if (min_runnable_load > (runnable_load + imbalance)) {
				/*
				 * The runnable load is significantly smaller
5759
				 * so we can pick this new CPU:
5760 5761 5762 5763 5764 5765 5766 5767
				 */
				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
5768
				 * blocked load into account through avg_load:
5769 5770
				 */
				min_avg_load = avg_load;
5771 5772 5773 5774 5775 5776 5777
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
5778 5779 5780
		}
	} while (group = group->next, group != sd->groups);

5781 5782 5783 5784 5785 5786
	/*
	 * 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.
5787 5788 5789 5790
	 *
	 * 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.
5791
	 */
5792 5793 5794
	if (sd_flag & SD_BALANCE_FORK)
		goto skip_spare;

5795
	if (this_spare > task_util(p) / 2 &&
5796
	    imbalance_scale*this_spare > 100*most_spare)
5797
		return NULL;
5798 5799

	if (most_spare > task_util(p) / 2)
5800 5801
		return most_spare_sg;

5802
skip_spare:
5803 5804 5805
	if (!idlest)
		return NULL;

5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817
	/*
	 * 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;

5818
	if (min_runnable_load > (this_runnable_load + imbalance))
5819
		return NULL;
5820 5821 5822 5823 5824

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

5825 5826 5827 5828
	return idlest;
}

/*
5829
 * find_idlest_group_cpu - find the idlest CPU among the CPUs in the group.
5830 5831
 */
static int
5832
find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
5833 5834
{
	unsigned long load, min_load = ULONG_MAX;
5835 5836 5837 5838
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5839 5840
	int i;

5841 5842
	/* Check if we have any choice: */
	if (group->group_weight == 1)
5843
		return cpumask_first(sched_group_span(group));
5844

5845
	/* Traverse only the allowed CPUs */
5846
	for_each_cpu_and(i, sched_group_span(group), &p->cpus_allowed) {
5847
		if (available_idle_cpu(i)) {
5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868
			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;
			}
5869
		} else if (shallowest_idle_cpu == -1) {
5870
			load = weighted_cpuload(cpu_rq(i));
5871
			if (load < min_load) {
5872 5873 5874
				min_load = load;
				least_loaded_cpu = i;
			}
5875 5876 5877
		}
	}

5878
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5879
}
5880

5881 5882 5883
static inline int find_idlest_cpu(struct sched_domain *sd, struct task_struct *p,
				  int cpu, int prev_cpu, int sd_flag)
{
5884
	int new_cpu = cpu;
5885

5886 5887 5888
	if (!cpumask_intersects(sched_domain_span(sd), &p->cpus_allowed))
		return prev_cpu;

5889
	/*
5890 5891
	 * We need task's util for capacity_spare_without, sync it up to
	 * prev_cpu's last_update_time.
5892 5893 5894 5895
	 */
	if (!(sd_flag & SD_BALANCE_FORK))
		sync_entity_load_avg(&p->se);

5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912
	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);
5913
		if (new_cpu == cpu) {
5914
			/* Now try balancing at a lower domain level of 'cpu': */
5915 5916 5917 5918
			sd = sd->child;
			continue;
		}

5919
		/* Now try balancing at a lower domain level of 'new_cpu': */
5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933
		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;
}

5934
#ifdef CONFIG_SCHED_SMT
5935
DEFINE_STATIC_KEY_FALSE(sched_smt_present);
5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963

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
5964
void __update_idle_core(struct rq *rq)
5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976
{
	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;

5977
		if (!available_idle_cpu(cpu))
5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993
			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);
5994
	int core, cpu;
5995

P
Peter Zijlstra 已提交
5996 5997 5998
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5999 6000 6001
	if (!test_idle_cores(target, false))
		return -1;

6002
	cpumask_and(cpus, sched_domain_span(sd), &p->cpus_allowed);
6003

6004
	for_each_cpu_wrap(core, cpus, target) {
6005 6006 6007 6008
		bool idle = true;

		for_each_cpu(cpu, cpu_smt_mask(core)) {
			cpumask_clear_cpu(cpu, cpus);
6009
			if (!available_idle_cpu(cpu))
6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031
				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
6032 6033 6034
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6035
	for_each_cpu(cpu, cpu_smt_mask(target)) {
6036
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6037
			continue;
6038
		if (available_idle_cpu(cpu))
6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055 6056 6057 6058 6059 6060 6061 6062
			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).
6063
 */
6064 6065
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
6066
	struct sched_domain *this_sd;
6067
	u64 avg_cost, avg_idle;
6068 6069
	u64 time, cost;
	s64 delta;
6070
	int cpu, nr = INT_MAX;
6071

6072 6073 6074 6075
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

6076 6077 6078 6079
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
6080 6081 6082 6083
	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)
6084 6085
		return -1;

6086 6087 6088 6089 6090 6091 6092 6093
	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;
	}

6094 6095
	time = local_clock();

6096
	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
6097 6098
		if (!--nr)
			return -1;
6099
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6100
			continue;
6101
		if (available_idle_cpu(cpu))
6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114
			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.
6115
 */
6116
static int select_idle_sibling(struct task_struct *p, int prev, int target)
6117
{
6118
	struct sched_domain *sd;
6119
	int i, recent_used_cpu;
6120

6121
	if (available_idle_cpu(target))
6122
		return target;
6123 6124

	/*
6125
	 * If the previous CPU is cache affine and idle, don't be stupid:
6126
	 */
6127
	if (prev != target && cpus_share_cache(prev, target) && available_idle_cpu(prev))
6128
		return prev;
6129

6130
	/* Check a recently used CPU as a potential idle candidate: */
6131 6132 6133 6134
	recent_used_cpu = p->recent_used_cpu;
	if (recent_used_cpu != prev &&
	    recent_used_cpu != target &&
	    cpus_share_cache(recent_used_cpu, target) &&
6135
	    available_idle_cpu(recent_used_cpu) &&
6136 6137 6138
	    cpumask_test_cpu(p->recent_used_cpu, &p->cpus_allowed)) {
		/*
		 * Replace recent_used_cpu with prev as it is a potential
6139
		 * candidate for the next wake:
6140 6141 6142 6143 6144
		 */
		p->recent_used_cpu = prev;
		return recent_used_cpu;
	}

6145
	sd = rcu_dereference(per_cpu(sd_llc, target));
6146 6147
	if (!sd)
		return target;
6148

6149 6150 6151
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6152

6153 6154 6155 6156 6157 6158 6159
	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;
6160

6161 6162
	return target;
}
6163

6164 6165 6166 6167 6168 6169 6170
/**
 * 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).
6171 6172 6173 6174 6175 6176 6177 6178 6179 6180
 *
 * 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.
 *
6181 6182 6183 6184 6185 6186 6187 6188
 * 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.
 *
6189 6190 6191 6192 6193 6194 6195 6196 6197 6198
 * 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).
6199 6200
 *
 * Return: the (estimated) utilization for the specified CPU
6201
 */
6202
static inline unsigned long cpu_util(int cpu)
6203
{
6204 6205 6206 6207 6208 6209 6210 6211
	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));
6212

6213
	return min_t(unsigned long, util, capacity_orig_of(cpu));
6214
}
6215

6216
/*
6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227
 * 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.
6228
 */
6229
static unsigned long cpu_util_without(int cpu, struct task_struct *p)
6230
{
6231 6232
	struct cfs_rq *cfs_rq;
	unsigned int util;
6233 6234

	/* Task has no contribution or is new */
6235
	if (cpu != task_cpu(p) || !READ_ONCE(p->se.avg.last_update_time))
6236 6237
		return cpu_util(cpu);

6238 6239 6240
	cfs_rq = &cpu_rq(cpu)->cfs;
	util = READ_ONCE(cfs_rq->avg.util_avg);

6241
	/* Discount task's util from CPU's util */
6242
	util -= min_t(unsigned int, util, task_util(p));
6243

6244 6245 6246 6247 6248 6249
	/*
	 * 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:
6250
	 *      cpu_util_without = (cpu_util - task_util) = 0
6251 6252 6253 6254 6255 6256
	 *
	 * 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:
6257
	 *      cpu_util_without = (cpu_util - task_util) >= 0
6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269
	 *
	 * 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.
	 */
6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296
	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);
	}
6297 6298 6299 6300 6301 6302 6303

	/*
	 * 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));
6304 6305
}

6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323
/*
 * 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;

6324 6325 6326
	/* Bring task utilization in sync with prev_cpu */
	sync_entity_load_avg(&p->se);

6327 6328 6329
	return min_cap * 1024 < task_util(p) * capacity_margin;
}

6330
/*
6331 6332 6333
 * 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.
6334
 *
6335 6336
 * 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.
6337
 *
6338
 * Returns the target CPU number.
6339 6340 6341
 *
 * preempt must be disabled.
 */
6342
static int
6343
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
6344
{
6345
	struct sched_domain *tmp, *sd = NULL;
6346
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
6347
	int new_cpu = prev_cpu;
6348
	int want_affine = 0;
6349
	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
6350

P
Peter Zijlstra 已提交
6351 6352
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
6353
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
6354
			      && cpumask_test_cpu(cpu, &p->cpus_allowed);
P
Peter Zijlstra 已提交
6355
	}
6356

6357
	rcu_read_lock();
6358
	for_each_domain(cpu, tmp) {
6359
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
6360
			break;
6361

6362
		/*
6363
		 * If both 'cpu' and 'prev_cpu' are part of this domain,
6364
		 * cpu is a valid SD_WAKE_AFFINE target.
6365
		 */
6366 6367
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
6368 6369 6370 6371
			if (cpu != prev_cpu)
				new_cpu = wake_affine(tmp, p, cpu, prev_cpu, sync);

			sd = NULL; /* Prefer wake_affine over balance flags */
6372
			break;
6373
		}
6374

6375
		if (tmp->flags & sd_flag)
6376
			sd = tmp;
M
Mike Galbraith 已提交
6377 6378
		else if (!want_affine)
			break;
6379 6380
	}

6381 6382
	if (unlikely(sd)) {
		/* Slow path */
6383
		new_cpu = find_idlest_cpu(sd, p, cpu, prev_cpu, sd_flag);
6384 6385 6386 6387 6388 6389 6390
	} 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;
6391
	}
6392
	rcu_read_unlock();
6393

6394
	return new_cpu;
6395
}
6396

6397 6398
static void detach_entity_cfs_rq(struct sched_entity *se);

