fair.c 237.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18 19
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
21 22
 */

23
#include <linux/sched.h>
24
#include <linux/latencytop.h>
25
#include <linux/cpumask.h>
26
#include <linux/cpuidle.h>
27 28 29
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
30
#include <linux/mempolicy.h>
31
#include <linux/migrate.h>
32
#include <linux/task_work.h>
33 34 35 36

#include <trace/events/sched.h>

#include "sched.h"
A
Arjan van de Ven 已提交
37

38
/*
39
 * Targeted preemption latency for CPU-bound tasks:
40
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
41
 *
42
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
43 44 45
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
46
 *
I
Ingo Molnar 已提交
47 48
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
49
 */
50 51
unsigned int sysctl_sched_latency = 6000000ULL;
unsigned int normalized_sysctl_sched_latency = 6000000ULL;
52

53 54 55 56 57 58 59 60 61 62 63 64
/*
 * The initial- and re-scaling of tunables is configurable
 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
 *
 * Options are:
 * SCHED_TUNABLESCALING_NONE - unscaled, always *1
 * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
 * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
 */
enum sched_tunable_scaling sysctl_sched_tunable_scaling
	= SCHED_TUNABLESCALING_LOG;

65
/*
66
 * Minimal preemption granularity for CPU-bound tasks:
67
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
68
 */
69 70
unsigned int sysctl_sched_min_granularity = 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
71 72

/*
73 74
 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
 */
75
static unsigned int sched_nr_latency = 8;
76 77

/*
78
 * After fork, child runs first. If set to 0 (default) then
79
 * parent will (try to) run first.
80
 */
81
unsigned int sysctl_sched_child_runs_first __read_mostly;
82 83 84

/*
 * SCHED_OTHER wake-up granularity.
85
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
86 87 88 89 90
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
 */
91
unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
92
unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
93

94 95
const_debug unsigned int sysctl_sched_migration_cost = 500000UL;

96 97 98 99 100 101 102
/*
 * The exponential sliding  window over which load is averaged for shares
 * distribution.
 * (default: 10msec)
 */
unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;

103 104 105 106 107 108 109 110 111 112 113 114 115 116
#ifdef CONFIG_CFS_BANDWIDTH
/*
 * Amount of runtime to allocate from global (tg) to local (per-cfs_rq) pool
 * each time a cfs_rq requests quota.
 *
 * Note: in the case that the slice exceeds the runtime remaining (either due
 * to consumption or the quota being specified to be smaller than the slice)
 * we will always only issue the remaining available time.
 *
 * default: 5 msec, units: microseconds
  */
unsigned int sysctl_sched_cfs_bandwidth_slice = 5000UL;
#endif

117 118 119 120 121 122
/*
 * The margin used when comparing utilization with CPU capacity:
 * util * 1024 < capacity * margin
 */
unsigned int capacity_margin = 1280; /* ~20% */

123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
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;
}

141 142 143 144 145 146 147 148 149
/*
 * 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:
 */
150
static unsigned int get_update_sysctl_factor(void)
151
{
152
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187
	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();
}

188
#define WMULT_CONST	(~0U)
189 190
#define WMULT_SHIFT	32

191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206
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;
}
207 208

/*
209 210 211 212
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
213
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
214 215 216 217 218
 * 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.
219
 */
220
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
221
{
222 223
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
224

225
	__update_inv_weight(lw);
226

227 228 229 230 231
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
232 233
	}

234 235
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
236

237 238 239 240
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
241

242
	return mul_u64_u32_shr(delta_exec, fact, shift);
243 244 245 246
}


const struct sched_class fair_sched_class;
247

248 249 250 251
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

252
#ifdef CONFIG_FAIR_GROUP_SCHED
253

254
/* cpu runqueue to which this cfs_rq is attached */
255 256
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
257
	return cfs_rq->rq;
258 259
}

260 261
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
262

263 264
static inline struct task_struct *task_of(struct sched_entity *se)
{
265
	SCHED_WARN_ON(!entity_is_task(se));
266 267 268
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289
/* 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;
}

290 291 292
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
293 294 295 296 297 298 299 300 301 302 303 304
		/*
		 * Ensure we either appear before our parent (if already
		 * enqueued) or force our parent to appear after us when it is
		 * enqueued.  The fact that we always enqueue bottom-up
		 * reduces this to two cases.
		 */
		if (cfs_rq->tg->parent &&
		    cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
		} else {
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
305
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
306
		}
307 308 309 310 311 312 313 314 315 316 317 318 319

		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
320 321 322 323 324
/* Iterate thr' all leaf cfs_rq's on a runqueue */
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
	list_for_each_entry_rcu(cfs_rq, &rq->leaf_cfs_rq_list, leaf_cfs_rq_list)

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
325
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
326 327 328
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
329
		return se->cfs_rq;
P
Peter Zijlstra 已提交
330

P
Peter Zijlstra 已提交
331
	return NULL;
P
Peter Zijlstra 已提交
332 333 334 335 336 337 338
}

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

339 340 341 342 343 344 345 346 347 348 349 350 351
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
352 353
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370

	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);
	}
}

371 372 373 374 375 376
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
377

378 379 380
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
381 382 383 384
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
385 386
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
387

P
Peter Zijlstra 已提交
388
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
389
{
P
Peter Zijlstra 已提交
390
	return &task_rq(p)->cfs;
391 392
}

P
Peter Zijlstra 已提交
393 394 395 396 397 398 399 400 401 402 403 404 405 406
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;
}

407 408 409 410 411 412 413 414
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
}

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

P
Peter Zijlstra 已提交
415 416 417 418 419 420 421 422
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
		for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)

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

423 424 425 426 427
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
428 429
#endif	/* CONFIG_FAIR_GROUP_SCHED */

430
static __always_inline
431
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
432 433 434 435 436

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

437
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
438
{
439
	s64 delta = (s64)(vruntime - max_vruntime);
440
	if (delta > 0)
441
		max_vruntime = vruntime;
442

443
	return max_vruntime;
444 445
}

446
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
447 448 449 450 451 452 453 454
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

455 456 457 458 459 460
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

461 462
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
463 464
	struct sched_entity *curr = cfs_rq->curr;

465 466
	u64 vruntime = cfs_rq->min_vruntime;

467 468 469 470 471 472
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
473 474 475 476 477 478

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

479
		if (!curr)
480 481 482 483 484
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

485
	/* ensure we never gain time by being placed backwards. */
486
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
487 488 489 490
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
491 492
}

493 494 495
/*
 * Enqueue an entity into the rb-tree:
 */
496
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
{
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_node;
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
	int leftmost = 1;

	/*
	 * Find the right place in the rbtree:
	 */
	while (*link) {
		parent = *link;
		entry = rb_entry(parent, struct sched_entity, run_node);
		/*
		 * We dont care about collisions. Nodes with
		 * the same key stay together.
		 */
513
		if (entity_before(se, entry)) {
514 515 516 517 518 519 520 521 522 523 524
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
525
	if (leftmost)
I
Ingo Molnar 已提交
526
		cfs_rq->rb_leftmost = &se->run_node;
527 528 529 530 531

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

532
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
533
{
P
Peter Zijlstra 已提交
534 535 536 537 538 539
	if (cfs_rq->rb_leftmost == &se->run_node) {
		struct rb_node *next_node;

		next_node = rb_next(&se->run_node);
		cfs_rq->rb_leftmost = next_node;
	}
I
Ingo Molnar 已提交
540

541 542 543
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

544
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
545
{
546 547 548 549 550 551
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
552 553
}

554 555 556 557 558 559 560 561 562 563 564
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
565
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
566
{
567
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
568

569 570
	if (!last)
		return NULL;
571 572

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

575 576 577 578
/**************************************************************
 * Scheduling class statistics methods:
 */

579
int sched_proc_update_handler(struct ctl_table *table, int write,
580
		void __user *buffer, size_t *lenp,
581 582
		loff_t *ppos)
{
583
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
584
	unsigned int factor = get_update_sysctl_factor();
585 586 587 588 589 590 591

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

592 593 594 595 596 597 598
#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

599 600 601
	return 0;
}
#endif
602

603
/*
604
 * delta /= w
605
 */
606
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
607
{
608
	if (unlikely(se->load.weight != NICE_0_LOAD))
609
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
610 611 612 613

	return delta;
}

614 615 616
/*
 * The idea is to set a period in which each task runs once.
 *
617
 * When there are too many tasks (sched_nr_latency) we have to stretch
618 619 620 621
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
622 623
static u64 __sched_period(unsigned long nr_running)
{
624 625 626 627
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
628 629
}

630 631 632 633
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
634
 * s = p*P[w/rw]
635
 */
P
Peter Zijlstra 已提交
636
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
637
{
M
Mike Galbraith 已提交
638
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
639

M
Mike Galbraith 已提交
640
	for_each_sched_entity(se) {
L
Lin Ming 已提交
641
		struct load_weight *load;
642
		struct load_weight lw;
L
Lin Ming 已提交
643 644 645

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
646

M
Mike Galbraith 已提交
647
		if (unlikely(!se->on_rq)) {
648
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
649 650 651 652

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
653
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
654 655
	}
	return slice;
656 657
}

658
/*
A
Andrei Epure 已提交
659
 * We calculate the vruntime slice of a to-be-inserted task.
660
 *
661
 * vs = s/w
662
 */
663
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
664
{
665
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
666 667
}

668
#ifdef CONFIG_SMP
669
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
670 671
static unsigned long task_h_load(struct task_struct *p);

672 673
/*
 * We choose a half-life close to 1 scheduling period.
674 675
 * Note: The tables runnable_avg_yN_inv and runnable_avg_yN_sum are
 * dependent on this value.
676 677 678
 */
#define LOAD_AVG_PERIOD 32
#define LOAD_AVG_MAX 47742 /* maximum possible load avg */
679
#define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_AVG_MAX */
680

681 682
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
683
{
684
	struct sched_avg *sa = &se->avg;
685

686 687 688 689 690 691 692
	sa->last_update_time = 0;
	/*
	 * sched_avg's period_contrib should be strictly less then 1024, so
	 * we give it 1023 to make sure it is almost a period (1024us), and
	 * will definitely be update (after enqueue).
	 */
	sa->period_contrib = 1023;
693 694 695 696 697 698 699 700
	/*
	 * Tasks are intialized with full load to be seen as heavy tasks until
	 * they get a chance to stabilize to their real load level.
	 * Group entities are intialized with zero load to reflect the fact that
	 * nothing has been attached to the task group yet.
	 */
	if (entity_is_task(se))
		sa->load_avg = scale_load_down(se->load.weight);
701
	sa->load_sum = sa->load_avg * LOAD_AVG_MAX;
702 703 704 705 706
	/*
	 * At this point, util_avg won't be used in select_task_rq_fair anyway
	 */
	sa->util_avg = 0;
	sa->util_sum = 0;
707
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
708
}
709

710 711
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
static int update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq);
712
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force);
713 714
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se);

715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743
/*
 * With new tasks being created, their initial util_avgs are extrapolated
 * based on the cfs_rq's current util_avg:
 *
 *   util_avg = cfs_rq->util_avg / (cfs_rq->load_avg + 1) * se.load.weight
 *
 * However, in many cases, the above util_avg does not give a desired
 * value. Moreover, the sum of the util_avgs may be divergent, such
 * as when the series is a harmonic series.
 *
 * To solve this problem, we also cap the util_avg of successive tasks to
 * only 1/2 of the left utilization budget:
 *
 *   util_avg_cap = (1024 - cfs_rq->avg.util_avg) / 2^n
 *
 * where n denotes the nth task.
 *
 * For example, a simplest series from the beginning would be like:
 *
 *  task  util_avg: 512, 256, 128,  64,  32,   16,    8, ...
 * cfs_rq util_avg: 512, 768, 896, 960, 992, 1008, 1016, ...
 *
 * Finally, that extrapolated util_avg is clamped to the cap (util_avg_cap)
 * if util_avg > util_avg_cap.
 */
void post_init_entity_util_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct sched_avg *sa = &se->avg;
744
	long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
745
	u64 now = cfs_rq_clock_task(cfs_rq);
746 747 748 749 750 751 752 753 754 755 756 757 758

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

			if (sa->util_avg > cap)
				sa->util_avg = cap;
		} else {
			sa->util_avg = cap;
		}
		sa->util_sum = sa->util_avg * LOAD_AVG_MAX;
	}
759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777

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

778
	update_cfs_rq_load_avg(now, cfs_rq, false);
779
	attach_entity_load_avg(cfs_rq, se);
780
	update_tg_load_avg(cfs_rq, false);
781 782
}

783
#else /* !CONFIG_SMP */
784
void init_entity_runnable_average(struct sched_entity *se)
785 786
{
}
787 788 789
void post_init_entity_util_avg(struct sched_entity *se)
{
}
790 791 792
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
{
}
793
#endif /* CONFIG_SMP */
794

795
/*
796
 * Update the current task's runtime statistics.
797
 */
798
static void update_curr(struct cfs_rq *cfs_rq)
799
{
800
	struct sched_entity *curr = cfs_rq->curr;
801
	u64 now = rq_clock_task(rq_of(cfs_rq));
802
	u64 delta_exec;
803 804 805 806

	if (unlikely(!curr))
		return;

807 808
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
809
		return;
810

I
Ingo Molnar 已提交
811
	curr->exec_start = now;
812

813 814 815 816
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
817
	schedstat_add(cfs_rq->exec_clock, delta_exec);
818 819 820 821

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

822 823 824
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

825
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
826
		cpuacct_charge(curtask, delta_exec);
827
		account_group_exec_runtime(curtask, delta_exec);
828
	}
829 830

	account_cfs_rq_runtime(cfs_rq, delta_exec);
831 832
}

833 834 835 836 837
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

838
static inline void
839
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
840
{
841 842 843 844 845 846 847
	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);
848 849

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
850 851
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
852

853
	schedstat_set(se->statistics.wait_start, wait_start);
854 855
}

856
static inline void
857 858 859
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
860 861
	u64 delta;

862 863 864 865
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
866 867 868 869 870 871 872 873 874

	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.
			 */
875
			schedstat_set(se->statistics.wait_start, delta);
876 877 878 879 880
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

881 882 883 884 885
	schedstat_set(se->statistics.wait_max,
		      max(schedstat_val(se->statistics.wait_max), delta));
	schedstat_inc(se->statistics.wait_count);
	schedstat_add(se->statistics.wait_sum, delta);
	schedstat_set(se->statistics.wait_start, 0);
886 887
}

888
static inline void
889 890 891
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
892 893 894 895 896 897 898
	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);
899 900 901 902

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

903 904
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
905 906 907 908

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

909 910
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
			schedstat_set(se->statistics.sleep_max, delta);
911

912 913
		schedstat_set(se->statistics.sleep_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
914 915 916 917 918 919

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
920 921
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
922 923 924 925

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

926 927
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
			schedstat_set(se->statistics.block_max, delta);
928

929 930
		schedstat_set(se->statistics.block_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
931 932 933

		if (tsk) {
			if (tsk->in_iowait) {
934 935
				schedstat_add(se->statistics.iowait_sum, delta);
				schedstat_inc(se->statistics.iowait_count);
936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
				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);
		}
	}
954 955
}

956 957 958
/*
 * Task is being enqueued - update stats:
 */
959
static inline void
960
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
961
{
962 963 964
	if (!schedstat_enabled())
		return;

965 966 967 968
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
969
	if (se != cfs_rq->curr)
970
		update_stats_wait_start(cfs_rq, se);
971 972 973

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
974 975 976
}

static inline void
977
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
978
{
979 980 981 982

	if (!schedstat_enabled())
		return;

983 984 985 986
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
987
	if (se != cfs_rq->curr)
988
		update_stats_wait_end(cfs_rq, se);
989

990 991
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
992

993 994 995 996 997 998
		if (tsk->state & TASK_INTERRUPTIBLE)
			schedstat_set(se->statistics.sleep_start,
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
			schedstat_set(se->statistics.block_start,
				      rq_clock(rq_of(cfs_rq)));
999 1000 1001
	}
}

1002 1003 1004 1005
/*
 * We are picking a new current task - update its stats:
 */
static inline void
1006
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1007 1008 1009 1010
{
	/*
	 * We are starting a new run period:
	 */
1011
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1012 1013 1014 1015 1016 1017
}

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

1018 1019
#ifdef CONFIG_NUMA_BALANCING
/*
1020 1021 1022
 * 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.
1023
 */
1024 1025
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1026 1027 1028

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1029

1030 1031 1032
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
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)
{
1057
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1058 1059 1060
	unsigned int scan, floor;
	unsigned int windows = 1;

1061 1062
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
	floor = 1000 / windows;

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

static unsigned int task_scan_max(struct task_struct *p)
{
	unsigned int smin = task_scan_min(p);
	unsigned int smax;

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
	return max(smin, smax);
}

1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090
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));
}

1091 1092 1093 1094 1095
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
1096
	pid_t gid;
1097
	int active_nodes;
1098 1099

	struct rcu_head rcu;
1100
	unsigned long total_faults;
1101
	unsigned long max_faults_cpu;
1102 1103 1104 1105 1106
	/*
	 * 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.
	 */
1107
	unsigned long *faults_cpu;
1108
	unsigned long faults[0];
1109 1110
};

1111 1112 1113 1114 1115 1116 1117 1118 1119
/* 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)

1120 1121 1122 1123 1124
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

1125 1126 1127 1128 1129 1130 1131
/*
 * The averaged statistics, shared & private, memory & cpu,
 * occupy the first half of the array. The second half of the
 * array is for current counters, which are averaged into the
 * first set by task_numa_placement.
 */
static inline int task_faults_idx(enum numa_faults_stats s, int nid, int priv)
1132
{
1133
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1134 1135 1136 1137
}

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

1141 1142
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1143 1144
}

1145 1146 1147 1148 1149
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

1150 1151
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1152 1153
}

1154 1155
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1156 1157
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
1158 1159
}

1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
/*
 * 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;
}

1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

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

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

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

		/*
		 * On systems with a backplane NUMA topology, compare groups
		 * of nodes, and move tasks towards the group with the most
		 * memory accesses. When comparing two nodes at distance
		 * "hoplimit", only nodes closer by than "hoplimit" are part
		 * of each group. Skip other nodes.
		 */
		if (sched_numa_topology_type == NUMA_BACKPLANE &&
					dist > maxdist)
			continue;

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

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

		score += faults;
	}

	return score;
}

1237 1238 1239 1240 1241 1242
/*
 * 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.
 */
1243 1244
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1245
{
1246
	unsigned long faults, total_faults;
1247

1248
	if (!p->numa_faults)
1249 1250 1251 1252 1253 1254 1255
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1256
	faults = task_faults(p, nid);
1257 1258
	faults += score_nearby_nodes(p, nid, dist, true);

1259
	return 1000 * faults / total_faults;
1260 1261
}

1262 1263
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1264
{
1265 1266 1267 1268 1269 1270 1271 1272
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1273 1274
		return 0;

1275
	faults = group_faults(p, nid);
1276 1277
	faults += score_nearby_nodes(p, nid, dist, false);

1278
	return 1000 * faults / total_faults;
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 1315 1316 1317 1318 1319 1320
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
	struct numa_group *ng = p->numa_group;
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

	this_cpupid = cpu_pid_to_cpupid(dst_cpu, current->pid);

	/*
	 * Multi-stage node selection is used in conjunction with a periodic
	 * migration fault to build a temporal task<->page relation. By using
	 * a two-stage filter we remove short/unlikely relations.
	 *
	 * Using P(p) ~ n_p / n_t as per frequentist probability, we can equate
	 * a task's usage of a particular page (n_p) per total usage of this
	 * page (n_t) (in a given time-span) to a probability.
	 *
	 * Our periodic faults will sample this probability and getting the
	 * same result twice in a row, given these samples are fully
	 * independent, is then given by P(n)^2, provided our sample period
	 * is sufficiently short compared to the usage pattern.
	 *
	 * This quadric squishes small probabilities, making it less likely we
	 * act on an unlikely task<->page relation.
	 */
	last_cpupid = page_cpupid_xchg_last(page, this_cpupid);
	if (!cpupid_pid_unset(last_cpupid) &&
				cpupid_to_nid(last_cpupid) != dst_nid)
		return false;

	/* Always allow migrate on private faults */
	if (cpupid_match_pid(p, last_cpupid))
		return true;

	/* A shared fault, but p->numa_group has not been set up yet. */
	if (!ng)
		return true;

	/*
1321 1322
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1323
	 */
1324 1325
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1326 1327 1328
		return true;

	/*
1329 1330 1331 1332 1333 1334
	 * 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)
1335
	 */
1336 1337
	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;
1338 1339
}

1340
static unsigned long weighted_cpuload(const int cpu);
1341 1342
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1343
static unsigned long capacity_of(int cpu);
1344 1345
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

1346
/* Cached statistics for all CPUs within a node */
1347
struct numa_stats {
1348
	unsigned long nr_running;
1349
	unsigned long load;
1350 1351

	/* Total compute capacity of CPUs on a node */
1352
	unsigned long compute_capacity;
1353 1354

	/* Approximate capacity in terms of runnable tasks on a node */
1355
	unsigned long task_capacity;
1356
	int has_free_capacity;
1357
};
1358

1359 1360 1361 1362 1363
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1364 1365
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1366 1367 1368 1369 1370 1371 1372

	memset(ns, 0, sizeof(*ns));
	for_each_cpu(cpu, cpumask_of_node(nid)) {
		struct rq *rq = cpu_rq(cpu);

		ns->nr_running += rq->nr_running;
		ns->load += weighted_cpuload(cpu);
1373
		ns->compute_capacity += capacity_of(cpu);
1374 1375

		cpus++;
1376 1377
	}

1378 1379 1380 1381 1382
	/*
	 * 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.
	 *
1383 1384
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1385 1386 1387 1388
	 */
	if (!cpus)
		return;