6399
/*
6400
 * Called immediately before a task is migrated to a new CPU; task_cpu(p) and
6401
 * cfs_rq_of(p) references at time of call are still valid and identify the
6402
 * previous CPU. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
6403
 */
6404
static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
6405
{
6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431
	/*
	 * 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;
	}

6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450
	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);
	}
6451 6452 6453

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
6454 6455

	/* We have migrated, no longer consider this task hot */
6456
	p->se.exec_start = 0;
6457 6458

	update_scan_period(p, new_cpu);
6459
}
6460 6461 6462 6463 6464

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
6465 6466
#endif /* CONFIG_SMP */

6467
static unsigned long wakeup_gran(struct sched_entity *se)
6468 6469 6470 6471
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
6472 6473
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
6474 6475 6476 6477 6478 6479 6480 6481 6482
	 *
	 * 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.
6483
	 */
6484
	return calc_delta_fair(gran, se);
6485 6486
}

6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508
/*
 * 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;

6509
	gran = wakeup_gran(se);
6510 6511 6512 6513 6514 6515
	if (vdiff > gran)
		return 1;

	return 0;
}

6516 6517
static void set_last_buddy(struct sched_entity *se)
{
6518 6519 6520
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6521 6522 6523
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6524
		cfs_rq_of(se)->last = se;
6525
	}
6526 6527 6528 6529
}

static void set_next_buddy(struct sched_entity *se)
{
6530 6531 6532
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6533 6534 6535
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6536
		cfs_rq_of(se)->next = se;
6537
	}
6538 6539
}

6540 6541
static void set_skip_buddy(struct sched_entity *se)
{
6542 6543
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
6544 6545
}

6546 6547 6548
/*
 * Preempt the current task with a newly woken task if needed:
 */
6549
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
6550 6551
{
	struct task_struct *curr = rq->curr;
6552
	struct sched_entity *se = &curr->se, *pse = &p->se;
6553
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
6554
	int scale = cfs_rq->nr_running >= sched_nr_latency;
6555
	int next_buddy_marked = 0;
6556

I
Ingo Molnar 已提交
6557 6558 6559
	if (unlikely(se == pse))
		return;

6560
	/*
6561
	 * This is possible from callers such as attach_tasks(), in which we
6562 6563 6564 6565 6566 6567 6568
	 * 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;

6569
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
6570
		set_next_buddy(pse);
6571 6572
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
6573

6574 6575 6576
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
6577 6578 6579 6580 6581 6582
	 *
	 * 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.
6583 6584 6585 6586
	 */
	if (test_tsk_need_resched(curr))
		return;

6587 6588 6589 6590 6591
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

6592
	/*
6593 6594
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
6595
	 */
6596
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
6597
		return;
6598

6599
	find_matching_se(&se, &pse);
6600
	update_curr(cfs_rq_of(se));
6601
	BUG_ON(!pse);
6602 6603 6604 6605 6606 6607 6608
	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);
6609
		goto preempt;
6610
	}
6611

6612
	return;
6613

6614
preempt:
6615
	resched_curr(rq);
6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629
	/*
	 * 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);
6630 6631
}

6632
static struct task_struct *
6633
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
6634 6635 6636
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
6637
	struct task_struct *p;
6638
	int new_tasks;
6639

6640
again:
6641
	if (!cfs_rq->nr_running)
6642
		goto idle;
6643

6644
#ifdef CONFIG_FAIR_GROUP_SCHED
6645
	if (prev->sched_class != &fair_sched_class)
6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664
		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.
		 */
6665 6666 6667 6668 6669
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6670

6671 6672 6673
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
6674
			 * Therefore the nr_running test will indeed
6675 6676
			 * be correct.
			 */
6677 6678 6679 6680 6681 6682
			if (unlikely(check_cfs_rq_runtime(cfs_rq))) {
				cfs_rq = &rq->cfs;

				if (!cfs_rq->nr_running)
					goto idle;

6683
				goto simple;
6684
			}
6685
		}
6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718

		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);
	}

6719
	goto done;
6720 6721
simple:
#endif
6722

6723
	put_prev_task(rq, prev);
6724

6725
	do {
6726
		se = pick_next_entity(cfs_rq, NULL);
6727
		set_next_entity(cfs_rq, se);
6728 6729 6730
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6731
	p = task_of(se);
6732

6733
done: __maybe_unused;
6734 6735 6736 6737 6738 6739 6740 6741 6742
#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

6743 6744
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6745 6746

	return p;
6747 6748

idle:
6749 6750
	new_tasks = idle_balance(rq, rf);

6751 6752 6753 6754 6755
	/*
	 * 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.
	 */
6756
	if (new_tasks < 0)
6757 6758
		return RETRY_TASK;

6759
	if (new_tasks > 0)
6760 6761 6762
		goto again;

	return NULL;
6763 6764 6765 6766 6767
}

/*
 * Account for a descheduled task:
 */
6768
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6769 6770 6771 6772 6773 6774
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6775
		put_prev_entity(cfs_rq, se);
6776 6777 6778
	}
}

6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803
/*
 * 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);
6804 6805 6806 6807 6808
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6809
		rq_clock_skip_update(rq);
6810 6811 6812 6813 6814
	}

	set_skip_buddy(se);
}

6815 6816 6817 6818
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6819 6820
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6821 6822 6823 6824 6825 6826 6827 6828 6829 6830
		return false;

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

	yield_task_fair(rq);

	return true;
}

6831
#ifdef CONFIG_SMP
6832
/**************************************************
P
Peter Zijlstra 已提交
6833 6834 6835 6836 6837
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
6838
 * per-CPU scheduler provides, namely provide a proportional amount of compute
P
Peter Zijlstra 已提交
6839 6840 6841 6842
 * 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)
 *
6843
 * Where W_i,n is the n-th weight average for CPU i. The instantaneous weight
P
Peter Zijlstra 已提交
6844 6845 6846 6847
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
6848
 * Where w_i,j is the weight of the j-th runnable task on CPU i. This weight
6849
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6850 6851 6852 6853 6854 6855
 *
 * 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)
 *
6856
 * C_i is the compute capacity of CPU i, typically it is the
P
Peter Zijlstra 已提交
6857 6858 6859 6860 6861 6862
 * 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):
 *
6863
 *   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
6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876
 *
 * 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)
6877
 * for all i,j solution, we create a tree of CPUs that follows the hardware
P
Peter Zijlstra 已提交
6878
 * topology where each level pairs two lower groups (or better). This results
6879
 * in O(log n) layers. Furthermore we reduce the number of CPUs going up the
P
Peter Zijlstra 已提交
6880
 * tree to only the first of the previous level and we decrease the frequency
6881
 * of load-balance at each level inv. proportional to the number of CPUs in
P
Peter Zijlstra 已提交
6882 6883 6884 6885 6886 6887 6888 6889
 * 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
6890
 *         |         |     `- number of CPUs doing load-balance
P
Peter Zijlstra 已提交
6891 6892 6893 6894 6895 6896 6897
 *         |         `- 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
6898
 * to every other CPU in at most O(log n) steps:
P
Peter Zijlstra 已提交
6899 6900 6901
 *
 * The adjacency matrix of the resulting graph is given by:
 *
6902
 *             log_2 n
P
Peter Zijlstra 已提交
6903 6904 6905 6906 6907 6908 6909
 *   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)
 *
6910
 * Showing there's indeed a path between every CPU in at most O(log n) steps.
P
Peter Zijlstra 已提交
6911 6912 6913 6914 6915 6916 6917 6918 6919
 * 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
6920
 * balancing is more aggressive and has the newly idle CPU iterate up the domain
P
Peter Zijlstra 已提交
6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940
 * 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)
 *
6941
 * 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
6942 6943 6944 6945 6946 6947
 *
 * 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.]
6948
 */
6949

6950 6951
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

6952 6953
enum fbq_type { regular, remote, all };

6954
#define LBF_ALL_PINNED	0x01
6955
#define LBF_NEED_BREAK	0x02
6956 6957
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
6958
#define LBF_NOHZ_STATS	0x10
6959
#define LBF_NOHZ_AGAIN	0x20
6960 6961 6962 6963 6964

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
6965
	int			src_cpu;
6966 6967 6968 6969

	int			dst_cpu;
	struct rq		*dst_rq;

6970 6971
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
6972
	enum cpu_idle_type	idle;
6973
	long			imbalance;
6974 6975 6976
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

6977
	unsigned int		flags;
6978 6979 6980 6981

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
6982 6983

	enum fbq_type		fbq_type;
6984
	struct list_head	tasks;
6985 6986
};

6987 6988 6989
/*
 * Is this task likely cache-hot:
 */
6990
static int task_hot(struct task_struct *p, struct lb_env *env)
6991 6992 6993
{
	s64 delta;

6994 6995
	lockdep_assert_held(&env->src_rq->lock);

6996 6997 6998 6999 7000 7001 7002 7003 7004
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
7005
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
7006 7007 7008 7009 7010 7011 7012 7013 7014
			(&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;

7015
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
7016 7017 7018 7019

	return delta < (s64)sysctl_sched_migration_cost;
}

7020
#ifdef CONFIG_NUMA_BALANCING
7021
/*
7022 7023 7024
 * 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.
7025
 */
7026
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
7027
{
7028
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
7029 7030
	unsigned long src_weight, dst_weight;
	int src_nid, dst_nid, dist;
7031

7032
	if (!static_branch_likely(&sched_numa_balancing))
7033 7034
		return -1;

7035
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
7036
		return -1;
7037 7038 7039 7040

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

7041
	if (src_nid == dst_nid)
7042
		return -1;
7043

7044 7045 7046 7047 7048 7049 7050
	/* 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;
	}
7051

7052 7053
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
7054
		return 0;
7055

7056
	/* Leaving a core idle is often worse than degrading locality. */
7057
	if (env->idle == CPU_IDLE)
7058 7059
		return -1;

7060
	dist = node_distance(src_nid, dst_nid);
7061
	if (numa_group) {
7062 7063
		src_weight = group_weight(p, src_nid, dist);
		dst_weight = group_weight(p, dst_nid, dist);
7064
	} else {
7065 7066
		src_weight = task_weight(p, src_nid, dist);
		dst_weight = task_weight(p, dst_nid, dist);
7067 7068
	}

7069
	return dst_weight < src_weight;
7070 7071
}

7072
#else
7073
static inline int migrate_degrades_locality(struct task_struct *p,
7074 7075
					     struct lb_env *env)
{
7076
	return -1;
7077
}
7078 7079
#endif

7080 7081 7082 7083
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
7084
int can_migrate_task(struct task_struct *p, struct lb_env *env)
7085
{
7086
	int tsk_cache_hot;
7087 7088 7089

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

7090 7091
	/*
	 * We do not migrate tasks that are:
7092
	 * 1) throttled_lb_pair, or
7093
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
7094 7095
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
7096
	 */
7097 7098 7099
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

7100
	if (!cpumask_test_cpu(env->dst_cpu, &p->cpus_allowed)) {
7101
		int cpu;
7102

7103
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
7104

7105 7106
		env->flags |= LBF_SOME_PINNED;

7107
		/*
7108
		 * Remember if this task can be migrated to any other CPU in
7109 7110 7111
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
7112 7113
		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
		 * already computed one in current iteration.
7114
		 */
7115
		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
7116 7117
			return 0;

7118
		/* Prevent to re-select dst_cpu via env's CPUs: */
7119
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
7120
			if (cpumask_test_cpu(cpu, &p->cpus_allowed)) {
7121
				env->flags |= LBF_DST_PINNED;
7122 7123 7124
				env->new_dst_cpu = cpu;
				break;
			}
7125
		}
7126

7127 7128
		return 0;
	}
7129 7130

	/* Record that we found atleast one task that could run on dst_cpu */
7131
	env->flags &= ~LBF_ALL_PINNED;
7132

7133
	if (task_running(env->src_rq, p)) {
7134
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
7135 7136 7137 7138 7139
		return 0;
	}

	/*
	 * Aggressive migration if:
7140 7141 7142
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
7143
	 */
7144 7145 7146
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
7147

7148
	if (tsk_cache_hot <= 0 ||
7149
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
7150
		if (tsk_cache_hot == 1) {
7151 7152
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
7153
		}
7154 7155 7156
		return 1;
	}

7157
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
7158
	return 0;
7159 7160
}

7161
/*
7162 7163 7164 7165 7166 7167 7168
 * 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;
7169
	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
7170 7171 7172
	set_task_cpu(p, env->dst_cpu);
}

7173
/*
7174
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
7175 7176
 * part of active balancing operations within "domain".
 *
7177
 * Returns a task if successful and NULL otherwise.
7178
 */
7179
static struct task_struct *detach_one_task(struct lb_env *env)
7180
{
7181
	struct task_struct *p;
7182

7183 7184
	lockdep_assert_held(&env->src_rq->lock);

7185 7186
	list_for_each_entry_reverse(p,
			&env->src_rq->cfs_tasks, se.group_node) {
7187 7188
		if (!can_migrate_task(p, env))
			continue;
7189

7190
		detach_task(p, env);
7191

7192
		/*
7193
		 * Right now, this is only the second place where
7194
		 * lb_gained[env->idle] is updated (other is detach_tasks)
7195
		 * so we can safely collect stats here rather than
7196
		 * inside detach_tasks().
7197
		 */
7198
		schedstat_inc(env->sd->lb_gained[env->idle]);
7199
		return p;
7200
	}
7201
	return NULL;
7202 7203
}

7204 7205
static const unsigned int sched_nr_migrate_break = 32;

7206
/*
7207 7208
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
7209
 *
7210
 * Returns number of detached tasks if successful and 0 otherwise.
7211
 */
7212
static int detach_tasks(struct lb_env *env)
7213
{
7214 7215
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
7216
	unsigned long load;
7217 7218 7219
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
7220

7221
	if (env->imbalance <= 0)
7222
		return 0;
7223

7224
	while (!list_empty(tasks)) {
7225 7226 7227 7228 7229 7230 7231
		/*
		 * 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;

7232
		p = list_last_entry(tasks, struct task_struct, se.group_node);
7233

7234 7235
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
7236
		if (env->loop > env->loop_max)
7237
			break;
7238 7239

		/* take a breather every nr_migrate tasks */
7240
		if (env->loop > env->loop_break) {
7241
			env->loop_break += sched_nr_migrate_break;
7242
			env->flags |= LBF_NEED_BREAK;
7243
			break;
7244
		}
7245

7246
		if (!can_migrate_task(p, env))
7247 7248 7249
			goto next;

		load = task_h_load(p);
7250

7251
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
7252 7253
			goto next;

7254
		if ((load / 2) > env->imbalance)
7255
			goto next;
7256

7257 7258 7259 7260
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
7261
		env->imbalance -= load;
7262 7263

#ifdef CONFIG_PREEMPT
7264 7265
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
7266
		 * kernels will stop after the first task is detached to minimize
7267 7268
		 * the critical section.
		 */
7269
		if (env->idle == CPU_NEWLY_IDLE)
7270
			break;
7271 7272
#endif

7273 7274 7275 7276
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
7277
		if (env->imbalance <= 0)
7278
			break;
7279 7280 7281

		continue;
next:
7282
		list_move(&p->se.group_node, tasks);
7283
	}
7284

7285
	/*
7286 7287 7288
	 * 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().
7289
	 */
7290
	schedstat_add(env->sd->lb_gained[env->idle], detached);
7291

7292 7293 7294 7295 7296 7297 7298 7299 7300 7301 7302
	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);
7303
	activate_task(rq, p, ENQUEUE_NOCLOCK);
7304
	p->on_rq = TASK_ON_RQ_QUEUED;
7305 7306 7307 7308 7309 7310 7311 7312 7313
	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)
{
7314 7315 7316
	struct rq_flags rf;

	rq_lock(rq, &rf);
7317
	update_rq_clock(rq);
7318
	attach_task(rq, p);
7319
	rq_unlock(rq, &rf);
7320 7321 7322 7323 7324 7325 7326 7327 7328 7329
}

/*
 * 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;
7330
	struct rq_flags rf;
7331

7332
	rq_lock(env->dst_rq, &rf);
7333
	update_rq_clock(env->dst_rq);
7334 7335 7336 7337

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
7338

7339 7340 7341
		attach_task(env->dst_rq, p);
	}

7342
	rq_unlock(env->dst_rq, &rf);
7343 7344
}

7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355
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;
}

7356
static inline bool others_have_blocked(struct rq *rq)
7357 7358 7359 7360
{
	if (READ_ONCE(rq->avg_rt.util_avg))
		return true;

7361 7362 7363
	if (READ_ONCE(rq->avg_dl.util_avg))
		return true;

7364
#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
7365 7366 7367 7368
	if (READ_ONCE(rq->avg_irq.util_avg))
		return true;
#endif

7369 7370 7371
	return false;
}

7372 7373
#ifdef CONFIG_FAIR_GROUP_SCHED

7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384
static inline bool cfs_rq_is_decayed(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->load.weight)
		return false;

	if (cfs_rq->avg.load_sum)
		return false;

	if (cfs_rq->avg.util_sum)
		return false;

7385
	if (cfs_rq->avg.runnable_load_sum)
7386 7387 7388 7389 7390
		return false;

	return true;
}

7391
static void update_blocked_averages(int cpu)
7392 7393
{
	struct rq *rq = cpu_rq(cpu);
7394
	struct cfs_rq *cfs_rq, *pos;
7395
	const struct sched_class *curr_class;
7396
	struct rq_flags rf;
7397
	bool done = true;
7398

7399
	rq_lock_irqsave(rq, &rf);
7400
	update_rq_clock(rq);
7401

7402 7403 7404 7405
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
7406
	for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) {
7407 7408
		struct sched_entity *se;

7409 7410 7411
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
7412

7413
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
7414
			update_tg_load_avg(cfs_rq, 0);
7415

7416 7417 7418
		/* Propagate pending load changes to the parent, if any: */
		se = cfs_rq->tg->se[cpu];
		if (se && !skip_blocked_update(se))
7419
			update_load_avg(cfs_rq_of(se), se, 0);
7420 7421 7422 7423 7424 7425 7426

		/*
		 * There can be a lot of idle CPU cgroups.  Don't let fully
		 * decayed cfs_rqs linger on the list.
		 */
		if (cfs_rq_is_decayed(cfs_rq))
			list_del_leaf_cfs_rq(cfs_rq);
7427 7428 7429

		/* Don't need periodic decay once load/util_avg are null */
		if (cfs_rq_has_blocked(cfs_rq))
7430
			done = false;
7431
	}
7432 7433 7434 7435

	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);
7436
	update_irq_load_avg(rq, 0);
7437
	/* Don't need periodic decay once load/util_avg are null */
7438
	if (others_have_blocked(rq))
7439
		done = false;
7440 7441 7442

#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7443 7444
	if (done)
		rq->has_blocked_load = 0;
7445
#endif
7446
	rq_unlock_irqrestore(rq, &rf);
7447 7448
}

7449
/*
7450
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
7451 7452 7453
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
7454
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
7455
{
7456 7457
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
7458
	unsigned long now = jiffies;
7459
	unsigned long load;
7460

7461
	if (cfs_rq->last_h_load_update == now)
7462 7463
		return;

7464 7465 7466 7467 7468 7469 7470
	cfs_rq->h_load_next = NULL;
	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
		cfs_rq->h_load_next = se;
		if (cfs_rq->last_h_load_update == now)
			break;
	}
7471

7472
	if (!se) {
7473
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
7474 7475 7476 7477 7478
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
7479 7480
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
7481 7482 7483 7484
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
7485 7486
}

7487
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
7488
{
7489
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
7490

7491
	update_cfs_rq_h_load(cfs_rq);
7492
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
7493
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
7494 7495
}
#else
7496
static inline void update_blocked_averages(int cpu)
7497
{
7498 7499
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
7500
	const struct sched_class *curr_class;
7501
	struct rq_flags rf;
7502

7503
	rq_lock_irqsave(rq, &rf);
7504
	update_rq_clock(rq);
7505
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
7506 7507 7508 7509

	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);
7510
	update_irq_load_avg(rq, 0);
7511 7512
#ifdef CONFIG_NO_HZ_COMMON
	rq->last_blocked_load_update_tick = jiffies;
7513
	if (!cfs_rq_has_blocked(cfs_rq) && !others_have_blocked(rq))
7514
		rq->has_blocked_load = 0;
7515
#endif
7516
	rq_unlock_irqrestore(rq, &rf);
7517 7518
}