1389 1390 1391 1392 1393 1394
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

	ns->task_capacity = min_t(unsigned, capacity,
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1395
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1396 1397
}

1398 1399
struct task_numa_env {
	struct task_struct *p;
1400

1401 1402
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1403

1404
	struct numa_stats src_stats, dst_stats;
1405

1406
	int imbalance_pct;
1407
	int dist;
1408 1409 1410

	struct task_struct *best_task;
	long best_imp;
1411 1412 1413
	int best_cpu;
};

1414 1415 1416 1417 1418
static void task_numa_assign(struct task_numa_env *env,
			     struct task_struct *p, long imp)
{
	if (env->best_task)
		put_task_struct(env->best_task);
1419 1420
	if (p)
		get_task_struct(p);
1421 1422 1423 1424 1425 1426

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

1427
static bool load_too_imbalanced(long src_load, long dst_load,
1428 1429
				struct task_numa_env *env)
{
1430 1431
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442
	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;
1443 1444

	/* We care about the slope of the imbalance, not the direction. */
1445 1446
	if (dst_load < src_load)
		swap(dst_load, src_load);
1447 1448

	/* Is the difference below the threshold? */
1449 1450
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1451 1452 1453 1454 1455
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1456
	 * Compare it with the old imbalance.
1457
	 */
1458
	orig_src_load = env->src_stats.load;
1459
	orig_dst_load = env->dst_stats.load;
1460

1461 1462
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1463

1464 1465 1466 1467 1468
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;

	/* Would this change make things worse? */
	return (imb > old_imb);
1469 1470
}

1471 1472 1473 1474 1475 1476
/*
 * 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
 */
1477 1478
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1479 1480 1481 1482
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1483
	long src_load, dst_load;
1484
	long load;
1485
	long imp = env->p->numa_group ? groupimp : taskimp;
1486
	long moveimp = imp;
1487
	int dist = env->dist;
1488 1489

	rcu_read_lock();
1490 1491
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1492 1493
		cur = NULL;

1494 1495 1496 1497 1498 1499 1500
	/*
	 * 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;

1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512
	/*
	 * "imp" is the fault differential for the source task between the
	 * source and destination node. Calculate the total differential for
	 * the source task and potential destination task. The more negative
	 * the value is, the more rmeote accesses that would be expected to
	 * be incurred if the tasks were swapped.
	 */
	if (cur) {
		/* Skip this swap candidate if cannot move to the source cpu */
		if (!cpumask_test_cpu(env->src_cpu, tsk_cpus_allowed(cur)))
			goto unlock;

1513 1514
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1515
		 * in any group then look only at task weights.
1516
		 */
1517
		if (cur->numa_group == env->p->numa_group) {
1518 1519
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1520 1521 1522 1523 1524 1525
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1526
		} else {
1527 1528 1529 1530 1531 1532
			/*
			 * 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)
1533 1534
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1535
			else
1536 1537
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1538
		}
1539 1540
	}

1541
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1542 1543 1544 1545
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1546
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1547
		    !env->dst_stats.has_free_capacity)
1548 1549 1550 1551 1552 1553
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1554 1555
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1556 1557 1558 1559 1560 1561
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1562 1563 1564
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1565

1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582
	if (moveimp > imp && moveimp > env->best_imp) {
		/*
		 * If the improvement from just moving env->p direction is
		 * better than swapping tasks around, check if a move is
		 * possible. Store a slightly smaller score than moveimp,
		 * so an actually idle CPU will win.
		 */
		if (!load_too_imbalanced(src_load, dst_load, env)) {
			imp = moveimp - 1;
			cur = NULL;
			goto assign;
		}
	}

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

1583
	if (cur) {
1584 1585 1586
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1587 1588
	}

1589
	if (load_too_imbalanced(src_load, dst_load, env))
1590 1591
		goto unlock;

1592 1593 1594 1595
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1596 1597 1598 1599 1600 1601
	if (!cur) {
		/*
		 * select_idle_siblings() uses an per-cpu cpumask that
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1602 1603
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1604 1605
		local_irq_enable();
	}
1606

1607 1608 1609 1610 1611 1612
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1613 1614
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1615 1616 1617 1618 1619 1620 1621 1622 1623
{
	int cpu;

	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
		if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(env->p)))
			continue;

		env->dst_cpu = cpu;
1624
		task_numa_compare(env, taskimp, groupimp);
1625 1626 1627
	}
}

1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644
/* Only move tasks to a NUMA node less busy than the current node. */
static bool numa_has_capacity(struct task_numa_env *env)
{
	struct numa_stats *src = &env->src_stats;
	struct numa_stats *dst = &env->dst_stats;

	if (src->has_free_capacity && !dst->has_free_capacity)
		return false;

	/*
	 * Only consider a task move if the source has a higher load
	 * than the destination, corrected for CPU capacity on each node.
	 *
	 *      src->load                dst->load
	 * --------------------- vs ---------------------
	 * src->compute_capacity    dst->compute_capacity
	 */
1645 1646 1647
	if (src->load * dst->compute_capacity * env->imbalance_pct >

	    dst->load * src->compute_capacity * 100)
1648 1649 1650 1651 1652
		return true;

	return false;
}

1653 1654 1655 1656
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1657

1658
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1659
		.src_nid = task_node(p),
1660 1661 1662 1663 1664

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1665
		.best_cpu = -1,
1666 1667
	};
	struct sched_domain *sd;
1668
	unsigned long taskweight, groupweight;
1669
	int nid, ret, dist;
1670
	long taskimp, groupimp;
1671

1672
	/*
1673 1674 1675 1676 1677 1678
	 * 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.
1679 1680
	 */
	rcu_read_lock();
1681
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1682 1683
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1684 1685
	rcu_read_unlock();

1686 1687 1688 1689 1690 1691 1692
	/*
	 * 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)) {
1693
		p->numa_preferred_nid = task_node(p);
1694 1695 1696
		return -EINVAL;
	}

1697
	env.dst_nid = p->numa_preferred_nid;
1698 1699 1700 1701 1702 1703
	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;
1704
	update_numa_stats(&env.dst_stats, env.dst_nid);
1705

1706
	/* Try to find a spot on the preferred nid. */
1707 1708
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1709

1710 1711 1712 1713 1714 1715 1716
	/*
	 * 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.
	 */
1717
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1718 1719 1720
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1721

1722
			dist = node_distance(env.src_nid, env.dst_nid);
1723 1724 1725 1726 1727
			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);
			}
1728

1729
			/* Only consider nodes where both task and groups benefit */
1730 1731
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1732
			if (taskimp < 0 && groupimp < 0)
1733 1734
				continue;

1735
			env.dist = dist;
1736 1737
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1738 1739
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1740 1741 1742
		}
	}

1743 1744 1745 1746 1747 1748 1749 1750
	/*
	 * 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.
	 */
1751
	if (p->numa_group) {
1752 1753
		struct numa_group *ng = p->numa_group;

1754 1755 1756 1757 1758
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

1759
		if (ng->active_nodes > 1 && numa_is_active_node(env.dst_nid, ng))
1760 1761 1762 1763 1764 1765
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1766

1767 1768 1769 1770 1771 1772
	/*
	 * Reset the scan period if the task is being rescheduled on an
	 * alternative node to recheck if the tasks is now properly placed.
	 */
	p->numa_scan_period = task_scan_min(p);

1773
	if (env.best_task == NULL) {
1774 1775 1776
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1777 1778 1779 1780
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1781 1782
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1783 1784
	put_task_struct(env.best_task);
	return ret;
1785 1786
}

1787 1788 1789
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1790 1791
	unsigned long interval = HZ;

1792
	/* This task has no NUMA fault statistics yet */
1793
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1794 1795
		return;

1796
	/* Periodically retry migrating the task to the preferred node */
1797 1798
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1799 1800

	/* Success if task is already running on preferred CPU */
1801
	if (task_node(p) == p->numa_preferred_nid)
1802 1803 1804
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1805
	task_numa_migrate(p);
1806 1807
}

1808
/*
1809
 * Find out how many nodes on the workload is actively running on. Do this by
1810 1811 1812 1813
 * 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.
 */
1814
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1815 1816
{
	unsigned long faults, max_faults = 0;
1817
	int nid, active_nodes = 0;
1818 1819 1820 1821 1822 1823 1824 1825 1826

	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);
1827 1828
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1829
	}
1830 1831 1832

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1833 1834
}

1835 1836 1837
/*
 * 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
1838 1839 1840
 * 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.
1841 1842
 */
#define NUMA_PERIOD_SLOTS 10
1843
#define NUMA_PERIOD_THRESHOLD 7
1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863

/*
 * Increase the scan period (slow down scanning) if the majority of
 * our memory is already on our local node, or if the majority of
 * the page accesses are shared with other processes.
 * Otherwise, decrease the scan period.
 */
static void update_task_scan_period(struct task_struct *p,
			unsigned long shared, unsigned long private)
{
	unsigned int period_slot;
	int ratio;
	int diff;

	unsigned long remote = p->numa_faults_locality[0];
	unsigned long local = p->numa_faults_locality[1];

	/*
	 * If there were no record hinting faults then either the task is
	 * completely idle or all activity is areas that are not of interest
1864 1865 1866
	 * 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
1867
	 */
1868
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901
		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period << 1);

		p->mm->numa_next_scan = jiffies +
			msecs_to_jiffies(p->numa_scan_period);

		return;
	}

	/*
	 * Prepare to scale scan period relative to the current period.
	 *	 == NUMA_PERIOD_THRESHOLD scan period stays the same
	 *       <  NUMA_PERIOD_THRESHOLD scan period decreases (scan faster)
	 *	 >= NUMA_PERIOD_THRESHOLD scan period increases (scan slower)
	 */
	period_slot = DIV_ROUND_UP(p->numa_scan_period, NUMA_PERIOD_SLOTS);
	ratio = (local * NUMA_PERIOD_SLOTS) / (local + remote);
	if (ratio >= NUMA_PERIOD_THRESHOLD) {
		int slot = ratio - NUMA_PERIOD_THRESHOLD;
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;

		/*
		 * Scale scan rate increases based on sharing. There is an
		 * inverse relationship between the degree of sharing and
		 * the adjustment made to the scanning period. Broadly
		 * speaking the intent is that there is little point
		 * scanning faster if shared accesses dominate as it may
		 * simply bounce migrations uselessly
		 */
1902
		ratio = DIV_ROUND_UP(private * NUMA_PERIOD_SLOTS, (private + shared + 1));
1903 1904 1905 1906 1907 1908 1909 1910
		diff = (diff * ratio) / NUMA_PERIOD_SLOTS;
	}

	p->numa_scan_period = clamp(p->numa_scan_period + diff,
			task_scan_min(p), task_scan_max(p));
	memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
}

1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928
/*
 * 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 {
1929 1930
		delta = p->se.avg.load_sum / p->se.load.weight;
		*period = LOAD_AVG_MAX;
1931 1932 1933 1934 1935 1936 1937 1938
	}

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

	return delta;
}

1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985
/*
 * 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;
1986
		nodemask_t max_group = NODE_MASK_NONE;
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019
		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. */
2020 2021
		if (!max_faults)
			break;
2022 2023 2024 2025 2026
		nodes = max_group;
	}
	return nid;
}

2027 2028
static void task_numa_placement(struct task_struct *p)
{
2029 2030
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
2031
	unsigned long fault_types[2] = { 0, 0 };
2032 2033
	unsigned long total_faults;
	u64 runtime, period;
2034
	spinlock_t *group_lock = NULL;
2035

2036 2037 2038 2039 2040
	/*
	 * 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:
	 */
2041
	seq = READ_ONCE(p->mm->numa_scan_seq);
2042 2043 2044
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2045
	p->numa_scan_period_max = task_scan_max(p);
2046

2047 2048 2049 2050
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2051 2052 2053
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2054
		spin_lock_irq(group_lock);
2055 2056
	}

2057 2058
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2059 2060
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2061
		unsigned long faults = 0, group_faults = 0;
2062
		int priv;
2063

2064
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2065
			long diff, f_diff, f_weight;
2066

2067 2068 2069 2070
			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);
2071

2072
			/* Decay existing window, copy faults since last scan */
2073 2074 2075
			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;
2076

2077 2078 2079 2080 2081 2082 2083 2084
			/*
			 * 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);
2085
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2086
				   (total_faults + 1);
2087 2088
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2089

2090 2091 2092
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2093
			p->total_numa_faults += diff;
2094
			if (p->numa_group) {
2095 2096 2097 2098 2099 2100 2101 2102 2103
				/*
				 * 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;
2104
				p->numa_group->total_faults += diff;
2105
				group_faults += p->numa_group->faults[mem_idx];
2106
			}
2107 2108
		}

2109 2110 2111 2112
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
2113 2114 2115 2116 2117 2118 2119

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

2120 2121
	update_task_scan_period(p, fault_types[0], fault_types[1]);

2122
	if (p->numa_group) {
2123
		numa_group_count_active_nodes(p->numa_group);
2124
		spin_unlock_irq(group_lock);
2125
		max_nid = preferred_group_nid(p, max_group_nid);
2126 2127
	}

2128 2129 2130 2131 2132 2133 2134
	if (max_faults) {
		/* Set the new preferred node */
		if (max_nid != p->numa_preferred_nid)
			sched_setnuma(p, max_nid);

		if (task_node(p) != p->numa_preferred_nid)
			numa_migrate_preferred(p);
2135
	}
2136 2137
}

2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148
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);
}

2149 2150
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2151 2152 2153 2154 2155 2156 2157 2158 2159
{
	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) +
2160
				    4*nr_node_ids*sizeof(unsigned long);
2161 2162 2163 2164 2165 2166

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

		atomic_set(&grp->refcount, 1);
2167 2168
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2169
		spin_lock_init(&grp->lock);
2170
		grp->gid = p->pid;
2171
		/* Second half of the array tracks nids where faults happen */
2172 2173
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2174

2175
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2176
			grp->faults[i] = p->numa_faults[i];
2177

2178
		grp->total_faults = p->total_numa_faults;
2179

2180 2181 2182 2183 2184
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2185
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2186 2187

	if (!cpupid_match_pid(tsk, cpupid))
2188
		goto no_join;
2189 2190 2191

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2192
		goto no_join;
2193 2194 2195

	my_grp = p->numa_group;
	if (grp == my_grp)
2196
		goto no_join;
2197 2198 2199 2200 2201 2202

	/*
	 * 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)
2203
		goto no_join;
2204 2205 2206 2207 2208

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2209
		goto no_join;
2210

2211 2212 2213 2214 2215 2216 2217
	/* 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;
2218

2219 2220 2221
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2222
	if (join && !get_numa_group(grp))
2223
		goto no_join;
2224 2225 2226 2227 2228 2229

	rcu_read_unlock();

	if (!join)
		return;

2230 2231
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2232

2233
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2234 2235
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2236
	}
2237 2238
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2239 2240 2241 2242 2243

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

	spin_unlock(&my_grp->lock);
2244
	spin_unlock_irq(&grp->lock);
2245 2246 2247 2248

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2249 2250 2251 2252 2253
	return;

no_join:
	rcu_read_unlock();
	return;
2254 2255 2256 2257 2258
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2259
	void *numa_faults = p->numa_faults;
2260 2261
	unsigned long flags;
	int i;
2262 2263

	if (grp) {
2264
		spin_lock_irqsave(&grp->lock, flags);
2265
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2266
			grp->faults[i] -= p->numa_faults[i];
2267
		grp->total_faults -= p->total_numa_faults;
2268

2269
		grp->nr_tasks--;
2270
		spin_unlock_irqrestore(&grp->lock, flags);
2271
		RCU_INIT_POINTER(p->numa_group, NULL);
2272 2273 2274
		put_numa_group(grp);
	}

2275
	p->numa_faults = NULL;
2276
	kfree(numa_faults);
2277 2278
}

2279 2280 2281
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2282
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2283 2284
{
	struct task_struct *p = current;
2285
	bool migrated = flags & TNF_MIGRATED;
2286
	int cpu_node = task_node(current);
2287
	int local = !!(flags & TNF_FAULT_LOCAL);
2288
	struct numa_group *ng;
2289
	int priv;
2290

2291
	if (!static_branch_likely(&sched_numa_balancing))
2292 2293
		return;

2294 2295 2296 2297
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2298
	/* Allocate buffer to track faults on a per-node basis */
2299 2300
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2301
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2302

2303 2304
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2305
			return;
2306

2307
		p->total_numa_faults = 0;
2308
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2309
	}
2310

2311 2312 2313 2314 2315 2316 2317 2318
	/*
	 * 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);
2319
		if (!priv && !(flags & TNF_NO_GROUP))
2320
			task_numa_group(p, last_cpupid, flags, &priv);
2321 2322
	}

2323 2324 2325 2326 2327 2328
	/*
	 * 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.
	 */
2329 2330 2331 2332
	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))
2333 2334
		local = 1;

2335
	task_numa_placement(p);
2336

2337 2338 2339 2340 2341
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
	if (time_after(jiffies, p->numa_migrate_retry))
2342 2343
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2344 2345
	if (migrated)
		p->numa_pages_migrated += pages;
2346 2347
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2348

2349 2350
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2351
	p->numa_faults_locality[local] += pages;
2352 2353
}

2354 2355
static void reset_ptenuma_scan(struct task_struct *p)
{
2356 2357 2358 2359 2360 2361 2362 2363
	/*
	 * 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:
	 */
2364
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2365 2366 2367
	p->mm->numa_scan_offset = 0;
}

2368 2369 2370 2371 2372 2373 2374 2375 2376
/*
 * 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;
2377
	u64 runtime = p->se.sum_exec_runtime;
2378
	struct vm_area_struct *vma;
2379
	unsigned long start, end;
2380
	unsigned long nr_pte_updates = 0;
2381
	long pages, virtpages;
2382

2383
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396

	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;

2397
	if (!mm->numa_next_scan) {
2398 2399
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2400 2401
	}

2402 2403 2404 2405 2406 2407 2408
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2409 2410 2411 2412
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
2413

2414
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2415 2416 2417
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2418 2419 2420 2421 2422 2423
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2424 2425 2426
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2427
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2428 2429
	if (!pages)
		return;
2430

2431

2432
	down_read(&mm->mmap_sem);
2433
	vma = find_vma(mm, start);
2434 2435
	if (!vma) {
		reset_ptenuma_scan(p);
2436
		start = 0;
2437 2438
		vma = mm->mmap;
	}
2439
	for (; vma; vma = vma->vm_next) {
2440
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2441
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2442
			continue;
2443
		}
2444

2445 2446 2447 2448 2449 2450 2451 2452 2453 2454
		/*
		 * 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
2455 2456 2457 2458 2459 2460
		/*
		 * 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;
2461

2462 2463 2464 2465
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2466
			nr_pte_updates = change_prot_numa(vma, start, end);
2467 2468

			/*
2469 2470 2471 2472 2473 2474
			 * 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.
2475 2476 2477
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2478
			virtpages -= (end - start) >> PAGE_SHIFT;
2479

2480
			start = end;
2481
			if (pages <= 0 || virtpages <= 0)
2482
				goto out;
2483 2484

			cond_resched();
2485
		} while (end != vma->vm_end);
2486
	}
2487

2488
out:
2489
	/*
P
Peter Zijlstra 已提交
2490 2491 2492 2493
	 * 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.
2494 2495
	 */
	if (vma)
2496
		mm->numa_scan_offset = start;
2497 2498 2499
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510

	/*
	 * 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;
	}
2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535
}

/*
 * 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;

2536
	if (now > curr->node_stamp + period) {
2537
		if (!curr->node_stamp)
2538
			curr->numa_scan_period = task_scan_min(curr);
2539
		curr->node_stamp += period;
2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550

		if (!time_before(jiffies, curr->mm->numa_next_scan)) {
			init_task_work(work, task_numa_work); /* TODO: move this into sched_fork() */
			task_work_add(curr, work, true);
		}
	}
}
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2551 2552 2553 2554 2555 2556 2557 2558

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)
{
}
2559 2560
#endif /* CONFIG_NUMA_BALANCING */

2561 2562 2563 2564
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2565
	if (!parent_entity(se))
2566
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2567
#ifdef CONFIG_SMP
2568 2569 2570 2571 2572 2573
	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);
	}
2574
#endif
2575 2576 2577 2578 2579 2580 2581
	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);
2582
	if (!parent_entity(se))
2583
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2584
#ifdef CONFIG_SMP
2585 2586
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2587
		list_del_init(&se->group_node);
2588
	}
2589
#endif
2590 2591 2592
	cfs_rq->nr_running--;
}

2593 2594
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2595
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2596
{
2597
	long tg_weight, load, shares;
2598 2599

	/*
2600 2601 2602
	 * This really should be: cfs_rq->avg.load_avg, but instead we use
	 * cfs_rq->load.weight, which is its upper bound. This helps ramp up
	 * the shares for small weight interactive tasks.
2603
	 */
2604
	load = scale_load_down(cfs_rq->load.weight);
2605

2606
	tg_weight = atomic_long_read(&tg->load_avg);
2607

2608 2609 2610
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2611 2612

	shares = (tg->shares * load);
2613 2614
	if (tg_weight)
		shares /= tg_weight;
2615 2616 2617 2618 2619 2620 2621 2622 2623

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

	return shares;
}
# else /* CONFIG_SMP */
2624
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
2625 2626 2627 2628
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
2629

P
Peter Zijlstra 已提交
2630 2631 2632
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
2633 2634 2635 2636
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
2637
		account_entity_dequeue(cfs_rq, se);
2638
	}
P
Peter Zijlstra 已提交
2639 2640 2641 2642 2643 2644 2645

	update_load_set(&se->load, weight);