7519
static unsigned long task_h_load(struct task_struct *p)
7520
{
7521
	return p->se.avg.load_avg;
7522
}
P
Peter Zijlstra 已提交
7523
#endif
7524 7525

/********** Helpers for find_busiest_group ************************/
7526 7527 7528 7529 7530 7531 7532

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

7533 7534 7535 7536 7537 7538 7539
/*
 * 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
7540
	unsigned long load_per_task;
7541
	unsigned long group_capacity;
7542
	unsigned long group_util; /* Total utilization of the group */
7543 7544 7545
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
7546
	enum group_type group_type;
7547
	int group_no_capacity;
7548 7549 7550 7551
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
7552 7553
};

J
Joonsoo Kim 已提交
7554 7555 7556 7557 7558 7559 7560
/*
 * 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 */
7561
	unsigned long total_running;
J
Joonsoo Kim 已提交
7562
	unsigned long total_load;	/* Total load of all groups in sd */
7563
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
7564 7565 7566
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
7567
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
7568 7569
};

7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580
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,
7581
		.total_running = 0UL,
7582
		.total_load = 0UL,
7583
		.total_capacity = 0UL,
7584 7585
		.busiest_stat = {
			.avg_load = 0UL,
7586 7587
			.sum_nr_running = 0,
			.group_type = group_other,
7588 7589 7590 7591
		},
	};
}

7592 7593 7594
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
7595
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
7596 7597
 *
 * Return: The load index.
7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619
 */
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;
}

7620
static unsigned long scale_rt_capacity(struct sched_domain *sd, int cpu)
7621 7622
{
	struct rq *rq = cpu_rq(cpu);
7623
	unsigned long max = arch_scale_cpu_capacity(sd, cpu);
7624 7625
	unsigned long used, free;
	unsigned long irq;
7626

7627
	irq = cpu_util_irq(rq);
7628

7629 7630
	if (unlikely(irq >= max))
		return 1;
7631

7632 7633
	used = READ_ONCE(rq->avg_rt.util_avg);
	used += READ_ONCE(rq->avg_dl.util_avg);
7634

7635 7636
	if (unlikely(used >= max))
		return 1;
7637

7638
	free = max - used;
7639 7640

	return scale_irq_capacity(free, irq, max);
7641 7642
}

7643
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
7644
{
7645
	unsigned long capacity = scale_rt_capacity(sd, cpu);
7646 7647
	struct sched_group *sdg = sd->groups;

7648
	cpu_rq(cpu)->cpu_capacity_orig = arch_scale_cpu_capacity(sd, cpu);
7649

7650 7651
	if (!capacity)
		capacity = 1;
7652

7653 7654
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
7655
	sdg->sgc->min_capacity = capacity;
7656 7657
}

7658
void update_group_capacity(struct sched_domain *sd, int cpu)
7659 7660 7661
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
7662
	unsigned long capacity, min_capacity;
7663 7664 7665 7666
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
7667
	sdg->sgc->next_update = jiffies + interval;
7668 7669

	if (!child) {
7670
		update_cpu_capacity(sd, cpu);
7671 7672 7673
		return;
	}

7674
	capacity = 0;
7675
	min_capacity = ULONG_MAX;
7676

P
Peter Zijlstra 已提交
7677 7678 7679 7680 7681 7682
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

7683
		for_each_cpu(cpu, sched_group_span(sdg)) {
7684
			struct sched_group_capacity *sgc;
7685
			struct rq *rq = cpu_rq(cpu);
7686

7687
			/*
7688
			 * build_sched_domains() -> init_sched_groups_capacity()
7689 7690 7691
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
7692 7693
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
7694
			 *
7695
			 * This avoids capacity from being 0 and
7696 7697 7698
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
7699
				capacity += capacity_of(cpu);
7700 7701 7702
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
7703
			}
7704

7705
			min_capacity = min(capacity, min_capacity);
7706
		}
P
Peter Zijlstra 已提交
7707 7708 7709 7710
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
7711
		 */
P
Peter Zijlstra 已提交
7712 7713 7714

		group = child->groups;
		do {
7715 7716 7717 7718
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
7719 7720 7721
			group = group->next;
		} while (group != child->groups);
	}
7722

7723
	sdg->sgc->capacity = capacity;
7724
	sdg->sgc->min_capacity = min_capacity;
7725 7726
}

7727
/*
7728 7729 7730
 * 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
7731 7732
 */
static inline int
7733
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
7734
{
7735 7736
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
7737 7738
}

7739 7740
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
7741
 * groups is inadequate due to ->cpus_allowed constraints.
7742
 *
7743 7744
 * 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.
7745 7746
 * Something like:
 *
7747 7748
 *	{ 0 1 2 3 } { 4 5 6 7 }
 *	        *     * * *
7749 7750 7751
 *
 * 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
7752
 * cpu 3 and leave one of the CPUs in the second group unused.
7753 7754
 *
 * The current solution to this issue is detecting the skew in the first group
7755 7756
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
7757 7758
 *
 * When this is so detected; this group becomes a candidate for busiest; see
7759
 * update_sd_pick_busiest(). And calculate_imbalance() and
7760
 * find_busiest_group() avoid some of the usual balance conditions to allow it
7761 7762 7763 7764 7765 7766 7767
 * 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.
 */

7768
static inline int sg_imbalanced(struct sched_group *group)
7769
{
7770
	return group->sgc->imbalance;
7771 7772
}

7773
/*
7774 7775 7776
 * 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
7777 7778
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
7779 7780 7781 7782 7783
 * 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.
7784
 */
7785 7786
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
7787
{
7788 7789
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
7790

7791
	if ((sgs->group_capacity * 100) >
7792
			(sgs->group_util * env->sd->imbalance_pct))
7793
		return true;
7794

7795 7796 7797 7798 7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810
	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;
7811

7812
	if ((sgs->group_capacity * 100) <
7813
			(sgs->group_util * env->sd->imbalance_pct))
7814
		return true;
7815

7816
	return false;
7817 7818
}

7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829
/*
 * 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;
}

7830 7831 7832
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7833
{
7834
	if (sgs->group_no_capacity)
7835 7836 7837 7838 7839 7840 7841 7842
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7843
static bool update_nohz_stats(struct rq *rq, bool force)
7844 7845 7846 7847
{
#ifdef CONFIG_NO_HZ_COMMON
	unsigned int cpu = rq->cpu;

7848 7849 7850
	if (!rq->has_blocked_load)
		return false;

7851
	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
7852
		return false;
7853

7854
	if (!force && !time_after(jiffies, rq->last_blocked_load_update_tick))
7855
		return true;
7856 7857

	update_blocked_averages(cpu);
7858 7859 7860 7861

	return rq->has_blocked_load;
#else
	return false;
7862 7863 7864
#endif
}

7865 7866
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7867
 * @env: The load balancing environment.
7868 7869 7870 7871
 * @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.
7872
 * @overload: Indicate more than one runnable task for any CPU.
7873
 */
7874 7875
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7876 7877
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7878
{
7879
	unsigned long load;
7880
	int i, nr_running;
7881

7882 7883
	memset(sgs, 0, sizeof(*sgs));

7884
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
7885 7886
		struct rq *rq = cpu_rq(i);

7887
		if ((env->flags & LBF_NOHZ_STATS) && update_nohz_stats(rq, false))
7888
			env->flags |= LBF_NOHZ_AGAIN;
7889

7890
		/* Bias balancing toward CPUs of our domain: */
7891
		if (local_group)
7892
			load = target_load(i, load_idx);
7893
		else
7894 7895 7896
			load = source_load(i, load_idx);

		sgs->group_load += load;
7897
		sgs->group_util += cpu_util(i);
7898
		sgs->sum_nr_running += rq->cfs.h_nr_running;
7899

7900 7901
		nr_running = rq->nr_running;
		if (nr_running > 1)
7902 7903
			*overload = true;

7904 7905 7906 7907
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
7908
		sgs->sum_weighted_load += weighted_cpuload(rq);
7909 7910 7911 7912
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
7913
			sgs->idle_cpus++;
7914 7915
	}

7916 7917
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
7918
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
7919

7920
	if (sgs->sum_nr_running)
7921
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
7922

7923
	sgs->group_weight = group->group_weight;
7924

7925
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
7926
	sgs->group_type = group_classify(group, sgs);
7927 7928
}

7929 7930
/**
 * update_sd_pick_busiest - return 1 on busiest group
7931
 * @env: The load balancing environment.
7932 7933
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
7934
 * @sgs: sched_group statistics
7935 7936 7937
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
7938 7939 7940
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
7941
 */
7942
static bool update_sd_pick_busiest(struct lb_env *env,
7943 7944
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
7945
				   struct sg_lb_stats *sgs)
7946
{
7947
	struct sg_lb_stats *busiest = &sds->busiest_stat;
7948

7949
	if (sgs->group_type > busiest->group_type)
7950 7951
		return true;

7952 7953 7954 7955 7956 7957
	if (sgs->group_type < busiest->group_type)
		return false;

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

7958 7959 7960 7961 7962 7963 7964 7965 7966 7967 7968 7969 7970 7971
	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:
7972 7973
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
7974 7975
		return true;

7976
	/* No ASYM_PACKING if target CPU is already busy */
7977 7978
	if (env->idle == CPU_NOT_IDLE)
		return true;
7979
	/*
T
Tim Chen 已提交
7980 7981 7982
	 * 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.
7983
	 */
T
Tim Chen 已提交
7984 7985
	if (sgs->sum_nr_running &&
	    sched_asym_prefer(env->dst_cpu, sg->asym_prefer_cpu)) {
7986 7987 7988
		if (!sds->busiest)
			return true;

7989
		/* Prefer to move from lowest priority CPU's work */
T
Tim Chen 已提交
7990 7991
		if (sched_asym_prefer(sds->busiest->asym_prefer_cpu,
				      sg->asym_prefer_cpu))
7992 7993 7994 7995 7996 7997
			return true;
	}

	return false;
}

7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008 8009 8010 8011 8012 8013 8014 8015 8016 8017 8018 8019 8020 8021 8022 8023 8024 8025 8026 8027
#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 */

8028
/**
8029
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
8030
 * @env: The load balancing environment.
8031 8032
 * @sds: variable to hold the statistics for this sched_domain.
 */
8033
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
8034
{
8035 8036
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
8037
	struct sg_lb_stats *local = &sds->local_stat;
J
Joonsoo Kim 已提交
8038
	struct sg_lb_stats tmp_sgs;
8039
	int load_idx, prefer_sibling = 0;
8040
	bool overload = false;
8041 8042 8043 8044

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

8045
#ifdef CONFIG_NO_HZ_COMMON
8046
	if (env->idle == CPU_NEWLY_IDLE && READ_ONCE(nohz.has_blocked))
8047 8048 8049
		env->flags |= LBF_NOHZ_STATS;
#endif

8050
	load_idx = get_sd_load_idx(env->sd, env->idle);
8051 8052

	do {
J
Joonsoo Kim 已提交
8053
		struct sg_lb_stats *sgs = &tmp_sgs;
8054 8055
		int local_group;

8056
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_span(sg));
J
Joonsoo Kim 已提交
8057 8058
		if (local_group) {
			sds->local = sg;
8059
			sgs = local;
8060 8061

			if (env->idle != CPU_NEWLY_IDLE ||
8062 8063
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
8064
		}
8065

8066 8067
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
8068

8069 8070 8071
		if (local_group)
			goto next_group;

8072 8073
		/*
		 * In case the child domain prefers tasks go to siblings
8074
		 * first, lower the sg capacity so that we'll try
8075 8076
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
8077 8078 8079 8080
		 * 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).
8081
		 */
8082
		if (prefer_sibling && sds->local &&
8083 8084
		    group_has_capacity(env, local) &&
		    (sgs->sum_nr_running > local->sum_nr_running + 1)) {
8085
			sgs->group_no_capacity = 1;
8086
			sgs->group_type = group_classify(sg, sgs);
8087
		}
8088

8089
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
8090
			sds->busiest = sg;
J
Joonsoo Kim 已提交
8091
			sds->busiest_stat = *sgs;
8092 8093
		}

8094 8095
next_group:
		/* Now, start updating sd_lb_stats */
8096
		sds->total_running += sgs->sum_nr_running;
8097
		sds->total_load += sgs->group_load;
8098
		sds->total_capacity += sgs->group_capacity;
8099

8100
		sg = sg->next;
8101
	} while (sg != env->sd->groups);
8102

8103 8104 8105 8106 8107 8108 8109 8110 8111
#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

8112 8113
	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
8114 8115 8116 8117 8118 8119

	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;
	}
8120 8121 8122 8123
}

/**
 * check_asym_packing - Check to see if the group is packed into the
8124
 *			sched domain.
8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138
 *
 * 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.
 *
8139
 * Return: 1 when packing is required and a task should be moved to
8140
 * this CPU.  The amount of the imbalance is returned in env->imbalance.
8141
 *
8142
 * @env: The load balancing environment.
8143 8144
 * @sds: Statistics of the sched_domain which is to be packed
 */
8145
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
8146 8147 8148
{
	int busiest_cpu;

8149
	if (!(env->sd->flags & SD_ASYM_PACKING))
8150 8151
		return 0;

8152 8153 8154
	if (env->idle == CPU_NOT_IDLE)
		return 0;

8155 8156 8157
	if (!sds->busiest)
		return 0;

T
Tim Chen 已提交
8158 8159
	busiest_cpu = sds->busiest->asym_prefer_cpu;
	if (sched_asym_prefer(busiest_cpu, env->dst_cpu))
8160 8161
		return 0;

8162
	env->imbalance = DIV_ROUND_CLOSEST(
8163
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
8164
		SCHED_CAPACITY_SCALE);
8165

8166
	return 1;
8167 8168 8169 8170 8171 8172
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
8173
 * @env: The load balancing environment.
8174 8175
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
8176 8177
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8178
{
8179
	unsigned long tmp, capa_now = 0, capa_move = 0;
8180
	unsigned int imbn = 2;
8181
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
8182
	struct sg_lb_stats *local, *busiest;
8183

J
Joonsoo Kim 已提交
8184 8185
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8186

J
Joonsoo Kim 已提交
8187 8188 8189 8190
	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;
8191

J
Joonsoo Kim 已提交
8192
	scaled_busy_load_per_task =
8193
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8194
		busiest->group_capacity;
J
Joonsoo Kim 已提交
8195

8196 8197
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
8198
		env->imbalance = busiest->load_per_task;
8199 8200 8201 8202 8203
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
8204
	 * however we may be able to increase total CPU capacity used by
8205 8206 8207
	 * moving them.
	 */

8208
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
8209
			min(busiest->load_per_task, busiest->avg_load);
8210
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
8211
			min(local->load_per_task, local->avg_load);
8212
	capa_now /= SCHED_CAPACITY_SCALE;
8213 8214

	/* Amount of load we'd subtract */
8215
	if (busiest->avg_load > scaled_busy_load_per_task) {
8216
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
8217
			    min(busiest->load_per_task,
8218
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
8219
	}
8220 8221

	/* Amount of load we'd add */
8222
	if (busiest->avg_load * busiest->group_capacity <
8223
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
8224 8225
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
8226
	} else {
8227
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8228
		      local->group_capacity;
J
Joonsoo Kim 已提交
8229
	}
8230
	capa_move += local->group_capacity *
8231
		    min(local->load_per_task, local->avg_load + tmp);
8232
	capa_move /= SCHED_CAPACITY_SCALE;
8233 8234

	/* Move if we gain throughput */
8235
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
8236
		env->imbalance = busiest->load_per_task;
8237 8238 8239 8240 8241
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
8242
 * @env: load balance environment
8243 8244
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
8245
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8246
{
8247
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
8248 8249 8250 8251
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8252

8253
	if (busiest->group_type == group_imbalanced) {
8254 8255
		/*
		 * In the group_imb case we cannot rely on group-wide averages
8256
		 * to ensure CPU-load equilibrium, look at wider averages. XXX
8257
		 */
J
Joonsoo Kim 已提交
8258 8259
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
8260 8261
	}

8262
	/*
8263 8264 8265 8266
	 * 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:
8267
	 */
8268 8269
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
8270 8271
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
8272 8273
	}

8274
	/*
8275
	 * If there aren't any idle CPUs, avoid creating some.
8276 8277 8278
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
8279
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
8280
		if (load_above_capacity > busiest->group_capacity) {
8281
			load_above_capacity -= busiest->group_capacity;
8282
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
8283 8284
			load_above_capacity /= busiest->group_capacity;
		} else
8285
			load_above_capacity = ~0UL;
8286 8287 8288
	}

	/*
8289
	 * We're trying to get all the CPUs to the average_load, so we don't
8290
	 * want to push ourselves above the average load, nor do we wish to
8291
	 * reduce the max loaded CPU below the average load. At the same time,
8292 8293
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
8294
	 */
8295
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
8296 8297

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
8298
	env->imbalance = min(
8299 8300
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
8301
	) / SCHED_CAPACITY_SCALE;
8302 8303 8304

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
8305
	 * there is no guarantee that any tasks will be moved so we'll have
8306 8307 8308
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
8309
	if (env->imbalance < busiest->load_per_task)
8310
		return fix_small_imbalance(env, sds);
8311
}
8312

8313 8314 8315 8316
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
8317
 * if there is an imbalance.
8318 8319 8320 8321
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
8322
 * @env: The load balancing environment.
8323
 *
8324
 * Return:	- The busiest group if imbalance exists.
8325
 */
J
Joonsoo Kim 已提交
8326
static struct sched_group *find_busiest_group(struct lb_env *env)
8327
{
J
Joonsoo Kim 已提交
8328
	struct sg_lb_stats *local, *busiest;
8329 8330
	struct sd_lb_stats sds;

8331
	init_sd_lb_stats(&sds);
8332 8333 8334 8335 8336

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
8337
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
8338 8339
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
8340

8341
	/* ASYM feature bypasses nice load balance check */
8342
	if (check_asym_packing(env, &sds))
8343 8344
		return sds.busiest;

8345
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
8346
	if (!sds.busiest || busiest->sum_nr_running == 0)
8347 8348
		goto out_balanced;

8349
	/* XXX broken for overlapping NUMA groups */
8350 8351
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
8352

P
Peter Zijlstra 已提交
8353 8354
	/*
	 * If the busiest group is imbalanced the below checks don't
8355
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
8356 8357
	 * isn't true due to cpus_allowed constraints and the like.
	 */
8358
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
8359 8360
		goto force_balance;

8361 8362 8363 8364 8365
	/*
	 * 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) &&
8366
	    busiest->group_no_capacity)
8367 8368
		goto force_balance;

8369
	/*
8370
	 * If the local group is busier than the selected busiest group
8371 8372
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
8373
	if (local->avg_load >= busiest->avg_load)
8374 8375
		goto out_balanced;

8376 8377 8378 8379
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
8380
	if (local->avg_load >= sds.avg_load)
8381 8382
		goto out_balanced;

8383
	if (env->idle == CPU_IDLE) {
8384
		/*
8385
		 * This CPU is idle. If the busiest group is not overloaded
8386
		 * and there is no imbalance between this and busiest group
8387
		 * wrt idle CPUs, it is balanced. The imbalance becomes
8388 8389
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
8390
		 */
8391 8392
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
8393
			goto out_balanced;
8394 8395 8396 8397 8398
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
8399 8400
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
8401
			goto out_balanced;
8402
	}
8403

8404
force_balance:
8405
	/* Looks like there is an imbalance. Compute it */
8406
	calculate_imbalance(env, &sds);
8407
	return env->imbalance ? sds.busiest : NULL;
8408 8409

out_balanced:
8410
	env->imbalance = 0;
8411 8412 8413 8414
	return NULL;
}

/*
8415
 * find_busiest_queue - find the busiest runqueue among the CPUs in the group.
8416
 */
8417
static struct rq *find_busiest_queue(struct lb_env *env,
8418
				     struct sched_group *group)
8419 8420
{
	struct rq *busiest = NULL, *rq;
8421
	unsigned long busiest_load = 0, busiest_capacity = 1;
8422 8423
	int i;

8424
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8425
		unsigned long capacity, wl;
8426 8427 8428 8429
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
8430

8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451 8452
		/*
		 * 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;

8453
		capacity = capacity_of(i);
8454

8455
		wl = weighted_cpuload(rq);
8456

8457 8458
		/*
		 * When comparing with imbalance, use weighted_cpuload()
8459
		 * which is not scaled with the CPU capacity.
8460
		 */
8461 8462 8463