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

2646 2647
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2648
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2649 2650 2651
{
	struct task_group *tg;
	struct sched_entity *se;
2652
	long shares;
P
Peter Zijlstra 已提交
2653 2654 2655

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
2656
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
2657
		return;
2658 2659 2660 2661
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
2662
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
2663 2664 2665 2666

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
2667
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
2668 2669 2670 2671
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2672
#ifdef CONFIG_SMP
2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692
/* Precomputed fixed inverse multiplies for multiplication by y^n */
static const u32 runnable_avg_yN_inv[] = {
	0xffffffff, 0xfa83b2da, 0xf5257d14, 0xefe4b99a, 0xeac0c6e6, 0xe5b906e6,
	0xe0ccdeeb, 0xdbfbb796, 0xd744fcc9, 0xd2a81d91, 0xce248c14, 0xc9b9bd85,
	0xc5672a10, 0xc12c4cc9, 0xbd08a39e, 0xb8fbaf46, 0xb504f333, 0xb123f581,
	0xad583ee9, 0xa9a15ab4, 0xa5fed6a9, 0xa2704302, 0x9ef5325f, 0x9b8d39b9,
	0x9837f050, 0x94f4efa8, 0x91c3d373, 0x8ea4398a, 0x8b95c1e3, 0x88980e80,
	0x85aac367, 0x82cd8698,
};

/*
 * Precomputed \Sum y^k { 1<=k<=n }.  These are floor(true_value) to prevent
 * over-estimates when re-combining.
 */
static const u32 runnable_avg_yN_sum[] = {
	    0, 1002, 1982, 2941, 3880, 4798, 5697, 6576, 7437, 8279, 9103,
	 9909,10698,11470,12226,12966,13690,14398,15091,15769,16433,17082,
	17718,18340,18949,19545,20128,20698,21256,21802,22336,22859,23371,
};

2693 2694 2695 2696 2697 2698 2699 2700 2701 2702
/*
 * Precomputed \Sum y^k { 1<=k<=n, where n%32=0). Values are rolled down to
 * lower integers. See Documentation/scheduler/sched-avg.txt how these
 * were generated:
 */
static const u32 __accumulated_sum_N32[] = {
	    0, 23371, 35056, 40899, 43820, 45281,
	46011, 46376, 46559, 46650, 46696, 46719,
};

2703 2704 2705 2706 2707 2708
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720
	unsigned int local_n;

	if (!n)
		return val;
	else if (unlikely(n > LOAD_AVG_PERIOD * 63))
		return 0;

	/* after bounds checking we can collapse to 32-bit */
	local_n = n;

	/*
	 * As y^PERIOD = 1/2, we can combine
2721 2722
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
2723 2724 2725 2726 2727 2728
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
2729 2730
	}

2731 2732
	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
	return val;
2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750
}

/*
 * For updates fully spanning n periods, the contribution to runnable
 * average will be: \Sum 1024*y^n
 *
 * We can compute this reasonably efficiently by combining:
 *   y^PERIOD = 1/2 with precomputed \Sum 1024*y^n {for  n <PERIOD}
 */
static u32 __compute_runnable_contrib(u64 n)
{
	u32 contrib = 0;

	if (likely(n <= LOAD_AVG_PERIOD))
		return runnable_avg_yN_sum[n];
	else if (unlikely(n >= LOAD_AVG_MAX_N))
		return LOAD_AVG_MAX;

2751 2752 2753
	/* Since n < LOAD_AVG_MAX_N, n/LOAD_AVG_PERIOD < 11 */
	contrib = __accumulated_sum_N32[n/LOAD_AVG_PERIOD];
	n %= LOAD_AVG_PERIOD;
2754 2755
	contrib = decay_load(contrib, n);
	return contrib + runnable_avg_yN_sum[n];
2756 2757
}

2758
#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
2759

2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787
/*
 * We can represent the historical contribution to runnable average as the
 * coefficients of a geometric series.  To do this we sub-divide our runnable
 * history into segments of approximately 1ms (1024us); label the segment that
 * occurred N-ms ago p_N, with p_0 corresponding to the current period, e.g.
 *
 * [<- 1024us ->|<- 1024us ->|<- 1024us ->| ...
 *      p0            p1           p2
 *     (now)       (~1ms ago)  (~2ms ago)
 *
 * Let u_i denote the fraction of p_i that the entity was runnable.
 *
 * We then designate the fractions u_i as our co-efficients, yielding the
 * following representation of historical load:
 *   u_0 + u_1*y + u_2*y^2 + u_3*y^3 + ...
 *
 * We choose y based on the with of a reasonably scheduling period, fixing:
 *   y^32 = 0.5
 *
 * This means that the contribution to load ~32ms ago (u_32) will be weighted
 * approximately half as much as the contribution to load within the last ms
 * (u_0).
 *
 * When a period "rolls over" and we have new u_0`, multiplying the previous
 * sum again by y is sufficient to update:
 *   load_avg = u_0` + y*(u_0 + u_1*y + u_2*y^2 + ... )
 *            = u_0 + u_1*y + u_2*y^2 + ... [re-labeling u_i --> u_{i+1}]
 */
2788 2789
static __always_inline int
__update_load_avg(u64 now, int cpu, struct sched_avg *sa,
2790
		  unsigned long weight, int running, struct cfs_rq *cfs_rq)
2791
{
2792
	u64 delta, scaled_delta, periods;
2793
	u32 contrib;
2794
	unsigned int delta_w, scaled_delta_w, decayed = 0;
2795
	unsigned long scale_freq, scale_cpu;
2796

2797
	delta = now - sa->last_update_time;
2798 2799 2800 2801 2802
	/*
	 * This should only happen when time goes backwards, which it
	 * unfortunately does during sched clock init when we swap over to TSC.
	 */
	if ((s64)delta < 0) {
2803
		sa->last_update_time = now;
2804 2805 2806 2807 2808 2809 2810 2811 2812 2813
		return 0;
	}

	/*
	 * Use 1024ns as the unit of measurement since it's a reasonable
	 * approximation of 1us and fast to compute.
	 */
	delta >>= 10;
	if (!delta)
		return 0;
2814
	sa->last_update_time = now;
2815

2816 2817 2818
	scale_freq = arch_scale_freq_capacity(NULL, cpu);
	scale_cpu = arch_scale_cpu_capacity(NULL, cpu);

2819
	/* delta_w is the amount already accumulated against our next period */
2820
	delta_w = sa->period_contrib;
2821 2822 2823
	if (delta + delta_w >= 1024) {
		decayed = 1;

2824 2825 2826
		/* how much left for next period will start over, we don't know yet */
		sa->period_contrib = 0;

2827 2828 2829 2830 2831 2832
		/*
		 * Now that we know we're crossing a period boundary, figure
		 * out how much from delta we need to complete the current
		 * period and accrue it.
		 */
		delta_w = 1024 - delta_w;
2833
		scaled_delta_w = cap_scale(delta_w, scale_freq);
2834
		if (weight) {
2835 2836 2837 2838 2839
			sa->load_sum += weight * scaled_delta_w;
			if (cfs_rq) {
				cfs_rq->runnable_load_sum +=
						weight * scaled_delta_w;
			}
2840
		}
2841
		if (running)
2842
			sa->util_sum += scaled_delta_w * scale_cpu;
2843 2844 2845 2846 2847 2848 2849

		delta -= delta_w;

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

2850
		sa->load_sum = decay_load(sa->load_sum, periods + 1);
2851 2852 2853 2854
		if (cfs_rq) {
			cfs_rq->runnable_load_sum =
				decay_load(cfs_rq->runnable_load_sum, periods + 1);
		}
2855
		sa->util_sum = decay_load((u64)(sa->util_sum), periods + 1);
2856 2857

		/* Efficiently calculate \sum (1..n_period) 1024*y^i */
2858
		contrib = __compute_runnable_contrib(periods);
2859
		contrib = cap_scale(contrib, scale_freq);
2860
		if (weight) {
2861
			sa->load_sum += weight * contrib;
2862 2863 2864
			if (cfs_rq)
				cfs_rq->runnable_load_sum += weight * contrib;
		}
2865
		if (running)
2866
			sa->util_sum += contrib * scale_cpu;
2867 2868 2869
	}

	/* Remainder of delta accrued against u_0` */
2870
	scaled_delta = cap_scale(delta, scale_freq);
2871
	if (weight) {
2872
		sa->load_sum += weight * scaled_delta;
2873
		if (cfs_rq)
2874
			cfs_rq->runnable_load_sum += weight * scaled_delta;
2875
	}
2876
	if (running)
2877
		sa->util_sum += scaled_delta * scale_cpu;
2878

2879
	sa->period_contrib += delta;
2880

2881 2882
	if (decayed) {
		sa->load_avg = div_u64(sa->load_sum, LOAD_AVG_MAX);
2883 2884 2885 2886
		if (cfs_rq) {
			cfs_rq->runnable_load_avg =
				div_u64(cfs_rq->runnable_load_sum, LOAD_AVG_MAX);
		}
2887
		sa->util_avg = sa->util_sum / LOAD_AVG_MAX;
2888
	}
2889

2890
	return decayed;
2891 2892
}

2893
#ifdef CONFIG_FAIR_GROUP_SCHED
2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908
/**
 * update_tg_load_avg - update the tg's load avg
 * @cfs_rq: the cfs_rq whose avg changed
 * @force: update regardless of how small the difference
 *
 * This function 'ensures': tg->load_avg := \Sum tg->cfs_rq[]->avg.load.
 * However, because tg->load_avg is a global value there are performance
 * considerations.
 *
 * In order to avoid having to look at the other cfs_rq's, we use a
 * differential update where we store the last value we propagated. This in
 * turn allows skipping updates if the differential is 'small'.
 *
 * Updating tg's load_avg is necessary before update_cfs_share() (which is
 * done) and effective_load() (which is not done because it is too costly).
2909
 */
2910
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
2911
{
2912
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
2913

2914 2915 2916 2917 2918 2919
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

2920 2921 2922
	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;
2923
	}
2924
}
2925

2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971
/*
 * Called within set_task_rq() right before setting a task's cpu. The
 * caller only guarantees p->pi_lock is held; no other assumptions,
 * including the state of rq->lock, should be made.
 */
void set_task_rq_fair(struct sched_entity *se,
		      struct cfs_rq *prev, struct cfs_rq *next)
{
	if (!sched_feat(ATTACH_AGE_LOAD))
		return;

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

#ifndef CONFIG_64BIT
		u64 p_last_update_time_copy;
		u64 n_last_update_time_copy;

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

			smp_rmb();

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

		} while (p_last_update_time != p_last_update_time_copy ||
			 n_last_update_time != n_last_update_time_copy);
#else
		p_last_update_time = prev->avg.last_update_time;
		n_last_update_time = next->avg.last_update_time;
#endif
		__update_load_avg(p_last_update_time, cpu_of(rq_of(prev)),
				  &se->avg, 0, 0, NULL);
		se->avg.last_update_time = n_last_update_time;
	}
}
2972
#else /* CONFIG_FAIR_GROUP_SCHED */
2973
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
2974
#endif /* CONFIG_FAIR_GROUP_SCHED */
2975

2976 2977
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq)
{
2978
	if (&this_rq()->cfs == cfs_rq) {
2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
		 * a real problem -- added to that it only calls on the local
		 * CPU, so if we enqueue remotely we'll miss an update, but
		 * the next tick/schedule should update.
		 *
		 * It will not get called when we go idle, because the idle
		 * thread is a different class (!fair), nor will the utilization
		 * number include things like RT tasks.
		 *
		 * As is, the util number is not freq-invariant (we'd have to
		 * implement arch_scale_freq_capacity() for that).
		 *
		 * See cpu_util().
		 */
2995
		cpufreq_update_util(rq_of(cfs_rq), 0);
2996 2997 2998
	}
}

2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015
/*
 * 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)

3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027
/**
 * update_cfs_rq_load_avg - update the cfs_rq's load/util averages
 * @now: current time, as per cfs_rq_clock_task()
 * @cfs_rq: cfs_rq to update
 * @update_freq: should we call cfs_rq_util_change() or will the call do so
 *
 * The cfs_rq avg is the direct sum of all its entities (blocked and runnable)
 * avg. The immediate corollary is that all (fair) tasks must be attached, see
 * post_init_entity_util_avg().
 *
 * cfs_rq->avg is used for task_h_load() and update_cfs_share() for example.
 *
3028 3029 3030 3031
 * 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.
3032
 */
3033 3034
static inline int
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
3035
{
3036
	struct sched_avg *sa = &cfs_rq->avg;
3037
	int decayed, removed_load = 0, removed_util = 0;
3038

3039
	if (atomic_long_read(&cfs_rq->removed_load_avg)) {
3040
		s64 r = atomic_long_xchg(&cfs_rq->removed_load_avg, 0);
3041 3042
		sub_positive(&sa->load_avg, r);
		sub_positive(&sa->load_sum, r * LOAD_AVG_MAX);
3043
		removed_load = 1;
3044
	}
3045

3046 3047
	if (atomic_long_read(&cfs_rq->removed_util_avg)) {
		long r = atomic_long_xchg(&cfs_rq->removed_util_avg, 0);
3048 3049
		sub_positive(&sa->util_avg, r);
		sub_positive(&sa->util_sum, r * LOAD_AVG_MAX);
3050
		removed_util = 1;
3051
	}
3052

3053
	decayed = __update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
3054
		scale_load_down(cfs_rq->load.weight), cfs_rq->curr != NULL, cfs_rq);
3055

3056 3057 3058 3059
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3060

3061 3062
	if (update_freq && (decayed || removed_util))
		cfs_rq_util_change(cfs_rq);
3063

3064
	return decayed || removed_load;
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082
}

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

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

3083
	if (update_cfs_rq_load_avg(now, cfs_rq, true) && update_tg)
3084
		update_tg_load_avg(cfs_rq, 0);
3085 3086
}

3087 3088 3089 3090 3091 3092 3093 3094
/**
 * attach_entity_load_avg - attach this entity to its cfs_rq load avg
 * @cfs_rq: cfs_rq to attach to
 * @se: sched_entity to attach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3095 3096
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3097 3098 3099
	if (!sched_feat(ATTACH_AGE_LOAD))
		goto skip_aging;

3100 3101 3102
	/*
	 * If we got migrated (either between CPUs or between cgroups) we'll
	 * have aged the average right before clearing @last_update_time.
3103 3104
	 *
	 * Or we're fresh through post_init_entity_util_avg().
3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115
	 */
	if (se->avg.last_update_time) {
		__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
				  &se->avg, 0, 0, NULL);

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

3116
skip_aging:
3117 3118 3119 3120 3121
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
	cfs_rq->avg.load_avg += se->avg.load_avg;
	cfs_rq->avg.load_sum += se->avg.load_sum;
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3122 3123

	cfs_rq_util_change(cfs_rq);
3124 3125
}

3126 3127 3128 3129 3130 3131 3132 3133
/**
 * 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.
 */
3134 3135 3136 3137 3138 3139
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
			  &se->avg, se->on_rq * scale_load_down(se->load.weight),
			  cfs_rq->curr == se, NULL);

3140 3141 3142 3143
	sub_positive(&cfs_rq->avg.load_avg, se->avg.load_avg);
	sub_positive(&cfs_rq->avg.load_sum, se->avg.load_sum);
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3144 3145

	cfs_rq_util_change(cfs_rq);
3146 3147
}

3148 3149 3150
/* Add the load generated by se into cfs_rq's load average */
static inline void
enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
3151
{
3152 3153
	struct sched_avg *sa = &se->avg;
	u64 now = cfs_rq_clock_task(cfs_rq);
3154
	int migrated, decayed;
3155

3156 3157
	migrated = !sa->last_update_time;
	if (!migrated) {
3158
		__update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
3159 3160
			se->on_rq * scale_load_down(se->load.weight),
			cfs_rq->curr == se, NULL);
3161
	}
3162

3163
	decayed = update_cfs_rq_load_avg(now, cfs_rq, !migrated);
3164

3165 3166 3167
	cfs_rq->runnable_load_avg += sa->load_avg;
	cfs_rq->runnable_load_sum += sa->load_sum;

3168 3169
	if (migrated)
		attach_entity_load_avg(cfs_rq, se);
3170

3171 3172
	if (decayed || migrated)
		update_tg_load_avg(cfs_rq, 0);
3173 3174
}

3175 3176 3177 3178 3179 3180 3181 3182 3183
/* Remove the runnable load generated by se from cfs_rq's runnable load average */
static inline void
dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_avg(se, 1);

	cfs_rq->runnable_load_avg =
		max_t(long, cfs_rq->runnable_load_avg - se->avg.load_avg, 0);
	cfs_rq->runnable_load_sum =
3184
		max_t(s64,  cfs_rq->runnable_load_sum - se->avg.load_sum, 0);
3185 3186
}

3187
#ifndef CONFIG_64BIT
3188 3189
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3190
	u64 last_update_time_copy;
3191
	u64 last_update_time;
3192

3193 3194 3195 3196 3197
	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);
3198 3199 3200

	return last_update_time;
}
3201
#else
3202 3203 3204 3205
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3206 3207
#endif

3208 3209 3210 3211 3212 3213 3214 3215 3216 3217
/*
 * Task first catches up with cfs_rq, and then subtract
 * itself from the cfs_rq (task must be off the queue now).
 */
void remove_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 last_update_time;

	/*
3218 3219 3220 3221 3222 3223 3224
	 * 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.
3225 3226 3227 3228
	 */

	last_update_time = cfs_rq_last_update_time(cfs_rq);

3229
	__update_load_avg(last_update_time, cpu_of(rq_of(cfs_rq)), &se->avg, 0, 0, NULL);
3230 3231
	atomic_long_add(se->avg.load_avg, &cfs_rq->removed_load_avg);
	atomic_long_add(se->avg.util_avg, &cfs_rq->removed_util_avg);
3232
}
3233

3234 3235 3236 3237 3238 3239 3240 3241 3242 3243
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
	return cfs_rq->runnable_load_avg;
}

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

3244 3245
static int idle_balance(struct rq *this_rq);

3246 3247
#else /* CONFIG_SMP */

3248 3249 3250 3251 3252 3253
static inline int
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
{
	return 0;
}

3254 3255
static inline void update_load_avg(struct sched_entity *se, int not_used)
{
3256
	cpufreq_update_util(rq_of(cfs_rq_of(se)), 0);
3257 3258
}

3259 3260
static inline void
enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3261 3262
static inline void
dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
3263
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3264

3265 3266 3267 3268 3269
static inline void
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3270 3271 3272 3273 3274
static inline int idle_balance(struct rq *rq)
{
	return 0;
}

3275
#endif /* CONFIG_SMP */
3276

P
Peter Zijlstra 已提交
3277 3278 3279 3280 3281 3282 3283 3284 3285
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)
3286
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3287 3288 3289
#endif
}

3290 3291 3292
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3293
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3294

3295 3296 3297 3298 3299 3300
	/*
	 * 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
3301
	if (initial && sched_feat(START_DEBIT))
3302
		vruntime += sched_vslice(cfs_rq, se);
3303

3304
	/* sleeps up to a single latency don't count. */
3305
	if (!initial) {
3306
		unsigned long thresh = sysctl_sched_latency;
3307

3308 3309 3310 3311 3312 3313
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3314

3315
		vruntime -= thresh;
3316 3317
	}

3318
	/* ensure we never gain time by being placed backwards. */
3319
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3320 3321
}

3322 3323
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335
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())  {
3336
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3337 3338 3339 3340 3341 3342 3343
			     "stat_blocked and stat_runtime require the "
			     "kernel parameter schedstats=enabled or "
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362

/*
 * 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)
 *
3363
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374
 *	  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.
 */

3375
static void
3376
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3377
{
3378 3379 3380
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3381
	/*
3382 3383
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3384
	 */
3385
	if (renorm && curr)
3386 3387
		se->vruntime += cfs_rq->min_vruntime;

3388 3389
	update_curr(cfs_rq);

3390
	/*
3391 3392 3393 3394
	 * 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.
3395
	 */
3396 3397 3398
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3399
	enqueue_entity_load_avg(cfs_rq, se);
3400 3401
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
3402

3403
	if (flags & ENQUEUE_WAKEUP)
3404
		place_entity(cfs_rq, se, 0);
3405

3406
	check_schedstat_required();
3407 3408
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3409
	if (!curr)
3410
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3411
	se->on_rq = 1;
3412

3413
	if (cfs_rq->nr_running == 1) {
3414
		list_add_leaf_cfs_rq(cfs_rq);
3415 3416
		check_enqueue_throttle(cfs_rq);
	}
3417 3418
}

3419
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3420
{
3421 3422
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3423
		if (cfs_rq->last != se)
3424
			break;
3425 3426

		cfs_rq->last = NULL;
3427 3428
	}
}
P
Peter Zijlstra 已提交
3429

3430 3431 3432 3433
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3434
		if (cfs_rq->next != se)
3435
			break;
3436 3437

		cfs_rq->next = NULL;
3438
	}
P
Peter Zijlstra 已提交
3439 3440
}

3441 3442 3443 3444
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3445
		if (cfs_rq->skip != se)
3446
			break;
3447 3448

		cfs_rq->skip = NULL;
3449 3450 3451
	}
}

P
Peter Zijlstra 已提交
3452 3453
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3454 3455 3456 3457 3458
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
3459 3460 3461

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
3462 3463
}

3464
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3465

3466
static void
3467
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3468
{
3469 3470 3471 3472
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
3473
	dequeue_entity_load_avg(cfs_rq, se);
3474