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
8464 8465
			continue;

8466
		/*
8467 8468 8469
		 * 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
8470
		 * potentially running at a lower capacity.
8471
		 *
8472
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
8473
		 * multiplication to rid ourselves of the division works out
8474 8475
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
8476
		 */
8477
		if (wl * busiest_capacity > busiest_load * capacity) {
8478
			busiest_load = wl;
8479
			busiest_capacity = capacity;
8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492
			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

8493
static int need_active_balance(struct lb_env *env)
8494
{
8495 8496 8497
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
8498 8499 8500

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
T
Tim Chen 已提交
8501 8502
		 * lower priority CPUs in order to pack all tasks in the
		 * highest priority CPUs.
8503
		 */
T
Tim Chen 已提交
8504 8505
		if ((sd->flags & SD_ASYM_PACKING) &&
		    sched_asym_prefer(env->dst_cpu, env->src_cpu))
8506
			return 1;
8507 8508
	}

8509 8510 8511 8512 8513 8514 8515 8516 8517 8518 8519 8520 8521
	/*
	 * 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;
	}

8522 8523 8524
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

8525 8526
static int active_load_balance_cpu_stop(void *data);

8527 8528 8529 8530 8531
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	int cpu, balance_cpu = -1;

8532 8533 8534 8535 8536 8537 8538
	/*
	 * 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;

8539
	/*
8540
	 * In the newly idle case, we will allow all the CPUs
8541 8542 8543 8544 8545
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

8546
	/* Try to find first idle CPU */
8547
	for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
8548
		if (!idle_cpu(cpu))
8549 8550 8551 8552 8553 8554 8555 8556 8557 8558
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
8559
	 * First idle CPU or the first CPU(busiest) in this sched group
8560 8561
	 * is eligible for doing load balancing at this and above domains.
	 */
8562
	return balance_cpu == env->dst_cpu;
8563 8564
}

8565 8566 8567 8568 8569 8570
/*
 * 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,
8571
			int *continue_balancing)
8572
{
8573
	int ld_moved, cur_ld_moved, active_balance = 0;
8574
	struct sched_domain *sd_parent = sd->parent;
8575 8576
	struct sched_group *group;
	struct rq *busiest;
8577
	struct rq_flags rf;
8578
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
8579

8580 8581
	struct lb_env env = {
		.sd		= sd,
8582 8583
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
8584
		.dst_grpmask    = sched_group_span(sd->groups),
8585
		.idle		= idle,
8586
		.loop_break	= sched_nr_migrate_break,
8587
		.cpus		= cpus,
8588
		.fbq_type	= all,
8589
		.tasks		= LIST_HEAD_INIT(env.tasks),
8590 8591
	};

8592
	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
8593

8594
	schedstat_inc(sd->lb_count[idle]);
8595 8596

redo:
8597 8598
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
8599
		goto out_balanced;
8600
	}
8601

8602
	group = find_busiest_group(&env);
8603
	if (!group) {
8604
		schedstat_inc(sd->lb_nobusyg[idle]);
8605 8606 8607
		goto out_balanced;
	}

8608
	busiest = find_busiest_queue(&env, group);
8609
	if (!busiest) {
8610
		schedstat_inc(sd->lb_nobusyq[idle]);
8611 8612 8613
		goto out_balanced;
	}

8614
	BUG_ON(busiest == env.dst_rq);
8615

8616
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
8617

8618 8619 8620
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

8621 8622 8623 8624 8625 8626 8627 8628
	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.
		 */
8629
		env.flags |= LBF_ALL_PINNED;
8630
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
8631

8632
more_balance:
8633
		rq_lock_irqsave(busiest, &rf);
8634
		update_rq_clock(busiest);
8635 8636 8637 8638 8639

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
8640
		cur_ld_moved = detach_tasks(&env);
8641 8642

		/*
8643 8644 8645 8646 8647
		 * 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.
8648
		 */
8649

8650
		rq_unlock(busiest, &rf);
8651 8652 8653 8654 8655 8656

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

8657
		local_irq_restore(rf.flags);
8658

8659 8660 8661 8662 8663
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

8664 8665 8666 8667
		/*
		 * 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
8668
		 * iterate on same src_cpu is dependent on number of CPUs in our
8669 8670 8671 8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682
		 * 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.
		 */
8683
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
8684

8685
			/* Prevent to re-select dst_cpu via env's CPUs */
8686 8687
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

8688
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
8689
			env.dst_cpu	 = env.new_dst_cpu;
8690
			env.flags	&= ~LBF_DST_PINNED;
8691 8692
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
8693

8694 8695 8696 8697 8698 8699
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
8700

8701 8702 8703 8704
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
8705
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
8706

8707
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
8708 8709 8710
				*group_imbalance = 1;
		}

8711
		/* All tasks on this runqueue were pinned by CPU affinity */
8712
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
8713
			cpumask_clear_cpu(cpu_of(busiest), cpus);
8714 8715 8716 8717 8718 8719 8720 8721 8722
			/*
			 * 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)) {
8723 8724
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
8725
				goto redo;
8726
			}
8727
			goto out_all_pinned;
8728 8729 8730 8731
		}
	}

	if (!ld_moved) {
8732
		schedstat_inc(sd->lb_failed[idle]);
8733 8734 8735 8736 8737 8738 8739 8740
		/*
		 * 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++;
8741

8742
		if (need_active_balance(&env)) {
8743 8744
			unsigned long flags;

8745 8746
			raw_spin_lock_irqsave(&busiest->lock, flags);

8747 8748 8749 8750
			/*
			 * Don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest CPU can't be
			 * moved to this_cpu:
8751
			 */
8752
			if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
8753 8754
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
8755
				env.flags |= LBF_ALL_PINNED;
8756 8757 8758
				goto out_one_pinned;
			}

8759 8760 8761 8762 8763
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
8764 8765 8766 8767 8768 8769
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
8770

8771
			if (active_balance) {
8772 8773 8774
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
8775
			}
8776

8777
			/* We've kicked active balancing, force task migration. */
8778 8779 8780 8781 8782 8783 8784 8785 8786 8787 8788 8789 8790
			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
8791
		 * detach_tasks).
8792 8793 8794 8795 8796 8797 8798 8799
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
8800 8801 8802 8803 8804 8805 8806 8807 8808 8809 8810 8811 8812 8813 8814 8815 8816
	/*
	 * We reach balance although we may have faced some affinity
	 * constraints. Clear the imbalance flag if it was set.
	 */
	if (sd_parent) {
		int *group_imbalance = &sd_parent->groups->sgc->imbalance;

		if (*group_imbalance)
			*group_imbalance = 0;
	}

out_all_pinned:
	/*
	 * We reach balance because all tasks are pinned at this level so
	 * we can't migrate them. Let the imbalance flag set so parent level
	 * can try to migrate them.
	 */
8817
	schedstat_inc(sd->lb_balanced[idle]);
8818 8819 8820 8821 8822

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
8823
	if (((env.flags & LBF_ALL_PINNED) &&
8824
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
8825 8826 8827
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

8828
	ld_moved = 0;
8829 8830 8831 8832
out:
	return ld_moved;
}

8833 8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848
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
8849
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
8850 8851 8852
{
	unsigned long interval, next;

8853 8854
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
8855 8856 8857 8858 8859 8860
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

8861
/*
8862
 * active_load_balance_cpu_stop is run by the CPU stopper. It pushes
8863 8864 8865
 * 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.
8866
 */
8867
static int active_load_balance_cpu_stop(void *data)
8868
{
8869 8870
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8871
	int target_cpu = busiest_rq->push_cpu;
8872
	struct rq *target_rq = cpu_rq(target_cpu);
8873
	struct sched_domain *sd;
8874
	struct task_struct *p = NULL;
8875
	struct rq_flags rf;
8876

8877
	rq_lock_irq(busiest_rq, &rf);
8878 8879 8880 8881 8882 8883 8884
	/*
	 * 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;
8885

8886
	/* Make sure the requested CPU hasn't gone down in the meantime: */
8887 8888 8889
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
8890 8891 8892

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
8893
		goto out_unlock;
8894 8895 8896 8897

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
8898
	 * Bjorn Helgaas on a 128-CPU setup.
8899 8900 8901 8902
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
8903
	rcu_read_lock();
8904 8905 8906 8907 8908 8909 8910
	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)) {
8911 8912
		struct lb_env env = {
			.sd		= sd,
8913 8914 8915 8916
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
8917
			.idle		= CPU_IDLE,
8918 8919 8920 8921 8922 8923 8924
			/*
			 * 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,
8925 8926
		};

8927
		schedstat_inc(sd->alb_count);
8928
		update_rq_clock(busiest_rq);
8929

8930
		p = detach_one_task(&env);
8931
		if (p) {
8932
			schedstat_inc(sd->alb_pushed);
8933 8934 8935
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
8936
			schedstat_inc(sd->alb_failed);
8937
		}
8938
	}
8939
	rcu_read_unlock();
8940 8941
out_unlock:
	busiest_rq->active_balance = 0;
8942
	rq_unlock(busiest_rq, &rf);
8943 8944 8945 8946 8947 8948

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

8949
	return 0;
8950 8951
}

8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972 8973 8974 8975 8976 8977 8978 8979 8980 8981 8982 8983 8984 8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044 9045 9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061 9062 9063 9064 9065 9066 9067 9068 9069
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
	}
}

9070 9071 9072 9073 9074
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

9075
#ifdef CONFIG_NO_HZ_COMMON
9076 9077 9078 9079 9080 9081
/*
 * 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.
 */
9082

9083
static inline int find_new_ilb(void)
9084
{
9085
	int ilb = cpumask_first(nohz.idle_cpus_mask);
9086

9087 9088 9089 9090
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
9091 9092
}

9093 9094 9095 9096 9097
/*
 * Kick a CPU to do the nohz balancing, if it is time for it. We pick the
 * nohz_load_balancer CPU (if there is one) otherwise fallback to any idle
 * CPU (if there is one).
 */
9098
static void kick_ilb(unsigned int flags)
9099 9100 9101 9102 9103
{
	int ilb_cpu;

	nohz.next_balance++;