3475
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
3476

P
Peter Zijlstra 已提交
3477
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3478

3479
	if (se != cfs_rq->curr)
3480
		__dequeue_entity(cfs_rq, se);
3481
	se->on_rq = 0;
3482
	account_entity_dequeue(cfs_rq, se);
3483 3484

	/*
3485 3486 3487 3488
	 * 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.
3489
	 */
3490
	if (!(flags & DEQUEUE_SLEEP))
3491
		se->vruntime -= cfs_rq->min_vruntime;
3492

3493 3494 3495
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

3496
	update_cfs_shares(cfs_rq);
3497 3498 3499 3500 3501 3502 3503 3504 3505

	/*
	 * Now advance min_vruntime if @se was the entity holding it back,
	 * except when: DEQUEUE_SAVE && !DEQUEUE_MOVE, in this case we'll be
	 * put back on, and if we advance min_vruntime, we'll be placed back
	 * further than we started -- ie. we'll be penalized.
	 */
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
		update_min_vruntime(cfs_rq);
3506 3507 3508 3509 3510
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
3511
static void
I
Ingo Molnar 已提交
3512
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3513
{
3514
	unsigned long ideal_runtime, delta_exec;
3515 3516
	struct sched_entity *se;
	s64 delta;
3517

P
Peter Zijlstra 已提交
3518
	ideal_runtime = sched_slice(cfs_rq, curr);
3519
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
3520
	if (delta_exec > ideal_runtime) {
3521
		resched_curr(rq_of(cfs_rq));
3522 3523 3524 3525 3526
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537
		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;

3538 3539
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
3540

3541 3542
	if (delta < 0)
		return;
3543

3544
	if (delta > ideal_runtime)
3545
		resched_curr(rq_of(cfs_rq));
3546 3547
}

3548
static void
3549
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
3550
{
3551 3552 3553 3554 3555 3556 3557
	/* '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.
		 */
3558
		update_stats_wait_end(cfs_rq, se);
3559
		__dequeue_entity(cfs_rq, se);
3560
		update_load_avg(se, 1);
3561 3562
	}

3563
	update_stats_curr_start(cfs_rq, se);
3564
	cfs_rq->curr = se;
3565

I
Ingo Molnar 已提交
3566 3567 3568 3569 3570
	/*
	 * 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):
	 */
3571
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
3572 3573 3574
		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
3575
	}
3576

3577
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
3578 3579
}

3580 3581 3582
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

3583 3584 3585 3586 3587 3588 3589
/*
 * 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
 */
3590 3591
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
3592
{
3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603
	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 */
3604

3605 3606 3607 3608 3609
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
3610 3611 3612 3613 3614 3615 3616 3617 3618 3619
		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;
		}

3620 3621 3622
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
3623

3624 3625 3626 3627 3628 3629
	/*
	 * 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;

3630 3631 3632 3633 3634 3635
	/*
	 * 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;

3636
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
3637 3638

	return se;
3639 3640
}

3641
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
3642

3643
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
3644 3645 3646 3647 3648 3649
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
3650
		update_curr(cfs_rq);
3651

3652 3653 3654
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

3655
	check_spread(cfs_rq, prev);
3656

3657
	if (prev->on_rq) {
3658
		update_stats_wait_start(cfs_rq, prev);
3659 3660
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
3661
		/* in !on_rq case, update occurred at dequeue */
3662
		update_load_avg(prev, 0);
3663
	}
3664
	cfs_rq->curr = NULL;
3665 3666
}

P
Peter Zijlstra 已提交
3667 3668
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
3669 3670
{
	/*
3671
	 * Update run-time statistics of the 'current'.
3672
	 */
3673
	update_curr(cfs_rq);
3674

3675 3676 3677
	/*
	 * Ensure that runnable average is periodically updated.
	 */
3678
	update_load_avg(curr, 1);
3679
	update_cfs_shares(cfs_rq);
3680

P
Peter Zijlstra 已提交
3681 3682 3683 3684 3685
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
3686
	if (queued) {
3687
		resched_curr(rq_of(cfs_rq));
3688 3689
		return;
	}
P
Peter Zijlstra 已提交
3690 3691 3692 3693 3694 3695 3696 3697
	/*
	 * 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
3698
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
3699
		check_preempt_tick(cfs_rq, curr);
3700 3701
}

3702 3703 3704 3705 3706 3707

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

#ifdef CONFIG_CFS_BANDWIDTH
3708 3709

#ifdef HAVE_JUMP_LABEL
3710
static struct static_key __cfs_bandwidth_used;
3711 3712 3713

static inline bool cfs_bandwidth_used(void)
{
3714
	return static_key_false(&__cfs_bandwidth_used);
3715 3716
}

3717
void cfs_bandwidth_usage_inc(void)
3718
{
3719 3720 3721 3722 3723 3724
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
3725 3726 3727 3728 3729 3730 3731
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

3732 3733
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
3734 3735
#endif /* HAVE_JUMP_LABEL */

3736 3737 3738 3739 3740 3741 3742 3743
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
3744 3745 3746 3747 3748 3749

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

P
Paul Turner 已提交
3750 3751 3752 3753 3754 3755 3756
/*
 * 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
 */
3757
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768
{
	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);
}

3769 3770 3771 3772 3773
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

3774 3775 3776 3777
/* 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))
3778
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
3779

3780
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
3781 3782
}

3783 3784
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
3785 3786 3787
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
3788
	u64 amount = 0, min_amount, expires;
3789 3790 3791 3792 3793 3794 3795

	/* 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;
3796
	else {
P
Peter Zijlstra 已提交
3797
		start_cfs_bandwidth(cfs_b);
3798 3799 3800 3801 3802 3803

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
3804
	}
P
Paul Turner 已提交
3805
	expires = cfs_b->runtime_expires;
3806 3807 3808
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
3809 3810 3811 3812 3813 3814 3815
	/*
	 * we may have advanced our local expiration to account for allowed
	 * spread between our sched_clock and the one on which runtime was
	 * issued.
	 */
	if ((s64)(expires - cfs_rq->runtime_expires) > 0)
		cfs_rq->runtime_expires = expires;
3816 3817

	return cfs_rq->runtime_remaining > 0;
3818 3819
}

P
Paul Turner 已提交
3820 3821 3822 3823 3824
/*
 * 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)
3825
{
P
Paul Turner 已提交
3826 3827 3828
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
3829
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
3830 3831
		return;

P
Paul Turner 已提交
3832 3833 3834 3835 3836 3837 3838 3839 3840
	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
3841 3842 3843
	 * whether the global deadline has advanced. It is valid to compare
	 * cfs_b->runtime_expires without any locks since we only care about
	 * exact equality, so a partial write will still work.
P
Paul Turner 已提交
3844 3845
	 */

3846
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
3847 3848 3849 3850 3851 3852 3853 3854
		/* 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;
	}
}

3855
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
3856 3857
{
	/* dock delta_exec before expiring quota (as it could span periods) */
3858
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
3859 3860 3861
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
3862 3863
		return;

3864 3865 3866 3867 3868
	/*
	 * 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))
3869
		resched_curr(rq_of(cfs_rq));
3870 3871
}

3872
static __always_inline
3873
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
3874
{
3875
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
3876 3877 3878 3879 3880
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

3881 3882
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
3883
	return cfs_bandwidth_used() && cfs_rq->throttled;
3884 3885
}

3886 3887 3888
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
3889
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916
}

/*
 * Ensure that neither of the group entities corresponding to src_cpu or
 * dest_cpu are members of a throttled hierarchy when performing group
 * load-balance operations.
 */
static inline int throttled_lb_pair(struct task_group *tg,
				    int src_cpu, int dest_cpu)
{
	struct cfs_rq *src_cfs_rq, *dest_cfs_rq;

	src_cfs_rq = tg->cfs_rq[src_cpu];
	dest_cfs_rq = tg->cfs_rq[dest_cpu];

	return throttled_hierarchy(src_cfs_rq) ||
	       throttled_hierarchy(dest_cfs_rq);
}

/* updated child weight may affect parent so we have to do this bottom up */
static int tg_unthrottle_up(struct task_group *tg, void *data)
{
	struct rq *rq = data;
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];

	cfs_rq->throttle_count--;
	if (!cfs_rq->throttle_count) {
3917
		/* adjust cfs_rq_clock_task() */
3918
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3919
					     cfs_rq->throttled_clock_task;
3920 3921 3922 3923 3924 3925 3926 3927 3928 3929
	}

	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)];

3930 3931
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3932
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3933 3934 3935 3936 3937
	cfs_rq->throttle_count++;

	return 0;
}

3938
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3939 3940 3941 3942 3943
{
	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
3944
	bool empty;
3945 3946 3947

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

3948
	/* freeze hierarchy runnable averages while throttled */
3949 3950 3951
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968

	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)
3969
		sub_nr_running(rq, task_delta);
3970 3971

	cfs_rq->throttled = 1;
3972
	cfs_rq->throttled_clock = rq_clock(rq);
3973
	raw_spin_lock(&cfs_b->lock);
3974
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
3975

3976 3977 3978 3979 3980
	/*
	 * Add to the _head_ of the list, so that an already-started
	 * distribute_cfs_runtime will not see us
	 */
	list_add_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
3981 3982 3983 3984 3985 3986 3987 3988

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

3989 3990 3991
	raw_spin_unlock(&cfs_b->lock);
}

3992
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3993 3994 3995 3996 3997 3998 3999
{
	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;

4000
	se = cfs_rq->tg->se[cpu_of(rq)];
4001 4002

	cfs_rq->throttled = 0;
4003 4004 4005

	update_rq_clock(rq);

4006
	raw_spin_lock(&cfs_b->lock);
4007
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4008 4009 4010
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4011 4012 4013
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031
	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)
4032
		add_nr_running(rq, task_delta);
4033 4034 4035

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4036
		resched_curr(rq);
4037 4038 4039 4040 4041 4042
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4043 4044
	u64 runtime;
	u64 starting_runtime = remaining;
4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074

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

		raw_spin_lock(&rq->lock);
		if (!cfs_rq_throttled(cfs_rq))
			goto next;

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

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

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

next:
		raw_spin_unlock(&rq->lock);

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

4075
	return starting_runtime - remaining;
4076 4077
}

4078 4079 4080 4081 4082 4083 4084 4085
/*
 * 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)
{
4086
	u64 runtime, runtime_expires;
4087
	int throttled;
4088 4089 4090

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4091
		goto out_deactivate;
4092

4093
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4094
	cfs_b->nr_periods += overrun;
4095

4096 4097 4098 4099 4100 4101
	/*
	 * 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
4102 4103 4104

	__refill_cfs_bandwidth_runtime(cfs_b);

4105 4106 4107
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4108
		return 0;
4109 4110
	}

4111 4112 4113
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4114 4115 4116
	runtime_expires = cfs_b->runtime_expires;

	/*
4117 4118 4119 4120 4121
	 * 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.
4122
	 */
4123 4124
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
4125 4126 4127 4128 4129 4130 4131
		raw_spin_unlock(&cfs_b->lock);
		/* we can't nest cfs_b->lock while distributing bandwidth */
		runtime = distribute_cfs_runtime(cfs_b, runtime,
						 runtime_expires);
		raw_spin_lock(&cfs_b->lock);

		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4132 4133

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4134
	}
4135

4136 4137 4138 4139 4140 4141 4142
	/*
	 * 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;
4143

4144 4145 4146 4147
	return 0;

out_deactivate:
	return 1;
4148
}
4149

4150 4151 4152 4153 4154 4155 4156
/* 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;

4157 4158 4159 4160
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4161
 * hrtimer base being cleared by hrtimer_start. In the case of
4162 4163
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188
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
4189 4190 4191
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220
}

/* 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)
{
4221 4222 4223
	if (!cfs_bandwidth_used())
		return;

4224
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239
		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 */
4240 4241 4242
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4243
		return;
4244
	}
4245

4246
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4247
		runtime = cfs_b->runtime;
4248

4249 4250 4251 4252 4253 4254 4255 4256 4257 4258
	expires = cfs_b->runtime_expires;
	raw_spin_unlock(&cfs_b->lock);

	if (!runtime)
		return;

	runtime = distribute_cfs_runtime(cfs_b, runtime, expires);

	raw_spin_lock(&cfs_b->lock);
	if (expires == cfs_b->runtime_expires)
4259
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4260 4261 4262
	raw_spin_unlock(&cfs_b->lock);
}

4263 4264 4265 4266 4267 4268 4269
/*
 * 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)
{
4270 4271 4272
	if (!cfs_bandwidth_used())
		return;

4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286
	/* 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);
}

4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300
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;
4301
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4302 4303
}

4304
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4305
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4306
{
4307
	if (!cfs_bandwidth_used())
4308
		return false;
4309

4310
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4311
		return false;
4312 4313 4314 4315 4316 4317

	/*
	 * 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))
4318
		return true;
4319 4320

	throttle_cfs_rq(cfs_rq);
4321
	return true;
4322
}
4323 4324 4325 4326 4327

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
4328

4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340
	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;

4341
	raw_spin_lock(&cfs_b->lock);
4342
	for (;;) {
P
Peter Zijlstra 已提交
4343
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4344 4345 4346 4347 4348
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4349 4350
	if (idle)
		cfs_b->period_active = 0;
4351
	raw_spin_unlock(&cfs_b->lock);
4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363

	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
4364
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375
	cfs_b->period_timer.function = sched_cfs_period_timer;
	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	cfs_b->slack_timer.function = sched_cfs_slack_timer;
}

static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
	cfs_rq->runtime_enabled = 0;
	INIT_LIST_HEAD(&cfs_rq->throttled_list);
}

P
Peter Zijlstra 已提交
4376
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4377
{
P
Peter Zijlstra 已提交
4378
	lockdep_assert_held(&cfs_b->lock);
4379

P
Peter Zijlstra 已提交
4380 4381 4382 4383 4384
	if (!cfs_b->period_active) {
		cfs_b->period_active = 1;
		hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
		hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
	}
4385 4386 4387 4388
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4389 4390 4391 4392
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4393 4394 4395 4396
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
	struct cfs_rq *cfs_rq;

	for_each_leaf_cfs_rq(rq, cfs_rq) {
		struct cfs_bandwidth *cfs_b = &cfs_rq->tg->cfs_bandwidth;

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

4410
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421
{
	struct cfs_rq *cfs_rq;

	for_each_leaf_cfs_rq(rq, cfs_rq) {
		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
4422
		cfs_rq->runtime_remaining = 1;
4423 4424 4425 4426 4427 4428
		/*
		 * Offline rq is schedulable till cpu is completely disabled
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

4429 4430 4431 4432 4433 4434
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
4435 4436
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
4437
	return rq_clock_task(rq_of(cfs_rq));
4438 4439
}

4440
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
4441
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
4442
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
4443
static inline void sync_throttle(struct task_group *tg, int cpu) {}
4444
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
4445 4446 4447 4448 4449

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460

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;
}
4461 4462 4463 4464 4465

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) {}
4466 4467
#endif

4468 4469 4470 4471 4472
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) {}
4473
static inline void update_runtime_enabled(struct rq *rq) {}
4474
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
4475 4476 4477

#endif /* CONFIG_CFS_BANDWIDTH */

4478 4479 4480 4481
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
4482 4483 4484 4485 4486 4487
#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);

4488
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
4489

4490
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
4491 4492 4493 4494 4495 4496
		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)
4497
				resched_curr(rq);
P
Peter Zijlstra 已提交
4498 4499
			return;
		}
4500
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
4501 4502
	}
}
4503 4504 4505 4506 4507 4508 4509 4510 4511 4512

/*
 * 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;

4513
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
4514 4515 4516 4517 4518
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
4519
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
4520 4521 4522 4523
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
4524 4525 4526 4527

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
4528 4529
#endif

4530 4531 4532 4533 4534
/*
 * 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:
 */
4535
static void
4536
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4537 4538
{
	struct cfs_rq *cfs_rq;
4539
	struct sched_entity *se = &p->se;
4540

4541 4542 4543 4544 4545 4546 4547 4548
	/*
	 * If in_iowait is set, the code below may not trigger any cpufreq
	 * utilization updates, so do it here explicitly with the IOWAIT flag
	 * passed.
	 */
	if (p->in_iowait)
		cpufreq_update_this_cpu(rq, SCHED_CPUFREQ_IOWAIT);

4549
	for_each_sched_entity(se) {
4550
		if (se->on_rq)
4551 4552
			break;
		cfs_rq = cfs_rq_of(se);
4553
		enqueue_entity(cfs_rq, se, flags);
4554 4555 4556 4557 4558 4559

		/*
		 * 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.
4560
		 */
4561 4562
		if (cfs_rq_throttled(cfs_rq))
			break;
4563
		cfs_rq->h_nr_running++;
4564

4565
		flags = ENQUEUE_WAKEUP;
4566
	}
P
Peter Zijlstra 已提交
4567

P
Peter Zijlstra 已提交
4568
	for_each_sched_entity(se) {
4569
		cfs_rq = cfs_rq_of(se);
4570
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
4571

4572 4573 4574
		if (cfs_rq_throttled(cfs_rq))
			break;

4575
		update_load_avg(se, 1);
4576
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4577 4578
	}

Y
Yuyang Du 已提交
4579
	if (!se)
4580
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
4581

4582
	hrtick_update(rq);
4583 4584
}

4585 4586
static void set_next_buddy(struct sched_entity *se);

4587 4588 4589 4590 4591
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
4592
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
4593 4594
{
	struct cfs_rq *cfs_rq;
4595
	struct sched_entity *se = &p->se;
4596
	int task_sleep = flags & DEQUEUE_SLEEP;
4597 4598 4599

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4600
		dequeue_entity(cfs_rq, se, flags);
4601 4602 4603 4604 4605 4606 4607 4608 4609

		/*
		 * 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;
4610
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4611

4612
		/* Don't dequeue parent if it has other entities besides us */
4613
		if (cfs_rq->load.weight) {
4614 4615
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
4616 4617 4618 4619
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
4620 4621
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
4622
			break;
4623
		}
4624
		flags |= DEQUEUE_SLEEP;
4625
	}
P
Peter Zijlstra 已提交
4626

P
Peter Zijlstra 已提交
4627
	for_each_sched_entity(se) {
4628
		cfs_rq = cfs_rq_of(se);
4629
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
4630

4631 4632 4633
		if (cfs_rq_throttled(cfs_rq))
			break;

4634
		update_load_avg(se, 1);
4635
		update_cfs_shares(cfs_rq);
P
Peter Zijlstra 已提交
4636 4637
	}

Y
Yuyang Du 已提交
4638
	if (!se)
4639
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
4640

4641
	hrtick_update(rq);
4642 4643
}

4644
#ifdef CONFIG_SMP
4645 4646 4647 4648 4649

/* 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);

4650
#ifdef CONFIG_NO_HZ_COMMON
4651 4652 4653 4654 4655
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
4656
 * The exact cpuload calculated at every tick would be:
4657
 *
4658 4659 4660 4661 4662 4663 4664
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
 * If a cpu misses updates for n ticks (as it was idle) and update gets
 * called on the n+1-th tick when cpu may be busy, then we have:
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
4665 4666 4667
 *
 * decay_load_missed() below does efficient calculation of
 *
4668 4669 4670 4671 4672 4673
 *   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())
4674
 *
4675
 * The calculation is approximated on a 128 point scale.
4676 4677
 */
#define DEGRADE_SHIFT		7
4678 4679 4680 4681 4682 4683 4684 4685 4686

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 }
};
4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715

/*
 * 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;
}
4716
#endif /* CONFIG_NO_HZ_COMMON */
4717

4718
/**
4719
 * __cpu_load_update - update the rq->cpu_load[] statistics
4720 4721 4722 4723
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
4724
 * Update rq->cpu_load[] statistics. This function is usually called every
4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750
 * 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
4751
 * term.
4752
 */
4753 4754
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
4755
{
4756
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767
	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 */

4768
		old_load = this_rq->cpu_load[i];
4769
#ifdef CONFIG_NO_HZ_COMMON
4770
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
4771 4772 4773 4774 4775 4776 4777 4778 4779
		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;
		}
4780
#endif
4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795
		new_load = this_load;
		/*
		 * Round up the averaging division if load is increasing. This
		 * prevents us from getting stuck on 9 if the load is 10, for
		 * example.
		 */
		if (new_load > old_load)
			new_load += scale - 1;

		this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i;
	}

	sched_avg_update(this_rq);
}

4796 4797 4798 4799 4800 4801
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
	return cfs_rq_runnable_load_avg(&cpu_rq(cpu)->cfs);
}

4802
#ifdef CONFIG_NO_HZ_COMMON
4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
 * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
 * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
 *
 * Therefore we need to avoid the delta approach from the regular tick when
 * possible since that would seriously skew the load calculation. This is why we
 * use cpu_load_update_periodic() for CPUs out of nohz. However we'll rely on
 * jiffies deltas for updates happening while in nohz mode (idle ticks, idle
 * loop exit, nohz_idle_balance, nohz full exit...)
 *
 * This means we might still be one tick off for nohz periods.
 */

static void cpu_load_update_nohz(struct rq *this_rq,
				 unsigned long curr_jiffies,
				 unsigned long load)
4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830
{
	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.
		 */
4831
		cpu_load_update(this_rq, load, pending_updates);
4832 4833 4834
	}
}

4835 4836 4837 4838
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
4839
static void cpu_load_update_idle(struct rq *this_rq)
4840 4841 4842 4843
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
4844
	if (weighted_cpuload(cpu_of(this_rq)))
4845 4846
		return;

4847
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
4848 4849 4850
}

/*
4851 4852 4853 4854
 * 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.
4855
 */
4856
void cpu_load_update_nohz_start(void)
4857 4858
{
	struct rq *this_rq = this_rq();
4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872

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

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
4873
	unsigned long curr_jiffies = READ_ONCE(jiffies);
4874 4875
	struct rq *this_rq = this_rq();
	unsigned long load;
4876 4877 4878 4879

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

4880
	load = weighted_cpuload(cpu_of(this_rq));
4881
	raw_spin_lock(&this_rq->lock);
4882
	update_rq_clock(this_rq);
4883
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
4884 4885
	raw_spin_unlock(&this_rq->lock);
}
4886 4887 4888 4889 4890 4891 4892 4893
#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)
{
4894
#ifdef CONFIG_NO_HZ_COMMON
4895 4896
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
4897
#endif
4898 4899
	cpu_load_update(this_rq, load, 1);
}
4900 4901 4902 4903

/*
 * Called from scheduler_tick()
 */
4904
void cpu_load_update_active(struct rq *this_rq)
4905
{
4906
	unsigned long load = weighted_cpuload(cpu_of(this_rq));
4907 4908 4909 4910 4911

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
4912 4913
}

4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946
/*
 * Return a low guess at the load of a migration-source cpu weighted
 * according to the scheduling class and "nice" value.
 *
 * We want to under-estimate the load of migration sources, to
 * balance conservatively.
 */
static unsigned long source_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long total = weighted_cpuload(cpu);

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

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

/*
 * Return a high guess at the load of a migration-target cpu weighted
 * according to the scheduling class and "nice" value.
 */
static unsigned long target_load(int cpu, int type)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long total = weighted_cpuload(cpu);

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

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

4947
static unsigned long capacity_of(int cpu)
4948
{
4949
	return cpu_rq(cpu)->cpu_capacity;
4950 4951
}

4952 4953 4954 4955 4956
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

4957 4958 4959
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
4960
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
4961
	unsigned long load_avg = weighted_cpuload(cpu);
4962 4963

	if (nr_running)
4964
		return load_avg / nr_running;
4965 4966 4967 4968

	return 0;
}

4969
#ifdef CONFIG_FAIR_GROUP_SCHED
4970 4971 4972 4973 4974 4975
/*
 * effective_load() calculates the load change as seen from the root_task_group
 *
 * Adding load to a group doesn't make a group heavier, but can cause movement
 * of group shares between cpus. Assuming the shares were perfectly aligned one
 * can calculate the shift in shares.
4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018
 *
 * Calculate the effective load difference if @wl is added (subtracted) to @tg
 * on this @cpu and results in a total addition (subtraction) of @wg to the
 * total group weight.
 *
 * Given a runqueue weight distribution (rw_i) we can compute a shares
 * distribution (s_i) using:
 *
 *   s_i = rw_i / \Sum rw_j						(1)
 *
 * Suppose we have 4 CPUs and our @tg is a direct child of the root group and
 * has 7 equal weight tasks, distributed as below (rw_i), with the resulting
 * shares distribution (s_i):
 *
 *   rw_i = {   2,   4,   1,   0 }
 *   s_i  = { 2/7, 4/7, 1/7,   0 }
 *
 * As per wake_affine() we're interested in the load of two CPUs (the CPU the
 * task used to run on and the CPU the waker is running on), we need to
 * compute the effect of waking a task on either CPU and, in case of a sync
 * wakeup, compute the effect of the current task going to sleep.
 *
 * So for a change of @wl to the local @cpu with an overall group weight change
 * of @wl we can compute the new shares distribution (s'_i) using:
 *
 *   s'_i = (rw_i + @wl) / (@wg + \Sum rw_j)				(2)
 *
 * Suppose we're interested in CPUs 0 and 1, and want to compute the load
 * differences in waking a task to CPU 0. The additional task changes the
 * weight and shares distributions like:
 *
 *   rw'_i = {   3,   4,   1,   0 }
 *   s'_i  = { 3/8, 4/8, 1/8,   0 }
 *
 * We can then compute the difference in effective weight by using:
 *
 *   dw_i = S * (s'_i - s_i)						(3)
 *
 * Where 'S' is the group weight as seen by its parent.
 *
 * Therefore the effective change in loads on CPU 0 would be 5/56 (3/8 - 2/7)
 * times the weight of the group. The effect on CPU 1 would be -4/56 (4/8 -
 * 4/7) times the weight of the group.
5019
 */
P
Peter Zijlstra 已提交
5020
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
5021
{
P
Peter Zijlstra 已提交
5022
	struct sched_entity *se = tg->se[cpu];
5023

5024
	if (!tg->parent)	/* the trivial, non-cgroup case */
5025 5026
		return wl;

P
Peter Zijlstra 已提交
5027
	for_each_sched_entity(se) {
5028 5029
		struct cfs_rq *cfs_rq = se->my_q;
		long W, w = cfs_rq_load_avg(cfs_rq);
P
Peter Zijlstra 已提交
5030

5031
		tg = cfs_rq->tg;
5032

5033 5034 5035
		/*
		 * W = @wg + \Sum rw_j
		 */
5036 5037 5038 5039 5040
		W = wg + atomic_long_read(&tg->load_avg);

		/* Ensure \Sum rw_j >= rw_i */
		W -= cfs_rq->tg_load_avg_contrib;
		W += w;
P
Peter Zijlstra 已提交
5041

5042 5043 5044
		/*
		 * w = rw_i + @wl
		 */
5045
		w += wl;
5046

5047 5048 5049 5050
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
5051
			wl = (w * (long)scale_load_down(tg->shares)) / W;
5052
		else
5053
			wl = scale_load_down(tg->shares);
5054

5055 5056 5057 5058 5059
		/*
		 * Per the above, wl is the new se->load.weight value; since
		 * those are clipped to [MIN_SHARES, ...) do so now. See
		 * calc_cfs_shares().
		 */
5060 5061
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
5062 5063 5064 5065

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
5066
		wl -= se->avg.load_avg;
5067 5068 5069 5070 5071 5072 5073 5074

		/*
		 * Recursively apply this logic to all parent groups to compute
		 * the final effective load change on the root group. Since
		 * only the @tg group gets extra weight, all parent groups can
		 * only redistribute existing shares. @wl is the shift in shares
		 * resulting from this level per the above.
		 */
P
Peter Zijlstra 已提交
5075 5076
		wg = 0;
	}
5077

P
Peter Zijlstra 已提交
5078
	return wl;
5079 5080
}
#else
P
Peter Zijlstra 已提交
5081

5082
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
5083
{
5084
	return wl;
5085
}
P
Peter Zijlstra 已提交
5086

5087 5088
#endif

P
Peter Zijlstra 已提交
5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105
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
5106 5107
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5108
 *
M
Mike Galbraith 已提交
5109
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121
 * 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
5122
 */
5123 5124
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5125 5126
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5127
	int factor = this_cpu_read(sd_llc_size);
5128

M
Mike Galbraith 已提交
5129 5130 5131 5132 5133
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5134 5135
}

5136 5137
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
		       int prev_cpu, int sync)
5138
{
5139
	s64 this_load, load;
5140
	s64 this_eff_load, prev_eff_load;
5141
	int idx, this_cpu;
5142
	struct task_group *tg;
5143
	unsigned long weight;
5144
	int balanced;
5145

5146 5147 5148 5149
	idx	  = sd->wake_idx;
	this_cpu  = smp_processor_id();
	load	  = source_load(prev_cpu, idx);
	this_load = target_load(this_cpu, idx);
5150

5151 5152 5153 5154 5155
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
5156 5157
	if (sync) {
		tg = task_group(current);
5158
		weight = current->se.avg.load_avg;
5159

5160
		this_load += effective_load(tg, this_cpu, -weight, -weight);
5161 5162
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
5163

5164
	tg = task_group(p);
5165
	weight = p->se.avg.load_avg;
5166

5167 5168
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
5169 5170 5171
	 * due to the sync cause above having dropped this_load to 0, we'll
	 * always have an imbalance, but there's really nothing you can do
	 * about that, so that's good too.
5172 5173 5174 5175
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
5176 5177
	this_eff_load = 100;
	this_eff_load *= capacity_of(prev_cpu);
5178

5179 5180
	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= capacity_of(this_cpu);
5181

5182
	if (this_load > 0) {
5183 5184 5185 5186
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);
5187
	}
5188

5189
	balanced = this_eff_load <= prev_eff_load;
5190

5191
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5192

5193 5194
	if (!balanced)
		return 0;
5195

5196 5197
	schedstat_inc(sd->ttwu_move_affine);
	schedstat_inc(p->se.statistics.nr_wakeups_affine);
5198 5199

	return 1;
5200 5201
}

5202 5203 5204 5205 5206
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5207
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5208
		  int this_cpu, int sd_flag)
5209
{
5210
	struct sched_group *idlest = NULL, *group = sd->groups;
5211
	unsigned long min_load = ULONG_MAX, this_load = 0;
5212
	int load_idx = sd->forkexec_idx;
5213
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
5214

5215 5216 5217
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5218 5219 5220 5221
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
5222

5223 5224
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
5225
					tsk_cpus_allowed(p)))
5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243
			continue;

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

		/* Tally up the load of all CPUs in the group */
		avg_load = 0;

		for_each_cpu(i, sched_group_cpus(group)) {
			/* Bias balancing toward cpus of our domain */
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

			avg_load += load;
		}

5244
		/* Adjust by relative CPU capacity of the group */
5245
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) / group->sgc->capacity;
5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266

		if (local_group) {
			this_load = avg_load;
		} else if (avg_load < min_load) {
			min_load = avg_load;
			idlest = group;
		}
	} while (group = group->next, group != sd->groups);

	if (!idlest || 100*this_load < imbalance*min_load)
		return NULL;
	return idlest;
}

/*
 * find_idlest_cpu - find the idlest cpu among the cpus in group.
 */
static int
find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
{
	unsigned long load, min_load = ULONG_MAX;
5267 5268 5269 5270
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5271 5272
	int i;

5273 5274 5275 5276
	/* Check if we have any choice: */
	if (group->group_weight == 1)
		return cpumask_first(sched_group_cpus(group));

5277
	/* Traverse only the allowed CPUs */
5278
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300
		if (idle_cpu(i)) {
			struct rq *rq = cpu_rq(i);
			struct cpuidle_state *idle = idle_get_state(rq);
			if (idle && idle->exit_latency < min_exit_latency) {
				/*
				 * We give priority to a CPU whose idle state
				 * has the smallest exit latency irrespective
				 * of any idle timestamp.
				 */
				min_exit_latency = idle->exit_latency;
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			} else if ((!idle || idle->exit_latency == min_exit_latency) &&
				   rq->idle_stamp > latest_idle_timestamp) {
				/*
				 * If equal or no active idle state, then
				 * the most recently idled CPU might have
				 * a warmer cache.
				 */
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			}
5301
		} else if (shallowest_idle_cpu == -1) {
5302 5303 5304 5305 5306
			load = weighted_cpuload(i);
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
5307 5308 5309
		}
	}

5310
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5311
}
5312

5313
/*
5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378
 * Implement a for_each_cpu() variant that starts the scan at a given cpu
 * (@start), and wraps around.
 *
 * This is used to scan for idle CPUs; such that not all CPUs looking for an
 * idle CPU find the same CPU. The down-side is that tasks tend to cycle
 * through the LLC domain.
 *
 * Especially tbench is found sensitive to this.
 */

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

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

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

	return next;
}

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

#ifdef CONFIG_SCHED_SMT

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

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

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

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

	return def;
}

/*
 * Scans the local SMT mask to see if the entire core is idle, and records this
 * information in sd_llc_shared->has_idle_cores.
 *
 * Since SMT siblings share all cache levels, inspecting this limited remote
 * state should be fairly cheap.
 */
P
Peter Zijlstra 已提交
5379
void __update_idle_core(struct rq *rq)
5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410
{
	int core = cpu_of(rq);
	int cpu;

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

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

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

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

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

P
Peter Zijlstra 已提交
5411 5412 5413
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446
	if (!test_idle_cores(target, false))
		return -1;

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

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

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

		if (idle)
			return core;
	}

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

	return -1;
}

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

P
Peter Zijlstra 已提交
5447 5448 5449
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477
	for_each_cpu(cpu, cpu_smt_mask(target)) {
		if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
			continue;
		if (idle_cpu(cpu))
			return cpu;
	}

	return -1;
}

#else /* CONFIG_SCHED_SMT */

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

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

#endif /* CONFIG_SCHED_SMT */

/*
 * Scan the LLC domain for idle CPUs; this is dynamically regulated by
 * comparing the average scan cost (tracked in sd->avg_scan_cost) against the
 * average idle time for this rq (as found in rq->avg_idle).
5478
 */
5479 5480
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
5481 5482
	struct sched_domain *this_sd;
	u64 avg_cost, avg_idle = this_rq()->avg_idle;
5483 5484 5485 5486
	u64 time, cost;
	s64 delta;
	int cpu, wrap;

5487 5488 5489 5490 5491 5492
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

	avg_cost = this_sd->avg_scan_cost;

5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
	if ((avg_idle / 512) < avg_cost)
		return -1;

	time = local_clock();

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

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

	return cpu;
}

/*
 * Try and locate an idle core/thread in the LLC cache domain.
5519
 */
5520
static int select_idle_sibling(struct task_struct *p, int prev, int target)
5521
{
5522
	struct sched_domain *sd;
5523
	int i;
5524

5525 5526
	if (idle_cpu(target))
		return target;
5527 5528

	/*
5529
	 * If the previous cpu is cache affine and idle, don't be stupid.
5530
	 */
5531 5532
	if (prev != target && cpus_share_cache(prev, target) && idle_cpu(prev))
		return prev;
5533

5534
	sd = rcu_dereference(per_cpu(sd_llc, target));
5535 5536
	if (!sd)
		return target;
5537

5538 5539 5540
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
5541

5542 5543 5544 5545 5546 5547 5548
	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;
5549

5550 5551
	return target;
}
5552

5553
/*
5554
 * cpu_util returns the amount of capacity of a CPU that is used by CFS
5555
 * tasks. The unit of the return value must be the one of capacity so we can
5556 5557
 * compare the utilization with the capacity of the CPU that is available for
 * CFS task (ie cpu_capacity).
5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577
 *
 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
 * recent utilization of currently non-runnable tasks on a CPU. It represents
 * the amount of utilization of a CPU in the range [0..capacity_orig] where
 * capacity_orig is the cpu_capacity available at the highest frequency
 * (arch_scale_freq_capacity()).
 * The utilization of a CPU converges towards a sum equal to or less than the
 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
 * the running time on this CPU scaled by capacity_curr.
 *
 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
 * higher than capacity_orig because of unfortunate rounding in
 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
 * the average stabilizes with the new running time. We need to check that the
 * utilization stays within the range of [0..capacity_orig] and cap it if
 * necessary. Without utilization capping, a group could be seen as overloaded
 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
 * available capacity. We allow utilization to overshoot capacity_curr (but not
 * capacity_orig) as it useful for predicting the capacity required after task
 * migrations (scheduler-driven DVFS).
5578
 */
5579
static int cpu_util(int cpu)
5580
{
5581
	unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
5582 5583
	unsigned long capacity = capacity_orig_of(cpu);

5584
	return (util >= capacity) ? capacity : util;
5585
}
5586

5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612
static inline int task_util(struct task_struct *p)
{
	return p->se.avg.util_avg;
}

/*
 * 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;

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

5613
/*
5614 5615 5616
 * 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.
5617
 *
5618 5619
 * 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.
5620
 *
5621
 * Returns the target cpu number.
5622 5623 5624
 *
 * preempt must be disabled.
 */
5625
static int
5626
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
5627
{
5628
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
5629
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
5630
	int new_cpu = prev_cpu;
5631
	int want_affine = 0;
5632
	int sync = wake_flags & WF_SYNC;
5633

P
Peter Zijlstra 已提交
5634 5635
	if (sd_flag & SD_BALANCE_WAKE) {
		record_wakee(p);
5636 5637
		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
			      && cpumask_test_cpu(cpu, tsk_cpus_allowed(p));
P
Peter Zijlstra 已提交
5638
	}
5639

5640
	rcu_read_lock();
5641
	for_each_domain(cpu, tmp) {
5642
		if (!(tmp->flags & SD_LOAD_BALANCE))
M
Mike Galbraith 已提交
5643
			break;
5644

5645
		/*
5646 5647
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
5648
		 */
5649 5650 5651
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
5652
			break;
5653
		}
5654

5655
		if (tmp->flags & sd_flag)
5656
			sd = tmp;
M
Mike Galbraith 已提交
5657 5658
		else if (!want_affine)
			break;
5659 5660
	}

M
Mike Galbraith 已提交
5661 5662
	if (affine_sd) {
		sd = NULL; /* Prefer wake_affine over balance flags */
5663
		if (cpu != prev_cpu && wake_affine(affine_sd, p, prev_cpu, sync))
M
Mike Galbraith 已提交
5664
			new_cpu = cpu;
5665
	}
5666

M
Mike Galbraith 已提交
5667 5668
	if (!sd) {
		if (sd_flag & SD_BALANCE_WAKE) /* XXX always ? */
5669
			new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
M
Mike Galbraith 已提交
5670 5671

	} else while (sd) {
5672
		struct sched_group *group;
5673
		int weight;
5674

5675
		if (!(sd->flags & sd_flag)) {
5676 5677 5678
			sd = sd->child;
			continue;
		}
5679

5680
		group = find_idlest_group(sd, p, cpu, sd_flag);
5681 5682 5683 5684
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
5685

5686
		new_cpu = find_idlest_cpu(group, p, cpu);
5687 5688 5689 5690
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
5691
		}
5692 5693 5694

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
5695
		weight = sd->span_weight;
5696 5697
		sd = NULL;
		for_each_domain(cpu, tmp) {
5698
			if (weight <= tmp->span_weight)
5699
				break;
5700
			if (tmp->flags & sd_flag)
5701 5702 5703
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
5704
	}
5705
	rcu_read_unlock();
5706

5707
	return new_cpu;
5708
}
5709 5710 5711 5712

/*
 * Called immediately before a task is migrated to a new cpu; task_cpu(p) and
 * cfs_rq_of(p) references at time of call are still valid and identify the
5713
 * previous cpu. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
5714
 */
5715
static void migrate_task_rq_fair(struct task_struct *p)
5716
{
5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742
	/*
	 * 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;
	}

5743
	/*
5744 5745 5746 5747 5748
	 * 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.
5749
	 */
5750 5751 5752 5753
	remove_entity_load_avg(&p->se);

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
5754 5755

	/* We have migrated, no longer consider this task hot */
5756
	p->se.exec_start = 0;
5757
}
5758 5759 5760 5761 5762

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
5763 5764
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
5765 5766
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
5767 5768 5769 5770
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
5771 5772
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
5773 5774 5775 5776 5777 5778 5779 5780 5781
	 *
	 * 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.
5782
	 */
5783
	return calc_delta_fair(gran, se);
5784 5785
}

5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807
/*
 * Should 'se' preempt 'curr'.
 *
 *             |s1
 *        |s2
 *   |s3
 *         g
 *      |<--->|c
 *
 *  w(c, s1) = -1
 *  w(c, s2) =  0
 *  w(c, s3) =  1
 *
 */
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
{
	s64 gran, vdiff = curr->vruntime - se->vruntime;

	if (vdiff <= 0)
		return -1;

P
Peter Zijlstra 已提交
5808
	gran = wakeup_gran(curr, se);
5809 5810 5811 5812 5813 5814
	if (vdiff > gran)
		return 1;

	return 0;
}

5815 5816
static void set_last_buddy(struct sched_entity *se)
{
5817 5818 5819 5820 5821
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
5822 5823 5824 5825
}

static void set_next_buddy(struct sched_entity *se)
{
5826 5827 5828 5829 5830
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
5831 5832
}

5833 5834
static void set_skip_buddy(struct sched_entity *se)
{
5835 5836
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
5837 5838
}

5839 5840 5841
/*
 * Preempt the current task with a newly woken task if needed:
 */
5842
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
5843 5844
{
	struct task_struct *curr = rq->curr;
5845
	struct sched_entity *se = &curr->se, *pse = &p->se;
5846
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
5847
	int scale = cfs_rq->nr_running >= sched_nr_latency;
5848
	int next_buddy_marked = 0;
5849

I
Ingo Molnar 已提交
5850 5851 5852
	if (unlikely(se == pse))
		return;

5853
	/*
5854
	 * This is possible from callers such as attach_tasks(), in which we
5855 5856 5857 5858 5859 5860 5861
	 * 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;

5862
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
5863
		set_next_buddy(pse);
5864 5865
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
5866

5867 5868 5869
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
5870 5871 5872 5873 5874 5875
	 *
	 * 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.
5876 5877 5878 5879
	 */
	if (test_tsk_need_resched(curr))
		return;

5880 5881 5882 5883 5884
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

5885
	/*
5886 5887
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
5888
	 */
5889
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
5890
		return;
5891

5892
	find_matching_se(&se, &pse);
5893
	update_curr(cfs_rq_of(se));
5894
	BUG_ON(!pse);
5895 5896 5897 5898 5899 5900 5901
	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);
5902
		goto preempt;
5903
	}
5904

5905
	return;
5906

5907
preempt:
5908
	resched_curr(rq);
5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922
	/*
	 * 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);
5923 5924
}

5925
static struct task_struct *
5926
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
5927 5928 5929
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
5930
	struct task_struct *p;
5931
	int new_tasks;
5932

5933
again:
5934 5935
#ifdef CONFIG_FAIR_GROUP_SCHED
	if (!cfs_rq->nr_running)
5936
		goto idle;
5937

5938
	if (prev->sched_class != &fair_sched_class)
5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957
		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.
		 */
5958 5959 5960 5961 5962
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
5963