9104
	ilb_cpu = find_new_ilb();
9105

9106 9107
	if (ilb_cpu >= nr_cpu_ids)
		return;
9108

9109
	flags = atomic_fetch_or(flags, nohz_flags(ilb_cpu));
P
Peter Zijlstra 已提交
9110
	if (flags & NOHZ_KICK_MASK)
9111
		return;
9112

9113 9114
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
9115
	 * This way we generate a sched IPI on the target CPU which
9116 9117 9118 9119
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
9120 9121 9122 9123 9124 9125 9126 9127 9128 9129 9130 9131 9132 9133 9134 9135 9136 9137 9138
}

/*
 * 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;
9139
	unsigned int flags = 0;
9140 9141 9142 9143 9144 9145 9146 9147

	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.
	 */
9148
	nohz_balance_exit_idle(rq);
9149 9150 9151 9152 9153 9154 9155 9156

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

9157 9158
	if (READ_ONCE(nohz.has_blocked) &&
	    time_after(now, READ_ONCE(nohz.next_blocked)))
9159 9160
		flags = NOHZ_STATS_KICK;

9161
	if (time_before(now, nohz.next_balance))
9162
		goto out;
9163 9164

	if (rq->nr_running >= 2) {
9165
		flags = NOHZ_KICK_MASK;
9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177
		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) {
9178
			flags = NOHZ_KICK_MASK;
9179 9180 9181 9182 9183 9184 9185 9186 9187
			goto unlock;
		}

	}

	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
9188
			flags = NOHZ_KICK_MASK;
9189 9190 9191 9192 9193 9194 9195 9196 9197 9198 9199 9200
			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)) {
9201
				flags = NOHZ_KICK_MASK;
9202 9203 9204 9205 9206 9207 9208
				goto unlock;
			}
		}
	}
unlock:
	rcu_read_unlock();
out:
9209 9210
	if (flags)
		kick_ilb(flags);
9211 9212
}

9213
static void set_cpu_sd_state_busy(int cpu)
9214
{
9215
	struct sched_domain *sd;
9216

9217 9218
	rcu_read_lock();
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
9219

9220 9221 9222 9223 9224 9225 9226
	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

	atomic_inc(&sd->shared->nr_busy_cpus);
unlock:
	rcu_read_unlock();
9227 9228
}

9229 9230 9231 9232 9233 9234 9235 9236 9237 9238 9239 9240 9241 9242 9243
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)
9244 9245 9246 9247
{
	struct sched_domain *sd;

	rcu_read_lock();
9248
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
9249 9250 9251 9252 9253

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

9254
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
9255
unlock:
9256 9257 9258
	rcu_read_unlock();
}

9259
/*
9260
 * This routine will record that the CPU is going idle with tick stopped.
9261
 * This info will be used in performing idle load balancing in the future.
9262
 */
9263
void nohz_balance_enter_idle(int cpu)
9264
{
9265 9266 9267 9268
	struct rq *rq = cpu_rq(cpu);

	SCHED_WARN_ON(cpu != smp_processor_id());

9269
	/* If this CPU is going down, then nothing needs to be done: */
9270 9271 9272
	if (!cpu_active(cpu))
		return;

9273
	/* Spare idle load balancing on CPUs that don't want to be disturbed: */
9274
	if (!housekeeping_cpu(cpu, HK_FLAG_SCHED))
9275 9276
		return;

9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289
	/*
	 * 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
	 */
9290
	if (rq->nohz_tick_stopped)
9291
		goto out;
9292

9293
	/* If we're a completely isolated CPU, we don't play: */
9294
	if (on_null_domain(rq))
9295 9296
		return;

9297 9298
	rq->nohz_tick_stopped = 1;

9299 9300
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
9301

9302 9303 9304 9305 9306 9307 9308
	/*
	 * 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();

9309
	set_cpu_sd_state_idle(cpu);
9310 9311 9312 9313 9314 9315 9316

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);
9317 9318 9319
}

/*
9320 9321 9322 9323 9324
 * 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.
9325
 */
9326 9327
static bool _nohz_idle_balance(struct rq *this_rq, unsigned int flags,
			       enum cpu_idle_type idle)
9328
{
9329
	/* Earliest time when we have to do rebalance again */
9330 9331
	unsigned long now = jiffies;
	unsigned long next_balance = now + 60*HZ;
9332
	bool has_blocked_load = false;
9333
	int update_next_balance = 0;
P
Peter Zijlstra 已提交
9334 9335
	int this_cpu = this_rq->cpu;
	int balance_cpu;
9336
	int ret = false;
P
Peter Zijlstra 已提交
9337
	struct rq *rq;
9338

P
Peter Zijlstra 已提交
9339
	SCHED_WARN_ON((flags & NOHZ_KICK_MASK) == NOHZ_BALANCE_KICK);
9340

9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356
	/*
	 * 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();

9357
	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
9358
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
9359 9360 9361
			continue;

		/*
9362 9363
		 * If this CPU gets work to do, stop the load balancing
		 * work being done for other CPUs. Next load
9364 9365
		 * balancing owner will pick it up.
		 */
9366 9367 9368 9369
		if (need_resched()) {
			has_blocked_load = true;
			goto abort;
		}
9370

V
Vincent Guittot 已提交
9371 9372
		rq = cpu_rq(balance_cpu);

9373
		has_blocked_load |= update_nohz_stats(rq, true);
9374

9375 9376 9377 9378 9379
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
9380 9381
			struct rq_flags rf;

9382
			rq_lock_irqsave(rq, &rf);
9383
			update_rq_clock(rq);
9384
			cpu_load_update_idle(rq);
9385
			rq_unlock_irqrestore(rq, &rf);
9386

P
Peter Zijlstra 已提交
9387 9388
			if (flags & NOHZ_BALANCE_KICK)
				rebalance_domains(rq, CPU_IDLE);
9389
		}
9390

9391 9392 9393 9394
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
9395
	}
9396

9397 9398 9399 9400 9401 9402
	/* 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
9403 9404 9405
	if (flags & NOHZ_BALANCE_KICK)
		rebalance_domains(this_rq, CPU_IDLE);

9406 9407 9408
	WRITE_ONCE(nohz.next_blocked,
		now + msecs_to_jiffies(LOAD_AVG_PERIOD));

9409 9410 9411
	/* The full idle balance loop has been done */
	ret = true;

9412 9413 9414 9415
abort:
	/* There is still blocked load, enable periodic update */
	if (has_blocked_load)
		WRITE_ONCE(nohz.has_blocked, 1);
9416

9417 9418 9419 9420 9421 9422 9423
	/*
	 * 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
9424

9425 9426 9427 9428 9429 9430 9431 9432 9433 9434 9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453
	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
9454
	return true;
9455
}
9456 9457 9458 9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470 9471 9472 9473 9474 9475 9476 9477 9478 9479 9480 9481 9482 9483 9484 9485 9486 9487 9488

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);
}

9489 9490 9491
#else /* !CONFIG_NO_HZ_COMMON */
static inline void nohz_balancer_kick(struct rq *rq) { }

9492
static inline bool nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
P
Peter Zijlstra 已提交
9493 9494 9495
{
	return false;
}
9496 9497

static inline void nohz_newidle_balance(struct rq *this_rq) { }
9498
#endif /* CONFIG_NO_HZ_COMMON */
9499

P
Peter Zijlstra 已提交
9500 9501 9502 9503 9504 9505 9506 9507 9508 9509 9510 9511 9512 9513 9514 9515 9516 9517 9518 9519 9520 9521 9522 9523 9524 9525 9526 9527 9528 9529 9530 9531 9532 9533
/*
 * 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) {
9534

P
Peter Zijlstra 已提交
9535 9536 9537 9538 9539 9540
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
			update_next_balance(sd, &next_balance);
		rcu_read_unlock();

9541 9542
		nohz_newidle_balance(this_rq);

P
Peter Zijlstra 已提交
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 9570 9571 9572 9573 9574 9575 9576 9577 9578 9579 9580 9581 9582 9583 9584 9585 9586 9587 9588 9589 9590 9591
		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;

9592
out:
P
Peter Zijlstra 已提交
9593 9594 9595 9596 9597 9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609 9610 9611 9612 9613 9614 9615 9616
	/*
	 * 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;
}

9617 9618 9619 9620
/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
9621
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
9622
{
9623
	struct rq *this_rq = this_rq();
9624
	enum cpu_idle_type idle = this_rq->idle_balance ?
9625 9626 9627
						CPU_IDLE : CPU_NOT_IDLE;

	/*
9628 9629
	 * If this CPU has a pending nohz_balance_kick, then do the
	 * balancing on behalf of the other idle CPUs whose ticks are
9630
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
9631
	 * give the idle CPUs a chance to load balance. Else we may
9632 9633
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
9634
	 */
P
Peter Zijlstra 已提交
9635 9636 9637 9638 9639
	if (nohz_idle_balance(this_rq, idle))
		return;

	/* normal load balance */
	update_blocked_averages(this_rq->cpu);
9640
	rebalance_domains(this_rq, idle);
9641 9642 9643 9644 9645
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
9646
void trigger_load_balance(struct rq *rq)
9647 9648
{
	/* Don't need to rebalance while attached to NULL domain */
9649 9650 9651 9652
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
9653
		raise_softirq(SCHED_SOFTIRQ);
9654 9655

	nohz_balancer_kick(rq);
9656 9657
}

9658 9659 9660
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
9661 9662

	update_runtime_enabled(rq);
9663 9664 9665 9666 9667
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
9668 9669 9670

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
9671 9672
}

9673
#endif /* CONFIG_SMP */
9674

9675
/*
9676 9677 9678 9679 9680 9681
 * 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.
9682
 */
P
Peter Zijlstra 已提交
9683
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
9684 9685 9686 9687 9688 9689
{
	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
9690
		entity_tick(cfs_rq, se, queued);
9691
	}
9692

9693
	if (static_branch_unlikely(&sched_numa_balancing))
9694
		task_tick_numa(rq, curr);
9695 9696 9697
}

/*
P
Peter Zijlstra 已提交
9698 9699 9700
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
9701
 */
P
Peter Zijlstra 已提交
9702
static void task_fork_fair(struct task_struct *p)
9703
{
9704 9705
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
9706
	struct rq *rq = this_rq();
9707
	struct rq_flags rf;
9708

9709
	rq_lock(rq, &rf);
9710 9711
	update_rq_clock(rq);