5964 5965 5966 5967 5968 5969 5970 5971 5972
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
			 * Therefore the 'simple' nr_running test will indeed
			 * be correct.
			 */
			if (unlikely(check_cfs_rq_runtime(cfs_rq)))
				goto simple;
		}
5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012

		se = pick_next_entity(cfs_rq, curr);
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

	p = task_of(se);

	/*
	 * Since we haven't yet done put_prev_entity and if the selected task
	 * is a different task than we started out with, try and touch the
	 * least amount of cfs_rqs.
	 */
	if (prev != p) {
		struct sched_entity *pse = &prev->se;

		while (!(cfs_rq = is_same_group(se, pse))) {
			int se_depth = se->depth;
			int pse_depth = pse->depth;

			if (se_depth <= pse_depth) {
				put_prev_entity(cfs_rq_of(pse), pse);
				pse = parent_entity(pse);
			}
			if (se_depth >= pse_depth) {
				set_next_entity(cfs_rq_of(se), se);
				se = parent_entity(se);
			}
		}

		put_prev_entity(cfs_rq, pse);
		set_next_entity(cfs_rq, se);
	}

	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);

	return p;
simple:
	cfs_rq = &rq->cfs;
#endif
6013

6014
	if (!cfs_rq->nr_running)
6015
		goto idle;
6016

6017
	put_prev_task(rq, prev);
6018

6019
	do {
6020
		se = pick_next_entity(cfs_rq, NULL);
6021
		set_next_entity(cfs_rq, se);
6022 6023 6024
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6025
	p = task_of(se);
6026

6027 6028
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6029 6030

	return p;
6031 6032

idle:
6033 6034 6035 6036 6037 6038
	/*
	 * 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.
	 */
6039
	lockdep_unpin_lock(&rq->lock, cookie);
6040
	new_tasks = idle_balance(rq);
6041
	lockdep_repin_lock(&rq->lock, cookie);
6042 6043 6044 6045 6046
	/*
	 * 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.
	 */
6047
	if (new_tasks < 0)
6048 6049
		return RETRY_TASK;

6050
	if (new_tasks > 0)
6051 6052 6053
		goto again;

	return NULL;
6054 6055 6056 6057 6058
}

/*
 * Account for a descheduled task:
 */
6059
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6060 6061 6062 6063 6064 6065
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6066
		put_prev_entity(cfs_rq, se);
6067 6068 6069
	}
}

6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094
/*
 * 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);
6095 6096 6097 6098 6099
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6100
		rq_clock_skip_update(rq, true);
6101 6102 6103 6104 6105
	}

	set_skip_buddy(se);
}

6106 6107 6108 6109
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6110 6111
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6112 6113 6114 6115 6116 6117 6118 6119 6120 6121
		return false;

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

	yield_task_fair(rq);

	return true;
}

6122
#ifdef CONFIG_SMP
6123
/**************************************************
P
Peter Zijlstra 已提交
6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139
 * Fair scheduling class load-balancing methods.
 *
 * BASICS
 *
 * The purpose of load-balancing is to achieve the same basic fairness the
 * per-cpu scheduler provides, namely provide a proportional amount of compute
 * time to each task. This is expressed in the following equation:
 *
 *   W_i,n/P_i == W_j,n/P_j for all i,j                               (1)
 *
 * Where W_i,n is the n-th weight average for cpu i. The instantaneous weight
 * W_i,0 is defined as:
 *
 *   W_i,0 = \Sum_j w_i,j                                             (2)
 *
 * Where w_i,j is the weight of the j-th runnable task on cpu i. This weight
6140
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6141 6142 6143 6144 6145 6146
 *
 * 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)
 *
6147
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
6148 6149 6150 6151 6152 6153
 * 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):
 *
6154
 *   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
6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192
 *
 * We them move tasks around to minimize the imbalance. In the continuous
 * function space it is obvious this converges, in the discrete case we get
 * a few fun cases generally called infeasible weight scenarios.
 *
 * [XXX expand on:
 *     - infeasible weights;
 *     - local vs global optima in the discrete case. ]
 *
 *
 * SCHED DOMAINS
 *
 * In order to solve the imbalance equation (4), and avoid the obvious O(n^2)
 * for all i,j solution, we create a tree of cpus that follows the hardware
 * topology where each level pairs two lower groups (or better). This results
 * in O(log n) layers. Furthermore we reduce the number of cpus going up the
 * tree to only the first of the previous level and we decrease the frequency
 * of load-balance at each level inv. proportional to the number of cpus in
 * the groups.
 *
 * This yields:
 *
 *     log_2 n     1     n
 *   \Sum       { --- * --- * 2^i } = O(n)                            (5)
 *     i = 0      2^i   2^i
 *                               `- size of each group
 *         |         |     `- number of cpus doing load-balance
 *         |         `- freq
 *         `- sum over all levels
 *
 * Coupled with a limit on how many tasks we can migrate every balance pass,
 * this makes (5) the runtime complexity of the balancer.
 *
 * An important property here is that each CPU is still (indirectly) connected
 * to every other cpu in at most O(log n) steps:
 *
 * The adjacency matrix of the resulting graph is given by:
 *
6193
 *             log_2 n
P
Peter Zijlstra 已提交
6194 6195 6196 6197 6198 6199 6200 6201 6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238
 *   A_i,j = \Union     (i % 2^k == 0) && i / 2^(k+1) == j / 2^(k+1)  (6)
 *             k = 0
 *
 * And you'll find that:
 *
 *   A^(log_2 n)_i,j != 0  for all i,j                                (7)
 *
 * Showing there's indeed a path between every cpu in at most O(log n) steps.
 * The task movement gives a factor of O(m), giving a convergence complexity
 * of:
 *
 *   O(nm log n),  n := nr_cpus, m := nr_tasks                        (8)
 *
 *
 * WORK CONSERVING
 *
 * In order to avoid CPUs going idle while there's still work to do, new idle
 * balancing is more aggressive and has the newly idle cpu iterate up the domain
 * tree itself instead of relying on other CPUs to bring it work.
 *
 * This adds some complexity to both (5) and (8) but it reduces the total idle
 * time.
 *
 * [XXX more?]
 *
 *
 * CGROUPS
 *
 * Cgroups make a horror show out of (2), instead of a simple sum we get:
 *
 *                                s_k,i
 *   W_i,0 = \Sum_j \Prod_k w_k * -----                               (9)
 *                                 S_k
 *
 * Where
 *
 *   s_k,i = \Sum_j w_i,j,k  and  S_k = \Sum_i s_k,i                 (10)
 *
 * w_i,j,k is the weight of the j-th runnable task in the k-th cgroup on cpu i.
 *
 * The big problem is S_k, its a global sum needed to compute a local (W_i)
 * property.
 *
 * [XXX write more on how we solve this.. _after_ merging pjt's patches that
 *      rewrite all of this once again.]
6239
 */
6240

6241 6242
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

6243 6244
enum fbq_type { regular, remote, all };

6245
#define LBF_ALL_PINNED	0x01
6246
#define LBF_NEED_BREAK	0x02
6247 6248
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
6249 6250 6251 6252 6253

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
6254
	int			src_cpu;
6255 6256 6257 6258

	int			dst_cpu;
	struct rq		*dst_rq;

6259 6260
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
6261
	enum cpu_idle_type	idle;
6262
	long			imbalance;
6263 6264 6265
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

6266
	unsigned int		flags;
6267 6268 6269 6270

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
6271 6272

	enum fbq_type		fbq_type;
6273
	struct list_head	tasks;
6274 6275
};

6276 6277 6278
/*
 * Is this task likely cache-hot:
 */
6279
static int task_hot(struct task_struct *p, struct lb_env *env)
6280 6281 6282
{
	s64 delta;

6283 6284
	lockdep_assert_held(&env->src_rq->lock);

6285 6286 6287 6288 6289 6290 6291 6292 6293
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
6294
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
6295 6296 6297 6298 6299 6300 6301 6302 6303
			(&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;

6304
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
6305 6306 6307 6308

	return delta < (s64)sysctl_sched_migration_cost;
}

6309
#ifdef CONFIG_NUMA_BALANCING
6310
/*
6311 6312 6313
 * 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.
6314
 */
6315
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
6316
{
6317
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
6318
	unsigned long src_faults, dst_faults;
6319 6320
	int src_nid, dst_nid;

6321
	if (!static_branch_likely(&sched_numa_balancing))
6322 6323
		return -1;

6324
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
6325
		return -1;
6326 6327 6328 6329

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

6330
	if (src_nid == dst_nid)
6331
		return -1;
6332

6333 6334 6335 6336 6337 6338 6339
	/* 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;
	}
6340

6341 6342
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
6343
		return 0;
6344

6345 6346 6347 6348 6349 6350
	if (numa_group) {
		src_faults = group_faults(p, src_nid);
		dst_faults = group_faults(p, dst_nid);
	} else {
		src_faults = task_faults(p, src_nid);
		dst_faults = task_faults(p, dst_nid);
6351 6352
	}

6353
	return dst_faults < src_faults;
6354 6355
}

6356
#else
6357
static inline int migrate_degrades_locality(struct task_struct *p,
6358 6359
					     struct lb_env *env)
{
6360
	return -1;
6361
}
6362 6363
#endif

6364 6365 6366 6367
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
6368
int can_migrate_task(struct task_struct *p, struct lb_env *env)
6369
{
6370
	int tsk_cache_hot;
6371 6372 6373

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

6374 6375
	/*
	 * We do not migrate tasks that are:
6376
	 * 1) throttled_lb_pair, or
6377
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
6378 6379
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
6380
	 */
6381 6382 6383
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

6384
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
6385
		int cpu;
6386

6387
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
6388

6389 6390
		env->flags |= LBF_SOME_PINNED;

6391 6392 6393 6394 6395 6396 6397 6398
		/*
		 * Remember if this task can be migrated to any other cpu in
		 * our sched_group. We may want to revisit it if we couldn't
		 * meet load balance goals by pulling other tasks on src_cpu.
		 *
		 * Also avoid computing new_dst_cpu if we have already computed
		 * one in current iteration.
		 */
6399
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
6400 6401
			return 0;

6402 6403 6404
		/* Prevent to re-select dst_cpu via env's cpus */
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
			if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p))) {
6405
				env->flags |= LBF_DST_PINNED;
6406 6407 6408
				env->new_dst_cpu = cpu;
				break;
			}
6409
		}
6410

6411 6412
		return 0;
	}
6413 6414

	/* Record that we found atleast one task that could run on dst_cpu */
6415
	env->flags &= ~LBF_ALL_PINNED;
6416

6417
	if (task_running(env->src_rq, p)) {
6418
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
6419 6420 6421 6422 6423
		return 0;
	}

	/*
	 * Aggressive migration if:
6424 6425 6426
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
6427
	 */
6428 6429 6430
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
6431

6432
	if (tsk_cache_hot <= 0 ||
6433
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
6434
		if (tsk_cache_hot == 1) {
6435 6436
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
6437
		}
6438 6439 6440
		return 1;
	}

6441
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
6442
	return 0;
6443 6444
}

6445
/*
6446 6447 6448 6449 6450 6451 6452
 * 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;
6453
	deactivate_task(env->src_rq, p, 0);
6454 6455 6456
	set_task_cpu(p, env->dst_cpu);
}

6457
/*
6458
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
6459 6460
 * part of active balancing operations within "domain".
 *
6461
 * Returns a task if successful and NULL otherwise.
6462
 */
6463
static struct task_struct *detach_one_task(struct lb_env *env)
6464 6465 6466
{
	struct task_struct *p, *n;

6467 6468
	lockdep_assert_held(&env->src_rq->lock);

6469 6470 6471
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
6472

6473
		detach_task(p, env);
6474

6475
		/*
6476
		 * Right now, this is only the second place where
6477
		 * lb_gained[env->idle] is updated (other is detach_tasks)
6478
		 * so we can safely collect stats here rather than
6479
		 * inside detach_tasks().
6480
		 */
6481
		schedstat_inc(env->sd->lb_gained[env->idle]);
6482
		return p;
6483
	}
6484
	return NULL;
6485 6486
}

6487 6488
static const unsigned int sched_nr_migrate_break = 32;

6489
/*
6490 6491
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
6492
 *
6493
 * Returns number of detached tasks if successful and 0 otherwise.
6494
 */
6495
static int detach_tasks(struct lb_env *env)
6496
{
6497 6498
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
6499
	unsigned long load;
6500 6501 6502
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
6503

6504
	if (env->imbalance <= 0)
6505
		return 0;
6506

6507
	while (!list_empty(tasks)) {
6508 6509 6510 6511 6512 6513 6514
		/*
		 * 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;

6515
		p = list_first_entry(tasks, struct task_struct, se.group_node);
6516

6517 6518
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
6519
		if (env->loop > env->loop_max)
6520
			break;
6521 6522

		/* take a breather every nr_migrate tasks */
6523
		if (env->loop > env->loop_break) {
6524
			env->loop_break += sched_nr_migrate_break;
6525
			env->flags |= LBF_NEED_BREAK;
6526
			break;
6527
		}
6528

6529
		if (!can_migrate_task(p, env))
6530 6531 6532
			goto next;

		load = task_h_load(p);
6533

6534
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
6535 6536
			goto next;

6537
		if ((load / 2) > env->imbalance)
6538
			goto next;
6539

6540 6541 6542 6543
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
6544
		env->imbalance -= load;
6545 6546

#ifdef CONFIG_PREEMPT
6547 6548
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
6549
		 * kernels will stop after the first task is detached to minimize
6550 6551
		 * the critical section.
		 */
6552
		if (env->idle == CPU_NEWLY_IDLE)
6553
			break;
6554 6555
#endif

6556 6557 6558 6559
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
6560
		if (env->imbalance <= 0)
6561
			break;
6562 6563 6564

		continue;
next:
6565
		list_move_tail(&p->se.group_node, tasks);
6566
	}
6567

6568
	/*
6569 6570 6571
	 * 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().
6572
	 */
6573
	schedstat_add(env->sd->lb_gained[env->idle], detached);
6574

6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586
	return detached;
}

/*
 * attach_task() -- attach the task detached by detach_task() to its new rq.
 */
static void attach_task(struct rq *rq, struct task_struct *p)
{
	lockdep_assert_held(&rq->lock);

	BUG_ON(task_rq(p) != rq);
	activate_task(rq, p, 0);
6587
	p->on_rq = TASK_ON_RQ_QUEUED;
6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615
	check_preempt_curr(rq, p, 0);
}

/*
 * attach_one_task() -- attaches the task returned from detach_one_task() to
 * its new rq.
 */
static void attach_one_task(struct rq *rq, struct task_struct *p)
{
	raw_spin_lock(&rq->lock);
	attach_task(rq, p);
	raw_spin_unlock(&rq->lock);
}

/*
 * attach_tasks() -- attaches all tasks detached by detach_tasks() to their
 * new rq.
 */
static void attach_tasks(struct lb_env *env)
{
	struct list_head *tasks = &env->tasks;
	struct task_struct *p;

	raw_spin_lock(&env->dst_rq->lock);

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
6616

6617 6618 6619 6620
		attach_task(env->dst_rq, p);
	}

	raw_spin_unlock(&env->dst_rq->lock);
6621 6622
}

P
Peter Zijlstra 已提交
6623
#ifdef CONFIG_FAIR_GROUP_SCHED
6624
static void update_blocked_averages(int cpu)
6625 6626
{
	struct rq *rq = cpu_rq(cpu);
6627 6628
	struct cfs_rq *cfs_rq;
	unsigned long flags;
6629

6630 6631
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
6632

6633 6634 6635 6636
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
6637
	for_each_leaf_cfs_rq(rq, cfs_rq) {
6638 6639 6640
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
6641

6642
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true))
6643 6644
			update_tg_load_avg(cfs_rq, 0);
	}
6645
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6646 6647
}

6648
/*
6649
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
6650 6651 6652
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
6653
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
6654
{
6655 6656
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
6657
	unsigned long now = jiffies;
6658
	unsigned long load;
6659

6660
	if (cfs_rq->last_h_load_update == now)
6661 6662
		return;

6663 6664 6665 6666 6667 6668 6669
	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;
	}
6670

6671
	if (!se) {
6672
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
6673 6674 6675 6676 6677
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
6678 6679
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
6680 6681 6682 6683
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
6684 6685
}

6686
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
6687
{
6688
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
6689

6690
	update_cfs_rq_h_load(cfs_rq);
6691
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
6692
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
6693 6694
}
#else
6695
static inline void update_blocked_averages(int cpu)
6696
{
6697 6698 6699 6700 6701 6702
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
	unsigned long flags;

	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
6703
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true);
6704
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6705 6706
}

6707
static unsigned long task_h_load(struct task_struct *p)
6708
{
6709
	return p->se.avg.load_avg;
6710
}
P
Peter Zijlstra 已提交
6711
#endif
6712 6713

/********** Helpers for find_busiest_group ************************/
6714 6715 6716 6717 6718 6719 6720

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

6721 6722 6723 6724 6725 6726 6727
/*
 * 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
6728
	unsigned long load_per_task;
6729
	unsigned long group_capacity;
6730
	unsigned long group_util; /* Total utilization of the group */
6731 6732 6733
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
6734
	enum group_type group_type;
6735
	int group_no_capacity;
6736 6737 6738 6739
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
6740 6741
};

J
Joonsoo Kim 已提交
6742 6743 6744 6745 6746 6747 6748 6749
/*
 * sd_lb_stats - Structure to store the statistics of a sched_domain
 *		 during load balancing.
 */
struct sd_lb_stats {
	struct sched_group *busiest;	/* Busiest group in this sd */
	struct sched_group *local;	/* Local group in this sd */
	unsigned long total_load;	/* Total load of all groups in sd */
6750
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
6751 6752 6753
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
6754
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
6755 6756
};

6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768
static inline void init_sd_lb_stats(struct sd_lb_stats *sds)
{
	/*
	 * Skimp on the clearing to avoid duplicate work. We can avoid clearing
	 * local_stat because update_sg_lb_stats() does a full clear/assignment.
	 * We must however clear busiest_stat::avg_load because
	 * update_sd_pick_busiest() reads this before assignment.
	 */
	*sds = (struct sd_lb_stats){
		.busiest = NULL,
		.local = NULL,
		.total_load = 0UL,
6769
		.total_capacity = 0UL,
6770 6771
		.busiest_stat = {
			.avg_load = 0UL,
6772 6773
			.sum_nr_running = 0,
			.group_type = group_other,
6774 6775 6776 6777
		},
	};
}

6778 6779 6780
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
6781
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
6782 6783
 *
 * Return: The load index.
6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805
 */
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;
}

6806
static unsigned long scale_rt_capacity(int cpu)
6807 6808
{
	struct rq *rq = cpu_rq(cpu);
6809
	u64 total, used, age_stamp, avg;
6810
	s64 delta;
6811

6812 6813 6814 6815
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
6816 6817
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
6818
	delta = __rq_clock_broken(rq) - age_stamp;
6819

6820 6821 6822 6823
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
6824

6825
	used = div_u64(avg, total);
6826

6827 6828
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
6829

6830
	return 1;
6831 6832
}

6833
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
6834
{
6835
	unsigned long capacity = arch_scale_cpu_capacity(sd, cpu);
6836 6837
	struct sched_group *sdg = sd->groups;

6838
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
6839

6840
	capacity *= scale_rt_capacity(cpu);
6841
	capacity >>= SCHED_CAPACITY_SHIFT;
6842

6843 6844
	if (!capacity)
		capacity = 1;
6845

6846 6847
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
6848 6849
}

6850
void update_group_capacity(struct sched_domain *sd, int cpu)
6851 6852 6853
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
6854
	unsigned long capacity;
6855 6856 6857 6858
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
6859
	sdg->sgc->next_update = jiffies + interval;
6860 6861

	if (!child) {
6862
		update_cpu_capacity(sd, cpu);
6863 6864 6865
		return;
	}

6866
	capacity = 0;
6867

P
Peter Zijlstra 已提交
6868 6869 6870 6871 6872 6873
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

6874
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
6875
			struct sched_group_capacity *sgc;
6876
			struct rq *rq = cpu_rq(cpu);
6877

6878
			/*
6879
			 * build_sched_domains() -> init_sched_groups_capacity()
6880 6881 6882
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
6883 6884
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
6885
			 *
6886
			 * This avoids capacity from being 0 and
6887 6888 6889
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
6890
				capacity += capacity_of(cpu);
6891 6892
				continue;
			}
6893

6894 6895
			sgc = rq->sd->groups->sgc;
			capacity += sgc->capacity;
6896
		}
P
Peter Zijlstra 已提交
6897 6898 6899 6900
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
6901
		 */
P
Peter Zijlstra 已提交
6902 6903 6904

		group = child->groups;
		do {
6905
			capacity += group->sgc->capacity;
P
Peter Zijlstra 已提交
6906 6907 6908
			group = group->next;
		} while (group != child->groups);
	}
6909

6910
	sdg->sgc->capacity = capacity;
6911 6912
}

6913
/*
6914 6915 6916
 * 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
6917 6918
 */
static inline int
6919
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
6920
{
6921 6922
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
6923 6924
}

6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
 * groups is inadequate due to tsk_cpus_allowed() constraints.
 *
 * Imagine a situation of two groups of 4 cpus each and 4 tasks each with a
 * cpumask covering 1 cpu of the first group and 3 cpus of the second group.
 * Something like:
 *
 * 	{ 0 1 2 3 } { 4 5 6 7 }
 * 	        *     * * *
 *
 * If we were to balance group-wise we'd place two tasks in the first group and
 * two tasks in the second group. Clearly this is undesired as it will overload
 * cpu 3 and leave one of the cpus in the second group unused.
 *
 * The current solution to this issue is detecting the skew in the first group
6941 6942
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
6943 6944
 *
 * When this is so detected; this group becomes a candidate for busiest; see
6945
 * update_sd_pick_busiest(). And calculate_imbalance() and
6946
 * find_busiest_group() avoid some of the usual balance conditions to allow it
6947 6948 6949 6950 6951 6952 6953
 * 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.
 */

6954
static inline int sg_imbalanced(struct sched_group *group)
6955
{
6956
	return group->sgc->imbalance;
6957 6958
}

6959
/*
6960 6961 6962
 * 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
6963 6964
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
6965 6966 6967 6968 6969
 * 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.
6970
 */
6971 6972
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
6973
{
6974 6975
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
6976

6977
	if ((sgs->group_capacity * 100) >
6978
			(sgs->group_util * env->sd->imbalance_pct))
6979
		return true;
6980

6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996
	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;
6997

6998
	if ((sgs->group_capacity * 100) <
6999
			(sgs->group_util * env->sd->imbalance_pct))
7000
		return true;
7001

7002
	return false;
7003 7004
}

7005 7006 7007
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7008
{
7009
	if (sgs->group_no_capacity)
7010 7011 7012 7013 7014 7015 7016 7017
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7018 7019
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7020
 * @env: The load balancing environment.
7021 7022 7023 7024
 * @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.
7025
 * @overload: Indicate more than one runnable task for any CPU.
7026
 */
7027 7028
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7029 7030
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7031
{
7032
	unsigned long load;
7033
	int i, nr_running;
7034

7035 7036
	memset(sgs, 0, sizeof(*sgs));

7037
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
7038 7039 7040
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
7041
		if (local_group)
7042
			load = target_load(i, load_idx);
7043
		else
7044 7045 7046
			load = source_load(i, load_idx);

		sgs->group_load += load;
7047
		sgs->group_util += cpu_util(i);
7048
		sgs->sum_nr_running += rq->cfs.h_nr_running;
7049

7050 7051
		nr_running = rq->nr_running;
		if (nr_running > 1)
7052 7053
			*overload = true;

7054 7055 7056 7057
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
7058
		sgs->sum_weighted_load += weighted_cpuload(i);
7059 7060 7061 7062
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
7063
			sgs->idle_cpus++;
7064 7065
	}

7066 7067
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
7068
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
7069

7070
	if (sgs->sum_nr_running)
7071
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
7072

7073
	sgs->group_weight = group->group_weight;
7074

7075
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
7076
	sgs->group_type = group_classify(group, sgs);
7077 7078
}

7079 7080
/**
 * update_sd_pick_busiest - return 1 on busiest group
7081
 * @env: The load balancing environment.
7082 7083
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
7084
 * @sgs: sched_group statistics
7085 7086 7087
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
7088 7089 7090
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
7091
 */
7092
static bool update_sd_pick_busiest(struct lb_env *env,
7093 7094
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
7095
				   struct sg_lb_stats *sgs)
7096
{
7097
	struct sg_lb_stats *busiest = &sds->busiest_stat;
7098

7099
	if (sgs->group_type > busiest->group_type)
7100 7101
		return true;

7102 7103 7104 7105 7106 7107 7108 7109
	if (sgs->group_type < busiest->group_type)
		return false;

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

	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
7110 7111
		return true;

7112 7113 7114
	/* No ASYM_PACKING if target cpu is already busy */
	if (env->idle == CPU_NOT_IDLE)
		return true;
7115 7116 7117 7118 7119
	/*
	 * ASYM_PACKING needs to move all the work to the lowest
	 * numbered CPUs in the group, therefore mark all groups
	 * higher than ourself as busy.
	 */
7120
	if (sgs->sum_nr_running && env->dst_cpu < group_first_cpu(sg)) {
7121 7122 7123
		if (!sds->busiest)
			return true;

7124 7125
		/* Prefer to move from highest possible cpu's work */
		if (group_first_cpu(sds->busiest) < group_first_cpu(sg))
7126 7127 7128 7129 7130 7131
			return true;
	}

	return false;
}

7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155 7156 7157 7158 7159 7160 7161
#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 */

7162
/**
7163
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
7164
 * @env: The load balancing environment.
7165 7166
 * @sds: variable to hold the statistics for this sched_domain.
 */
7167
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
7168
{
7169 7170
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
7171
	struct sg_lb_stats tmp_sgs;
7172
	int load_idx, prefer_sibling = 0;
7173
	bool overload = false;
7174 7175 7176 7177

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

7178
	load_idx = get_sd_load_idx(env->sd, env->idle);
7179 7180

	do {
J
Joonsoo Kim 已提交
7181
		struct sg_lb_stats *sgs = &tmp_sgs;
7182 7183
		int local_group;

7184
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
7185 7186 7187
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
7188 7189

			if (env->idle != CPU_NEWLY_IDLE ||
7190 7191
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
7192
		}
7193

7194 7195
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
7196

7197 7198 7199
		if (local_group)
			goto next_group;

7200 7201
		/*
		 * In case the child domain prefers tasks go to siblings
7202
		 * first, lower the sg capacity so that we'll try
7203 7204
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
7205 7206 7207 7208
		 * 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).
7209
		 */
7210
		if (prefer_sibling && sds->local &&
7211 7212 7213
		    group_has_capacity(env, &sds->local_stat) &&
		    (sgs->sum_nr_running > 1)) {
			sgs->group_no_capacity = 1;
7214
			sgs->group_type = group_classify(sg, sgs);
7215
		}
7216

7217
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
7218
			sds->busiest = sg;
J
Joonsoo Kim 已提交
7219
			sds->busiest_stat = *sgs;
7220 7221
		}

7222 7223 7224
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
7225
		sds->total_capacity += sgs->group_capacity;
7226

7227
		sg = sg->next;
7228
	} while (sg != env->sd->groups);
7229 7230 7231

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
7232 7233 7234 7235 7236 7237 7238

	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;
	}

7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257
}

/**
 * check_asym_packing - Check to see if the group is packed into the
 *			sched doman.
 *
 * This is primarily intended to used at the sibling level.  Some
 * cores like POWER7 prefer to use lower numbered SMT threads.  In the
 * case of POWER7, it can move to lower SMT modes only when higher
 * threads are idle.  When in lower SMT modes, the threads will
 * perform better since they share less core resources.  Hence when we
 * have idle threads, we want them to be the higher ones.
 *
 * This packing function is run on idle threads.  It checks to see if
 * the busiest CPU in this domain (core in the P7 case) has a higher
 * CPU number than the packing function is being run on.  Here we are
 * assuming lower CPU number will be equivalent to lower a SMT thread
 * number.
 *
7258
 * Return: 1 when packing is required and a task should be moved to
7259 7260
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
7261
 * @env: The load balancing environment.
7262 7263
 * @sds: Statistics of the sched_domain which is to be packed
 */
7264
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
7265 7266 7267
{
	int busiest_cpu;

7268
	if (!(env->sd->flags & SD_ASYM_PACKING))
7269 7270
		return 0;

7271 7272 7273
	if (env->idle == CPU_NOT_IDLE)
		return 0;

7274 7275 7276 7277
	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
7278
	if (env->dst_cpu > busiest_cpu)
7279 7280
		return 0;

7281
	env->imbalance = DIV_ROUND_CLOSEST(
7282
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
7283
		SCHED_CAPACITY_SCALE);
7284

7285
	return 1;
7286 7287 7288 7289 7290 7291
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
7292
 * @env: The load balancing environment.
7293 7294
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
7295 7296
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
7297
{
7298
	unsigned long tmp, capa_now = 0, capa_move = 0;
7299
	unsigned int imbn = 2;
7300
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
7301
	struct sg_lb_stats *local, *busiest;
7302

J
Joonsoo Kim 已提交
7303 7304
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
7305

J
Joonsoo Kim 已提交
7306 7307 7308 7309
	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;
7310

J
Joonsoo Kim 已提交
7311
	scaled_busy_load_per_task =
7312
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
7313
		busiest->group_capacity;
J
Joonsoo Kim 已提交
7314

7315 7316
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
7317
		env->imbalance = busiest->load_per_task;
7318 7319 7320 7321 7322
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
7323
	 * however we may be able to increase total CPU capacity used by
7324 7325 7326
	 * moving them.
	 */

7327
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
7328
			min(busiest->load_per_task, busiest->avg_load);
7329
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
7330
			min(local->load_per_task, local->avg_load);
7331
	capa_now /= SCHED_CAPACITY_SCALE;
7332 7333

	/* Amount of load we'd subtract */
7334
	if (busiest->avg_load > scaled_busy_load_per_task) {
7335
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
7336
			    min(busiest->load_per_task,
7337
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
7338
	}
7339 7340

	/* Amount of load we'd add */
7341
	if (busiest->avg_load * busiest->group_capacity <
7342
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
7343 7344
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
7345
	} else {
7346
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
7347
		      local->group_capacity;
J
Joonsoo Kim 已提交
7348
	}
7349
	capa_move += local->group_capacity *
7350
		    min(local->load_per_task, local->avg_load + tmp);
7351
	capa_move /= SCHED_CAPACITY_SCALE;
7352 7353

	/* Move if we gain throughput */
7354
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
7355
		env->imbalance = busiest->load_per_task;
7356 7357 7358 7359 7360
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
7361
 * @env: load balance environment
7362 7363
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
7364
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
7365
{
7366
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
7367 7368 7369 7370
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
7371

7372
	if (busiest->group_type == group_imbalanced) {
7373 7374 7375 7376
		/*
		 * In the group_imb case we cannot rely on group-wide averages
		 * to ensure cpu-load equilibrium, look at wider averages. XXX
		 */
J
Joonsoo Kim 已提交
7377 7378
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
7379 7380
	}

7381
	/*
7382 7383 7384 7385
	 * 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:
7386
	 */
7387 7388
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
7389 7390
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
7391 7392
	}

7393 7394 7395 7396 7397
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
7398
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
7399
		if (load_above_capacity > busiest->group_capacity) {
7400
			load_above_capacity -= busiest->group_capacity;
7401
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
7402 7403
			load_above_capacity /= busiest->group_capacity;
		} else
7404
			load_above_capacity = ~0UL;
7405 7406 7407 7408 7409 7410
	}

	/*
	 * We're trying to get all the cpus to the average_load, so we don't
	 * want to push ourselves above the average load, nor do we wish to
	 * reduce the max loaded cpu below the average load. At the same time,
7411 7412
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
7413
	 */
7414
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
7415 7416

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
7417
	env->imbalance = min(
7418 7419
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
7420
	) / SCHED_CAPACITY_SCALE;
7421 7422 7423

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
7424
	 * there is no guarantee that any tasks will be moved so we'll have
7425 7426 7427
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
7428
	if (env->imbalance < busiest->load_per_task)
7429
		return fix_small_imbalance(env, sds);
7430
}
7431

7432 7433 7434 7435
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
7436
 * if there is an imbalance.
7437 7438 7439 7440
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
7441
 * @env: The load balancing environment.
7442
 *
7443
 * Return:	- The busiest group if imbalance exists.
7444
 */
J
Joonsoo Kim 已提交
7445
static struct sched_group *find_busiest_group(struct lb_env *env)
7446
{
J
Joonsoo Kim 已提交
7447
	struct sg_lb_stats *local, *busiest;
7448 7449
	struct sd_lb_stats sds;

7450
	init_sd_lb_stats(&sds);
7451 7452 7453 7454 7455

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
7456
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
7457 7458
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
7459

7460
	/* ASYM feature bypasses nice load balance check */
7461
	if (check_asym_packing(env, &sds))
7462 7463
		return sds.busiest;

7464
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
7465
	if (!sds.busiest || busiest->sum_nr_running == 0)
7466 7467
		goto out_balanced;

7468 7469
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
7470

P
Peter Zijlstra 已提交
7471 7472
	/*
	 * If the busiest group is imbalanced the below checks don't
7473
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
7474 7475
	 * isn't true due to cpus_allowed constraints and the like.
	 */
7476
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
7477 7478
		goto force_balance;

7479
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
7480 7481
	if (env->idle == CPU_NEWLY_IDLE && group_has_capacity(env, local) &&
	    busiest->group_no_capacity)
7482 7483
		goto force_balance;

7484
	/*
7485
	 * If the local group is busier than the selected busiest group
7486 7487
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
7488
	if (local->avg_load >= busiest->avg_load)
7489 7490
		goto out_balanced;

7491 7492 7493 7494
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
7495
	if (local->avg_load >= sds.avg_load)
7496 7497
		goto out_balanced;

7498
	if (env->idle == CPU_IDLE) {
7499
		/*
7500 7501 7502 7503 7504
		 * This cpu is idle. If the busiest group is not overloaded
		 * and there is no imbalance between this and busiest group
		 * wrt idle cpus, it is balanced. The imbalance becomes
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
7505
		 */
7506 7507
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
7508
			goto out_balanced;
7509 7510 7511 7512 7513
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
7514 7515
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
7516
			goto out_balanced;
7517
	}
7518

7519
force_balance:
7520
	/* Looks like there is an imbalance. Compute it */
7521
	calculate_imbalance(env, &sds);
7522 7523 7524
	return sds.busiest;

out_balanced:
7525
	env->imbalance = 0;
7526 7527 7528 7529 7530 7531
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
7532
static struct rq *find_busiest_queue(struct lb_env *env,
7533
				     struct sched_group *group)
7534 7535
{
	struct rq *busiest = NULL, *rq;
7536
	unsigned long busiest_load = 0, busiest_capacity = 1;
7537 7538
	int i;

7539
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
7540
		unsigned long capacity, wl;
7541 7542 7543 7544
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
7545

7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566 7567
		/*
		 * 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;

7568
		capacity = capacity_of(i);
7569

7570
		wl = weighted_cpuload(i);
7571

7572 7573
		/*
		 * When comparing with imbalance, use weighted_cpuload()
7574
		 * which is not scaled with the cpu capacity.
7575
		 */
7576 7577 7578

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
7579 7580
			continue;

7581 7582
		/*
		 * For the load comparisons with the other cpu's, consider
7583 7584 7585
		 * the weighted_cpuload() scaled with the cpu capacity, so
		 * that the load can be moved away from the cpu that is
		 * potentially running at a lower capacity.
7586
		 *
7587
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
7588
		 * multiplication to rid ourselves of the division works out
7589 7590
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
7591
		 */
7592
		if (wl * busiest_capacity > busiest_load * capacity) {
7593
			busiest_load = wl;
7594
			busiest_capacity = capacity;
7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607
			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

7608
static int need_active_balance(struct lb_env *env)
7609
{
7610 7611 7612
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
7613 7614 7615 7616 7617 7618

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
7619
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
7620
			return 1;
7621 7622
	}

7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635
	/*
	 * 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;
	}

7636 7637 7638
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

7639 7640
static int active_load_balance_cpu_stop(void *data);

7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	struct cpumask *sg_cpus, *sg_mask;
	int cpu, balance_cpu = -1;

	/*
	 * In the newly idle case, we will allow all the cpu's
	 * to do the newly idle load balance.
	 */
	if (env->idle == CPU_NEWLY_IDLE)
		return 1;

	sg_cpus = sched_group_cpus(sg);
	sg_mask = sched_group_mask(sg);
	/* Try to find first idle cpu */
	for_each_cpu_and(cpu, sg_cpus, env->cpus) {
		if (!cpumask_test_cpu(cpu, sg_mask) || !idle_cpu(cpu))
			continue;

		balance_cpu = cpu;
		break;
	}

	if (balance_cpu == -1)
		balance_cpu = group_balance_cpu(sg);

	/*
	 * First idle cpu or the first cpu(busiest) in this sched group
	 * is eligible for doing load balancing at this and above domains.
	 */
7672
	return balance_cpu == env->dst_cpu;
7673 7674
}

7675 7676 7677 7678 7679 7680
/*
 * 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,
7681
			int *continue_balancing)
7682
{
7683
	int ld_moved, cur_ld_moved, active_balance = 0;
7684
	struct sched_domain *sd_parent = sd->parent;
7685 7686 7687
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
7688
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
7689

7690 7691
	struct lb_env env = {
		.sd		= sd,
7692 7693
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
7694
		.dst_grpmask    = sched_group_cpus(sd->groups),
7695
		.idle		= idle,
7696
		.loop_break	= sched_nr_migrate_break,
7697
		.cpus		= cpus,
7698
		.fbq_type	= all,
7699
		.tasks		= LIST_HEAD_INIT(env.tasks),
7700 7701
	};

7702 7703 7704 7705
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
7706
	if (idle == CPU_NEWLY_IDLE)
7707 7708
		env.dst_grpmask = NULL;

7709 7710
	cpumask_copy(cpus, cpu_active_mask);

7711
	schedstat_inc(sd->lb_count[idle]);
7712 7713

redo:
7714 7715
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
7716
		goto out_balanced;
7717
	}
7718

7719
	group = find_busiest_group(&env);
7720
	if (!group) {
7721
		schedstat_inc(sd->lb_nobusyg[idle]);
7722 7723 7724
		goto out_balanced;
	}

7725
	busiest = find_busiest_queue(&env, group);
7726
	if (!busiest) {
7727
		schedstat_inc(sd->lb_nobusyq[idle]);
7728 7729 7730
		goto out_balanced;
	}

7731
	BUG_ON(busiest == env.dst_rq);
7732

7733
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
7734

7735 7736 7737
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

7738 7739 7740 7741 7742 7743 7744 7745
	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.
		 */
7746
		env.flags |= LBF_ALL_PINNED;
7747
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
7748

7749
more_balance:
7750
		raw_spin_lock_irqsave(&busiest->lock, flags);
7751 7752 7753 7754 7755

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
7756
		cur_ld_moved = detach_tasks(&env);
7757 7758

		/*
7759 7760 7761 7762 7763
		 * 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.
7764
		 */
7765 7766 7767 7768 7769 7770 7771 7772

		raw_spin_unlock(&busiest->lock);

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

7773
		local_irq_restore(flags);
7774

7775 7776 7777 7778 7779
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798
		/*
		 * Revisit (affine) tasks on src_cpu that couldn't be moved to
		 * us and move them to an alternate dst_cpu in our sched_group
		 * where they can run. The upper limit on how many times we
		 * iterate on same src_cpu is dependent on number of cpus in our
		 * sched_group.
		 *
		 * This changes load balance semantics a bit on who can move
		 * load to a given_cpu. In addition to the given_cpu itself
		 * (or a ilb_cpu acting on its behalf where given_cpu is
		 * nohz-idle), we now have balance_cpu in a position to move
		 * load to given_cpu. In rare situations, this may cause
		 * conflicts (balance_cpu and given_cpu/ilb_cpu deciding
		 * _independently_ and at _same_ time to move some load to
		 * given_cpu) causing exceess load to be moved to given_cpu.
		 * This however should not happen so much in practice and
		 * moreover subsequent load balance cycles should correct the
		 * excess load moved.
		 */
7799
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
7800

7801 7802 7803
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

7804
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
7805
			env.dst_cpu	 = env.new_dst_cpu;
7806
			env.flags	&= ~LBF_DST_PINNED;
7807 7808
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
7809

7810 7811 7812 7813 7814 7815
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
7816

7817 7818 7819 7820
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
7821
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
7822

7823
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
7824 7825 7826
				*group_imbalance = 1;
		}

7827
		/* All tasks on this runqueue were pinned by CPU affinity */
7828
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
7829
			cpumask_clear_cpu(cpu_of(busiest), cpus);
7830 7831 7832
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
7833
				goto redo;
7834
			}
7835
			goto out_all_pinned;
7836 7837 7838 7839
		}
	}

	if (!ld_moved) {
7840
		schedstat_inc(sd->lb_failed[idle]);
7841 7842 7843 7844 7845 7846 7847 7848
		/*
		 * 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++;
7849

7850
		if (need_active_balance(&env)) {
7851 7852
			raw_spin_lock_irqsave(&busiest->lock, flags);

7853 7854 7855
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
7856 7857
			 */
			if (!cpumask_test_cpu(this_cpu,
7858
					tsk_cpus_allowed(busiest->curr))) {
7859 7860
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
7861
				env.flags |= LBF_ALL_PINNED;
7862 7863 7864
				goto out_one_pinned;
			}

7865 7866 7867 7868 7869
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
7870 7871 7872 7873 7874 7875
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
7876

7877
			if (active_balance) {
7878 7879 7880
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
7881
			}
7882

7883
			/* We've kicked active balancing, force task migration. */
7884 7885 7886 7887 7888 7889 7890 7891 7892 7893 7894 7895 7896
			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
7897
		 * detach_tasks).
7898 7899 7900 7901 7902 7903 7904 7905
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917 7918 7919 7920 7921 7922
	/*
	 * 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.
	 */
7923
	schedstat_inc(sd->lb_balanced[idle]);
7924 7925 7926 7927 7928

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
7929
	if (((env.flags & LBF_ALL_PINNED) &&
7930
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
7931 7932 7933
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

7934
	ld_moved = 0;
7935 7936 7937 7938
out:
	return ld_moved;
}

7939 7940 7941 7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954
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
7955
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
7956 7957 7958
{
	unsigned long interval, next;

7959 7960
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
7961 7962 7963 7964 7965 7966
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

7967 7968 7969 7970
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
7971
static int idle_balance(struct rq *this_rq)
7972
{
7973 7974
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
7975 7976
	struct sched_domain *sd;
	int pulled_task = 0;
7977
	u64 curr_cost = 0;
7978

7979 7980 7981 7982 7983 7984
	/*
	 * 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);

7985 7986
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
7987 7988 7989
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
7990
			update_next_balance(sd, &next_balance);
7991 7992
		rcu_read_unlock();