9712 9713
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
9714 9715
	if (curr) {
		update_curr(cfs_rq);
9716
		se->vruntime = curr->vruntime;
9717
	}
9718
	place_entity(cfs_rq, se, 1);
9719

P
Peter Zijlstra 已提交
9720
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
9721
		/*
9722 9723 9724
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
9725
		swap(curr->vruntime, se->vruntime);
9726
		resched_curr(rq);
9727
	}
9728

9729
	se->vruntime -= cfs_rq->min_vruntime;
9730
	rq_unlock(rq, &rf);
9731 9732
}

9733 9734 9735 9736
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
9737 9738
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
9739
{
9740
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
9741 9742
		return;

9743 9744 9745 9746 9747
	/*
	 * 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
9748
	if (rq->curr == p) {
9749
		if (p->prio > oldprio)
9750
			resched_curr(rq);
9751
	} else
9752
		check_preempt_curr(rq, p, 0);
9753 9754
}

9755
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
9756 9757 9758 9759
{
	struct sched_entity *se = &p->se;

	/*
9760 9761 9762 9763 9764 9765 9766 9767 9768 9769
	 * 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
9770
	 *
9771 9772 9773 9774
	 * - 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
9775
	 */
9776 9777
	if (!se->sum_exec_runtime ||
	    (p->state == TASK_WAKING && p->sched_remote_wakeup))
9778 9779 9780 9781 9782
		return true;

	return false;
}

9783 9784 9785 9786 9787 9788 9789 9790 9791 9792 9793 9794 9795 9796 9797 9798 9799 9800
#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;

9801
		update_load_avg(cfs_rq, se, UPDATE_TG);
9802 9803 9804 9805 9806 9807
	}
}
#else
static void propagate_entity_cfs_rq(struct sched_entity *se) { }
#endif

9808
static void detach_entity_cfs_rq(struct sched_entity *se)
9809 9810 9811
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

9812
	/* Catch up with the cfs_rq and remove our load when we leave */
9813
	update_load_avg(cfs_rq, se, 0);
9814
	detach_entity_load_avg(cfs_rq, se);
9815
	update_tg_load_avg(cfs_rq, false);
9816
	propagate_entity_cfs_rq(se);
P
Peter Zijlstra 已提交
9817 9818
}

9819
static void attach_entity_cfs_rq(struct sched_entity *se)
9820
{
9821
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
9822 9823

#ifdef CONFIG_FAIR_GROUP_SCHED
9824 9825 9826 9827 9828 9829
	/*
	 * 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
9830

9831
	/* Synchronize entity with its cfs_rq */
9832
	update_load_avg(cfs_rq, se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
9833
	attach_entity_load_avg(cfs_rq, se, 0);
9834
	update_tg_load_avg(cfs_rq, false);
9835
	propagate_entity_cfs_rq(se);
9836 9837 9838 9839 9840 9841 9842 9843 9844 9845 9846 9847 9848 9849 9850 9851 9852 9853 9854 9855 9856 9857 9858 9859 9860
}

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);
9861 9862 9863 9864

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
9865

9866 9867 9868 9869 9870 9871 9872 9873
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);
9874

9875
	if (task_on_rq_queued(p)) {
9876
		/*
9877 9878 9879
		 * 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.
9880
		 */
9881 9882 9883 9884
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
9885
	}
9886 9887
}

9888 9889 9890 9891 9892 9893 9894 9895 9896
/* 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;

9897 9898 9899 9900 9901 9902 9903
	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);
	}
9904 9905
}

9906 9907
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
9908
	cfs_rq->tasks_timeline = RB_ROOT_CACHED;
9909 9910 9911 9912
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
9913
#ifdef CONFIG_SMP
9914
	raw_spin_lock_init(&cfs_rq->removed.lock);
9915
#endif
9916 9917
}

P
Peter Zijlstra 已提交
9918
#ifdef CONFIG_FAIR_GROUP_SCHED
9919 9920 9921 9922 9923 9924 9925 9926
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;
}

9927
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
9928
{
9929
	detach_task_cfs_rq(p);
9930
	set_task_rq(p, task_cpu(p));
9931 9932 9933 9934 9935

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
9936
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
9937
}
9938

9939 9940 9941 9942 9943 9944 9945 9946 9947 9948 9949 9950 9951
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;
	}
}

9952 9953 9954 9955 9956 9957 9958 9959 9960
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]);
9961
		if (tg->se)
9962 9963 9964 9965 9966 9967 9968 9969 9970 9971
			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;
9972
	struct cfs_rq *cfs_rq;
9973 9974
	int i;

K
Kees Cook 已提交
9975
	tg->cfs_rq = kcalloc(nr_cpu_ids, sizeof(cfs_rq), GFP_KERNEL);
9976 9977
	if (!tg->cfs_rq)
		goto err;
K
Kees Cook 已提交
9978
	tg->se = kcalloc(nr_cpu_ids, sizeof(se), GFP_KERNEL);
9979 9980 9981 9982 9983 9984 9985 9986 9987 9988 9989 9990 9991 9992 9993 9994 9995 9996 9997 9998
	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]);
9999
		init_entity_runnable_average(se);
10000 10001 10002 10003 10004 10005 10006 10007 10008 10009
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

10010 10011 10012 10013 10014 10015 10016 10017 10018 10019 10020
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];

		raw_spin_lock_irq(&rq->lock);
10021
		update_rq_clock(rq);
10022
		attach_entity_cfs_rq(se);
10023
		sync_throttle(tg, i);
10024 10025 10026 10027
		raw_spin_unlock_irq(&rq->lock);
	}
}

10028
void unregister_fair_sched_group(struct task_group *tg)
10029 10030
{
	unsigned long flags;
10031 10032
	struct rq *rq;
	int cpu;
10033

10034 10035 10036
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
10037

10038 10039 10040 10041 10042 10043 10044 10045 10046 10047 10048 10049 10050
		/*
		 * 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);
	}
10051 10052 10053 10054 10055 10056 10057 10058 10059 10060 10061 10062 10063 10064 10065 10066 10067 10068 10069
}

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
10070
	if (!parent) {
10071
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
10072 10073
		se->depth = 0;
	} else {
10074
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
10075 10076
		se->depth = parent->depth + 1;
	}
10077 10078

	se->my_q = cfs_rq;
10079 10080
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
10081 10082 10083 10084 10085 10086 10087 10088 10089 10090 10091 10092 10093 10094 10095 10096 10097 10098 10099 10100 10101 10102 10103 10104
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;

	/*
	 * We can't change the weight of the root cgroup.
	 */
	if (!tg->se[0])
		return -EINVAL;

	shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	mutex_lock(&shares_mutex);
	if (tg->shares == shares)
		goto done;

	tg->shares = shares;
	for_each_possible_cpu(i) {
		struct rq *rq = cpu_rq(i);
10105 10106
		struct sched_entity *se = tg->se[i];
		struct rq_flags rf;
10107 10108

		/* Propagate contribution to hierarchy */
10109
		rq_lock_irqsave(rq, &rf);
10110
		update_rq_clock(rq);
10111
		for_each_sched_entity(se) {
10112
			update_load_avg(cfs_rq_of(se), se, UPDATE_TG);
10113
			update_cfs_group(se);
10114
		}
10115
		rq_unlock_irqrestore(rq, &rf);
10116 10117 10118 10119 10120 10121 10122 10123 10124 10125 10126 10127 10128 10129 10130
	}

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;
}

10131 10132
void online_fair_sched_group(struct task_group *tg) { }

10133
void unregister_fair_sched_group(struct task_group *tg) { }
10134 10135 10136

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
10137

10138
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
10139 10140 10141 10142 10143 10144 10145 10146 10147
{
	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)
10148
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
10149 10150 10151 10152

	return rr_interval;
}

10153 10154 10155
/*
 * All the scheduling class methods:
 */
10156
const struct sched_class fair_sched_class = {
10157
	.next			= &idle_sched_class,
10158 10159 10160
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
10161
	.yield_to_task		= yield_to_task_fair,
10162

I
Ingo Molnar 已提交
10163
	.check_preempt_curr	= check_preempt_wakeup,
10164 10165 10166 10167

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

10168
#ifdef CONFIG_SMP
L
Li Zefan 已提交
10169
	.select_task_rq		= select_task_rq_fair,
10170
	.migrate_task_rq	= migrate_task_rq_fair,
10171

10172 10173
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
10174

10175
	.task_dead		= task_dead_fair,
10176
	.set_cpus_allowed	= set_cpus_allowed_common,
10177
#endif
10178

10179
	.set_curr_task          = set_curr_task_fair,
10180
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
10181
	.task_fork		= task_fork_fair,
10182 10183

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
10184
	.switched_from		= switched_from_fair,
10185
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
10186

10187 10188
	.get_rr_interval	= get_rr_interval_fair,

10189 10190
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
10191
#ifdef CONFIG_FAIR_GROUP_SCHED
10192
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
10193
#endif
10194 10195 10196
};

#ifdef CONFIG_SCHED_DEBUG
10197
void print_cfs_stats(struct seq_file *m, int cpu)
10198
{
10199
	struct cfs_rq *cfs_rq, *pos;
10200

10201
	rcu_read_lock();
10202
	for_each_leaf_cfs_rq_safe(cpu_rq(cpu), cfs_rq, pos)
10203
		print_cfs_rq(m, cpu, cfs_rq);
10204
	rcu_read_unlock();
10205
}
10206 10207 10208 10209 10210 10211 10212 10213 10214 10215 10216 10217 10218 10219 10220 10221 10222 10223 10224 10225 10226

#ifdef CONFIG_NUMA_BALANCING
void show_numa_stats(struct task_struct *p, struct seq_file *m)
{
	int node;
	unsigned long tsf = 0, tpf = 0, gsf = 0, gpf = 0;

	for_each_online_node(node) {
		if (p->numa_faults) {
			tsf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 0)];
			tpf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		if (p->numa_group) {
			gsf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 0)],
			gpf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		print_numa_stats(m, node, tsf, tpf, gsf, gpf);
	}
}
#endif /* CONFIG_NUMA_BALANCING */
#endif /* CONFIG_SCHED_DEBUG */
10227 10228 10229 10230 10231 10232

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

10233
#ifdef CONFIG_NO_HZ_COMMON
10234
	nohz.next_balance = jiffies;
10235
	nohz.next_blocked = jiffies;
10236 10237 10238 10239 10240
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}