7993
		goto out;
7994
	}
7995

7996 7997
	raw_spin_unlock(&this_rq->lock);

7998
	update_blocked_averages(this_cpu);
7999
	rcu_read_lock();
8000
	for_each_domain(this_cpu, sd) {
8001
		int continue_balancing = 1;
8002
		u64 t0, domain_cost;
8003 8004 8005 8006

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

8007
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
8008
			update_next_balance(sd, &next_balance);
8009
			break;
8010
		}
8011

8012
		if (sd->flags & SD_BALANCE_NEWIDLE) {
8013 8014
			t0 = sched_clock_cpu(this_cpu);

8015
			pulled_task = load_balance(this_cpu, this_rq,
8016 8017
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
8018 8019 8020 8021 8022 8023

			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;
8024
		}
8025

8026
		update_next_balance(sd, &next_balance);
8027 8028 8029 8030 8031 8032

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
8033 8034
			break;
	}
8035
	rcu_read_unlock();
8036 8037 8038

	raw_spin_lock(&this_rq->lock);

8039 8040 8041
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

8042
	/*
8043 8044 8045
	 * 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.
8046
	 */
8047
	if (this_rq->cfs.h_nr_running && !pulled_task)
8048
		pulled_task = 1;
8049

8050 8051 8052
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
8053
		this_rq->next_balance = next_balance;
8054

8055
	/* Is there a task of a high priority class? */
8056
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
8057 8058
		pulled_task = -1;

8059
	if (pulled_task)
8060 8061
		this_rq->idle_stamp = 0;

8062
	return pulled_task;
8063 8064 8065
}

/*
8066 8067 8068 8069
 * active_load_balance_cpu_stop is run by cpu stopper. It pushes
 * running tasks off the busiest CPU onto idle CPUs. It requires at
 * least 1 task to be running on each physical CPU where possible, and
 * avoids physical / logical imbalances.
8070
 */
8071
static int active_load_balance_cpu_stop(void *data)
8072
{
8073 8074
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8075
	int target_cpu = busiest_rq->push_cpu;
8076
	struct rq *target_rq = cpu_rq(target_cpu);
8077
	struct sched_domain *sd;
8078
	struct task_struct *p = NULL;
8079 8080 8081 8082 8083 8084 8085

	raw_spin_lock_irq(&busiest_rq->lock);

	/* make sure the requested cpu hasn't gone down in the meantime */
	if (unlikely(busiest_cpu != smp_processor_id() ||
		     !busiest_rq->active_balance))
		goto out_unlock;
8086 8087 8088

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
8089
		goto out_unlock;
8090 8091 8092 8093 8094 8095 8096 8097 8098

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
	 * Bjorn Helgaas on a 128-cpu setup.
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
8099
	rcu_read_lock();
8100 8101 8102 8103 8104 8105 8106
	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)) {
8107 8108
		struct lb_env env = {
			.sd		= sd,
8109 8110 8111 8112
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
8113 8114 8115
			.idle		= CPU_IDLE,
		};

8116
		schedstat_inc(sd->alb_count);
8117

8118
		p = detach_one_task(&env);
8119
		if (p) {
8120
			schedstat_inc(sd->alb_pushed);
8121 8122 8123
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
8124
			schedstat_inc(sd->alb_failed);
8125
		}
8126
	}
8127
	rcu_read_unlock();
8128 8129
out_unlock:
	busiest_rq->active_balance = 0;
8130 8131 8132 8133 8134 8135 8136
	raw_spin_unlock(&busiest_rq->lock);

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

8137
	return 0;
8138 8139
}

8140 8141 8142 8143 8144
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

8145
#ifdef CONFIG_NO_HZ_COMMON
8146 8147 8148 8149 8150 8151
/*
 * 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.
 */
8152
static struct {
8153
	cpumask_var_t idle_cpus_mask;
8154
	atomic_t nr_cpus;
8155 8156
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
8157

8158
static inline int find_new_ilb(void)
8159
{
8160
	int ilb = cpumask_first(nohz.idle_cpus_mask);
8161

8162 8163 8164 8165
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
8166 8167
}

8168 8169 8170 8171 8172
/*
 * 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).
 */
8173
static void nohz_balancer_kick(void)
8174 8175 8176 8177 8178
{
	int ilb_cpu;

	nohz.next_balance++;

8179
	ilb_cpu = find_new_ilb();
8180

8181 8182
	if (ilb_cpu >= nr_cpu_ids)
		return;
8183

8184
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
8185 8186 8187 8188 8189 8190 8191 8192
		return;
	/*
	 * Use smp_send_reschedule() instead of resched_cpu().
	 * This way we generate a sched IPI on the target cpu which
	 * is idle. And the softirq performing nohz idle load balance
	 * will be run before returning from the IPI.
	 */
	smp_send_reschedule(ilb_cpu);
8193 8194 8195
	return;
}

8196
void nohz_balance_exit_idle(unsigned int cpu)
8197 8198
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
8199 8200 8201 8202 8203 8204 8205
		/*
		 * Completely isolated CPUs don't ever set, so we must test.
		 */
		if (likely(cpumask_test_cpu(cpu, nohz.idle_cpus_mask))) {
			cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
			atomic_dec(&nohz.nr_cpus);
		}
8206 8207 8208 8209
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

8210 8211 8212
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
8213
	int cpu = smp_processor_id();
8214 8215

	rcu_read_lock();
8216
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8217 8218 8219 8220 8221

	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

8222
	atomic_inc(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8223
unlock:
8224 8225 8226 8227 8228 8229
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
8230
	int cpu = smp_processor_id();
8231 8232

	rcu_read_lock();
8233
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8234 8235 8236 8237 8238

	if (!sd || sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 1;

8239
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8240
unlock:
8241 8242 8243
	rcu_read_unlock();
}

8244
/*
8245
 * This routine will record that the cpu is going idle with tick stopped.
8246
 * This info will be used in performing idle load balancing in the future.
8247
 */
8248
void nohz_balance_enter_idle(int cpu)
8249
{
8250 8251 8252 8253 8254 8255
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

8256 8257
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
8258

8259 8260 8261 8262 8263 8264
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

8265 8266 8267
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
8268 8269 8270 8271 8272
}
#endif

static DEFINE_SPINLOCK(balancing);

8273 8274 8275 8276
/*
 * 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.
 */
8277
void update_max_interval(void)
8278 8279 8280 8281
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

8282 8283 8284 8285
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
8286
 * Balancing parameters are set up in init_sched_domains.
8287
 */
8288
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
8289
{
8290
	int continue_balancing = 1;
8291
	int cpu = rq->cpu;
8292
	unsigned long interval;
8293
	struct sched_domain *sd;
8294 8295 8296
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
8297 8298
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
8299

8300
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
8301

8302
	rcu_read_lock();
8303
	for_each_domain(cpu, sd) {
8304 8305 8306 8307 8308 8309 8310 8311 8312 8313 8314 8315
		/*
		 * 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;

8316 8317 8318
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329
		/*
		 * 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;
		}

8330
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
8331 8332 8333 8334 8335 8336 8337 8338

		need_serialize = sd->flags & SD_SERIALIZE;
		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
8339
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
8340
				/*
8341
				 * The LBF_DST_PINNED logic could have changed
8342 8343
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
8344
				 */
8345
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
8346 8347
			}
			sd->last_balance = jiffies;
8348
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
8349 8350 8351 8352 8353 8354 8355 8356
		}
		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;
		}
8357 8358
	}
	if (need_decay) {
8359
		/*
8360 8361
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
8362
		 */
8363 8364
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
8365
	}
8366
	rcu_read_unlock();
8367 8368 8369 8370 8371 8372

	/*
	 * 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.
	 */
8373
	if (likely(update_next_balance)) {
8374
		rq->next_balance = next_balance;
8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388

#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
	}
8389 8390
}

8391
#ifdef CONFIG_NO_HZ_COMMON
8392
/*
8393
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
8394 8395
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
8396
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
8397
{
8398
	int this_cpu = this_rq->cpu;
8399 8400
	struct rq *rq;
	int balance_cpu;
8401 8402 8403
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
8404

8405 8406 8407
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
8408 8409

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
8410
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
8411 8412 8413 8414 8415 8416 8417
			continue;

		/*
		 * If this cpu gets work to do, stop the load balancing
		 * work being done for other cpus. Next load
		 * balancing owner will pick it up.
		 */
8418
		if (need_resched())
8419 8420
			break;

V
Vincent Guittot 已提交
8421 8422
		rq = cpu_rq(balance_cpu);

8423 8424 8425 8426 8427 8428 8429
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
			raw_spin_lock_irq(&rq->lock);
			update_rq_clock(rq);
8430
			cpu_load_update_idle(rq);
8431 8432 8433
			raw_spin_unlock_irq(&rq->lock);
			rebalance_domains(rq, CPU_IDLE);
		}
8434

8435 8436 8437 8438
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
8439
	}
8440 8441 8442 8443 8444 8445 8446 8447

	/*
	 * 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;
8448 8449
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
8450 8451 8452
}

/*
8453
 * Current heuristic for kicking the idle load balancer in the presence
8454
 * of an idle cpu in the system.
8455
 *   - This rq has more than one task.
8456 8457 8458 8459
 *   - 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.
8460 8461
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
8462
 */
8463
static inline bool nohz_kick_needed(struct rq *rq)
8464 8465
{
	unsigned long now = jiffies;
8466
	struct sched_domain_shared *sds;
8467
	struct sched_domain *sd;
8468
	int nr_busy, cpu = rq->cpu;
8469
	bool kick = false;
8470

8471
	if (unlikely(rq->idle_balance))
8472
		return false;
8473

8474 8475 8476 8477
       /*
	* 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.
	*/
8478
	set_cpu_sd_state_busy();
8479
	nohz_balance_exit_idle(cpu);
8480 8481 8482 8483 8484 8485

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
8486
		return false;
8487 8488

	if (time_before(now, nohz.next_balance))
8489
		return false;
8490

8491
	if (rq->nr_running >= 2)
8492
		return true;
8493

8494
	rcu_read_lock();
8495 8496 8497 8498 8499 8500 8501
	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);
8502 8503 8504 8505 8506
		if (nr_busy > 1) {
			kick = true;
			goto unlock;
		}

8507
	}
8508

8509 8510 8511 8512 8513 8514 8515 8516
	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
			kick = true;
			goto unlock;
		}
	}
8517

8518
	sd = rcu_dereference(per_cpu(sd_asym, cpu));
8519
	if (sd && (cpumask_first_and(nohz.idle_cpus_mask,
8520 8521 8522 8523
				  sched_domain_span(sd)) < cpu)) {
		kick = true;
		goto unlock;
	}
8524

8525
unlock:
8526
	rcu_read_unlock();
8527
	return kick;
8528 8529
}
#else
8530
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
8531 8532 8533 8534 8535 8536
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
8537
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
8538
{
8539
	struct rq *this_rq = this_rq();
8540
	enum cpu_idle_type idle = this_rq->idle_balance ?
8541 8542 8543
						CPU_IDLE : CPU_NOT_IDLE;

	/*
8544
	 * If this cpu has a pending nohz_balance_kick, then do the
8545
	 * balancing on behalf of the other idle cpus whose ticks are
8546 8547 8548 8549
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
	 * give the idle cpus a chance to load balance. Else we may
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
8550
	 */
8551
	nohz_idle_balance(this_rq, idle);
8552
	rebalance_domains(this_rq, idle);
8553 8554 8555 8556 8557
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
8558
void trigger_load_balance(struct rq *rq)
8559 8560
{
	/* Don't need to rebalance while attached to NULL domain */
8561 8562 8563 8564
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
8565
		raise_softirq(SCHED_SOFTIRQ);
8566
#ifdef CONFIG_NO_HZ_COMMON
8567
	if (nohz_kick_needed(rq))
8568
		nohz_balancer_kick();
8569
#endif
8570 8571
}

8572 8573 8574
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
8575 8576

	update_runtime_enabled(rq);
8577 8578 8579 8580 8581
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
8582 8583 8584

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
8585 8586
}

8587
#endif /* CONFIG_SMP */
8588

8589 8590 8591
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
8592
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
8593 8594 8595 8596 8597 8598
{
	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
8599
		entity_tick(cfs_rq, se, queued);
8600
	}
8601

8602
	if (static_branch_unlikely(&sched_numa_balancing))
8603
		task_tick_numa(rq, curr);
8604 8605 8606
}

/*
P
Peter Zijlstra 已提交
8607 8608 8609
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
8610
 */
P
Peter Zijlstra 已提交
8611
static void task_fork_fair(struct task_struct *p)
8612
{
8613 8614
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
8615
	struct rq *rq = this_rq();
8616

8617
	raw_spin_lock(&rq->lock);
8618 8619
	update_rq_clock(rq);

8620 8621
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
8622 8623
	if (curr) {
		update_curr(cfs_rq);
8624
		se->vruntime = curr->vruntime;
8625
	}
8626
	place_entity(cfs_rq, se, 1);
8627

P
Peter Zijlstra 已提交
8628
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
8629
		/*
8630 8631 8632
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
8633
		swap(curr->vruntime, se->vruntime);
8634
		resched_curr(rq);
8635
	}
8636

8637
	se->vruntime -= cfs_rq->min_vruntime;
8638
	raw_spin_unlock(&rq->lock);
8639 8640
}

8641 8642 8643 8644
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
8645 8646
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
8647
{
8648
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
8649 8650
		return;

8651 8652 8653 8654 8655
	/*
	 * 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
8656
	if (rq->curr == p) {
8657
		if (p->prio > oldprio)
8658
			resched_curr(rq);
8659
	} else
8660
		check_preempt_curr(rq, p, 0);
8661 8662
}

8663
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
8664 8665 8666 8667
{
	struct sched_entity *se = &p->se;

	/*
8668 8669 8670 8671 8672 8673 8674 8675 8676 8677
	 * 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
8678
	 *
8679 8680 8681 8682
	 * - 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
8683
	 */
8684 8685 8686 8687 8688 8689 8690 8691 8692 8693
	if (!se->sum_exec_runtime || p->state == TASK_WAKING)
		return true;

	return false;
}

static void detach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
8694
	u64 now = cfs_rq_clock_task(cfs_rq);
8695 8696

	if (!vruntime_normalized(p)) {
P
Peter Zijlstra 已提交
8697 8698 8699 8700 8701 8702 8703
		/*
		 * 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;
	}
8704

8705
	/* Catch up with the cfs_rq and remove our load when we leave */
8706
	update_cfs_rq_load_avg(now, cfs_rq, false);
8707
	detach_entity_load_avg(cfs_rq, se);
8708
	update_tg_load_avg(cfs_rq, false);
P
Peter Zijlstra 已提交
8709 8710
}

8711
static void attach_task_cfs_rq(struct task_struct *p)
8712
{
8713
	struct sched_entity *se = &p->se;
8714
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
8715
	u64 now = cfs_rq_clock_task(cfs_rq);
8716 8717

#ifdef CONFIG_FAIR_GROUP_SCHED
8718 8719 8720 8721 8722 8723
	/*
	 * 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
8724

8725
	/* Synchronize task with its cfs_rq */
8726
	update_cfs_rq_load_avg(now, cfs_rq, false);
8727
	attach_entity_load_avg(cfs_rq, se);
8728
	update_tg_load_avg(cfs_rq, false);
8729 8730 8731 8732

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
8733

8734 8735 8736 8737 8738 8739 8740 8741
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);
8742

8743
	if (task_on_rq_queued(p)) {
8744
		/*
8745 8746 8747
		 * 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.
8748
		 */
8749 8750 8751 8752
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
8753
	}
8754 8755
}

8756 8757 8758 8759 8760 8761 8762 8763 8764
/* 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;

8765 8766 8767 8768 8769 8770 8771
	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);
	}
8772 8773
}

8774 8775 8776 8777 8778 8779 8780
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
	cfs_rq->tasks_timeline = RB_ROOT;
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
8781
#ifdef CONFIG_SMP
8782 8783
	atomic_long_set(&cfs_rq->removed_load_avg, 0);
	atomic_long_set(&cfs_rq->removed_util_avg, 0);
8784
#endif
8785 8786
}

P
Peter Zijlstra 已提交
8787
#ifdef CONFIG_FAIR_GROUP_SCHED
8788 8789 8790 8791 8792 8793 8794 8795
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;
}

8796
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
8797
{
8798
	detach_task_cfs_rq(p);
8799
	set_task_rq(p, task_cpu(p));
8800 8801 8802 8803 8804

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
8805
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
8806
}
8807

8808 8809 8810 8811 8812 8813 8814 8815 8816 8817 8818 8819 8820
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;
	}
}

8821 8822 8823 8824 8825 8826 8827 8828 8829
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]);
8830
		if (tg->se)
8831 8832 8833 8834 8835 8836 8837 8838 8839 8840
			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;
8841
	struct cfs_rq *cfs_rq;
8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867
	int i;

	tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->cfs_rq)
		goto err;
	tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL);
	if (!tg->se)
		goto err;

	tg->shares = NICE_0_LOAD;

	init_cfs_bandwidth(tg_cfs_bandwidth(tg));

	for_each_possible_cpu(i) {
		cfs_rq = kzalloc_node(sizeof(struct cfs_rq),
				      GFP_KERNEL, cpu_to_node(i));
		if (!cfs_rq)
			goto err;

		se = kzalloc_node(sizeof(struct sched_entity),
				  GFP_KERNEL, cpu_to_node(i));
		if (!se)
			goto err_free_rq;

		init_cfs_rq(cfs_rq);
		init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]);
8868
		init_entity_runnable_average(se);
8869 8870 8871 8872 8873 8874 8875 8876 8877 8878
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

8879 8880 8881 8882 8883 8884 8885 8886 8887 8888 8889 8890
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];

		raw_spin_lock_irq(&rq->lock);
		post_init_entity_util_avg(se);
8891
		sync_throttle(tg, i);
8892 8893 8894 8895
		raw_spin_unlock_irq(&rq->lock);
	}
}

8896
void unregister_fair_sched_group(struct task_group *tg)
8897 8898
{
	unsigned long flags;
8899 8900
	struct rq *rq;
	int cpu;
8901

8902 8903 8904
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
8905

8906 8907 8908 8909 8910 8911 8912 8913 8914 8915 8916 8917 8918
		/*
		 * 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);
	}
8919 8920 8921 8922 8923 8924 8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935 8936 8937
}

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
8938
	if (!parent) {
8939
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
8940 8941
		se->depth = 0;
	} else {
8942
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
8943 8944
		se->depth = parent->depth + 1;
	}
8945 8946

	se->my_q = cfs_rq;
8947 8948
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
8949 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
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;
	unsigned long flags;

	/*
	 * We can't change the weight of the root cgroup.
	 */
	if (!tg->se[0])
		return -EINVAL;

	shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	mutex_lock(&shares_mutex);
	if (tg->shares == shares)
		goto done;

	tg->shares = shares;
	for_each_possible_cpu(i) {
		struct rq *rq = cpu_rq(i);
		struct sched_entity *se;

		se = tg->se[i];
		/* Propagate contribution to hierarchy */
		raw_spin_lock_irqsave(&rq->lock, flags);
8979 8980 8981

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
8982
		for_each_sched_entity(se)
8983 8984 8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999
			update_cfs_shares(group_cfs_rq(se));
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}

done:
	mutex_unlock(&shares_mutex);
	return 0;
}
#else /* CONFIG_FAIR_GROUP_SCHED */

void free_fair_sched_group(struct task_group *tg) { }

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	return 1;
}

9000 9001
void online_fair_sched_group(struct task_group *tg) { }

9002
void unregister_fair_sched_group(struct task_group *tg) { }
9003 9004 9005

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
9006

9007
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
9008 9009 9010 9011 9012 9013 9014 9015 9016
{
	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)
9017
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
9018 9019 9020 9021

	return rr_interval;
}

9022 9023 9024
/*
 * All the scheduling class methods:
 */
9025
const struct sched_class fair_sched_class = {
9026
	.next			= &idle_sched_class,
9027 9028 9029
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
9030
	.yield_to_task		= yield_to_task_fair,
9031

I
Ingo Molnar 已提交
9032
	.check_preempt_curr	= check_preempt_wakeup,
9033 9034 9035 9036

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

9037
#ifdef CONFIG_SMP
L
Li Zefan 已提交
9038
	.select_task_rq		= select_task_rq_fair,
9039
	.migrate_task_rq	= migrate_task_rq_fair,
9040

9041 9042
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
9043

9044
	.task_dead		= task_dead_fair,
9045
	.set_cpus_allowed	= set_cpus_allowed_common,
9046
#endif
9047

9048
	.set_curr_task          = set_curr_task_fair,
9049
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
9050
	.task_fork		= task_fork_fair,
9051 9052

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
9053
	.switched_from		= switched_from_fair,
9054
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
9055

9056 9057
	.get_rr_interval	= get_rr_interval_fair,

9058 9059
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
9060
#ifdef CONFIG_FAIR_GROUP_SCHED
9061
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
9062
#endif
9063 9064 9065
};

#ifdef CONFIG_SCHED_DEBUG
9066
void print_cfs_stats(struct seq_file *m, int cpu)
9067 9068 9069
{
	struct cfs_rq *cfs_rq;

9070
	rcu_read_lock();
9071
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
9072
		print_cfs_rq(m, cpu, cfs_rq);
9073
	rcu_read_unlock();
9074
}
9075 9076 9077 9078 9079 9080 9081 9082 9083 9084 9085 9086 9087 9088 9089 9090 9091 9092 9093 9094 9095

#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 */
9096 9097 9098 9099 9100 9101

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

9102
#ifdef CONFIG_NO_HZ_COMMON
9103
	nohz.next_balance = jiffies;
9104 9105 9106 9107 9108
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}