fair.c 186.4 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 20
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
21 22
 */

A
Arjan van de Ven 已提交
23
#include <linux/latencytop.h>
24
#include <linux/sched.h>
25
#include <linux/cpumask.h>
26 27 28
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
29
#include <linux/mempolicy.h>
30
#include <linux/migrate.h>
31
#include <linux/task_work.h>
32 33 34 35

#include <trace/events/sched.h>

#include "sched.h"
A
Arjan van de Ven 已提交
36

37
/*
38
 * Targeted preemption latency for CPU-bound tasks:
39
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
40
 *
41
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
42 43 44
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
45
 *
I
Ingo Molnar 已提交
46 47
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
48
 */
49 50
unsigned int sysctl_sched_latency = 6000000ULL;
unsigned int normalized_sysctl_sched_latency = 6000000ULL;
51

52 53 54 55 56 57 58 59 60 61 62 63
/*
 * 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;

64
/*
65
 * Minimal preemption granularity for CPU-bound tasks:
66
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
67
 */
68 69
unsigned int sysctl_sched_min_granularity = 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
70 71

/*
72 73
 * is kept at sysctl_sched_latency / sysctl_sched_min_granularity
 */
74
static unsigned int sched_nr_latency = 8;
75 76

/*
77
 * After fork, child runs first. If set to 0 (default) then
78
 * parent will (try to) run first.
79
 */
80
unsigned int sysctl_sched_child_runs_first __read_mostly;
81 82 83

/*
 * SCHED_OTHER wake-up granularity.
84
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
85 86 87 88 89
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
 */
90
unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
91
unsigned int normalized_sysctl_sched_wakeup_granularity = 1000000UL;
92

93 94
const_debug unsigned int sysctl_sched_migration_cost = 500000UL;

95 96 97 98 99 100 101
/*
 * The exponential sliding  window over which load is averaged for shares
 * distribution.
 * (default: 10msec)
 */
unsigned int __read_mostly sysctl_sched_shares_window = 10000000UL;

102 103 104 105 106 107 108 109 110 111 112 113 114 115
#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

116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
static inline void update_load_add(struct load_weight *lw, unsigned long inc)
{
	lw->weight += inc;
	lw->inv_weight = 0;
}

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

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

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
/*
 * 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:
 */
static int get_update_sysctl_factor(void)
{
	unsigned int cpus = min_t(int, num_online_cpus(), 8);
	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();
}

#if BITS_PER_LONG == 32
# define WMULT_CONST	(~0UL)
#else
# define WMULT_CONST	(1UL << 32)
#endif

#define WMULT_SHIFT	32

/*
 * Shift right and round:
 */
#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))

/*
 * delta *= weight / lw
 */
static unsigned long
calc_delta_mine(unsigned long delta_exec, unsigned long weight,
		struct load_weight *lw)
{
	u64 tmp;

	/*
	 * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched
	 * entities since MIN_SHARES = 2. Treat weight as 1 if less than
	 * 2^SCHED_LOAD_RESOLUTION.
	 */
	if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION)))
		tmp = (u64)delta_exec * scale_load_down(weight);
	else
		tmp = (u64)delta_exec;

	if (!lw->inv_weight) {
		unsigned long 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;
	}

	/*
	 * Check whether we'd overflow the 64-bit multiplication:
	 */
	if (unlikely(tmp > WMULT_CONST))
		tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight,
			WMULT_SHIFT/2);
	else
		tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT);

	return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
}


const struct sched_class fair_sched_class;
238

239 240 241 242
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

243
#ifdef CONFIG_FAIR_GROUP_SCHED
244

245
/* cpu runqueue to which this cfs_rq is attached */
246 247
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
248
	return cfs_rq->rq;
249 250
}

251 252
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
253

254 255 256 257 258 259 260 261
static inline struct task_struct *task_of(struct sched_entity *se)
{
#ifdef CONFIG_SCHED_DEBUG
	WARN_ON_ONCE(!entity_is_task(se));
#endif
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
/* Walk up scheduling entities hierarchy */
#define for_each_sched_entity(se) \
		for (; se; se = se->parent)

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

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

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

283 284
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
				       int force_update);
285

286 287 288
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
289 290 291 292 293 294 295 296 297 298 299 300
		/*
		 * 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,
301
				&rq_of(cfs_rq)->leaf_cfs_rq_list);
302
		}
303 304

		cfs_rq->on_list = 1;
305
		/* We should have no load, but we need to update last_decay. */
306
		update_cfs_rq_blocked_load(cfs_rq, 0);
307 308 309 310 311 312 313 314 315 316 317
	}
}

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
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
/* 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 ? */
static inline int
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
		return 1;

	return 0;
}

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

337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
/* return depth at which a sched entity is present in the hierarchy */
static inline int depth_se(struct sched_entity *se)
{
	int depth = 0;

	for_each_sched_entity(se)
		depth++;

	return depth;
}

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 */
	se_depth = depth_se(*se);
	pse_depth = depth_se(*pse);

	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);
	}
}

380 381 382 383 384 385
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
386

387 388 389
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
390 391 392 393
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
394 395
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
396

P
Peter Zijlstra 已提交
397
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
398
{
P
Peter Zijlstra 已提交
399
	return &task_rq(p)->cfs;
400 401
}

P
Peter Zijlstra 已提交
402 403 404 405 406 407 408 409 410 411 412 413 414 415
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;
}

416 417 418 419 420 421 422 423
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
424 425 426 427 428 429 430 431 432 433 434 435 436 437
#define for_each_leaf_cfs_rq(rq, cfs_rq) \
		for (cfs_rq = &rq->cfs; cfs_rq; cfs_rq = NULL)

static inline int
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	return 1;
}

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

438 439 440 441 442
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
443 444
#endif	/* CONFIG_FAIR_GROUP_SCHED */

445 446
static __always_inline
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec);
447 448 449 450 451

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

452
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
453
{
454
	s64 delta = (s64)(vruntime - max_vruntime);
455
	if (delta > 0)
456
		max_vruntime = vruntime;
457

458
	return max_vruntime;
459 460
}

461
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
462 463 464 465 466 467 468 469
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

470 471 472 473 474 475
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

476 477 478 479 480 481 482 483 484 485 486 487
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
	u64 vruntime = cfs_rq->min_vruntime;

	if (cfs_rq->curr)
		vruntime = cfs_rq->curr->vruntime;

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

P
Peter Zijlstra 已提交
488
		if (!cfs_rq->curr)
489 490 491 492 493
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

494
	/* ensure we never gain time by being placed backwards. */
495
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
496 497 498 499
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
500 501
}

502 503 504
/*
 * Enqueue an entity into the rb-tree:
 */
505
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521
{
	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.
		 */
522
		if (entity_before(se, entry)) {
523 524 525 526 527 528 529 530 531 532 533
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
			leftmost = 0;
		}
	}

	/*
	 * Maintain a cache of leftmost tree entries (it is frequently
	 * used):
	 */
534
	if (leftmost)
I
Ingo Molnar 已提交
535
		cfs_rq->rb_leftmost = &se->run_node;
536 537 538 539 540

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

541
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
542
{
P
Peter Zijlstra 已提交
543 544 545 546 547 548
	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
549

550 551 552
	rb_erase(&se->run_node, &cfs_rq->tasks_timeline);
}

553
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
554
{
555 556 557 558 559 560
	struct rb_node *left = cfs_rq->rb_leftmost;

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
561 562
}

563 564 565 566 567 568 569 570 571 572 573
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
574
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
575
{
576
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline);
577

578 579
	if (!last)
		return NULL;
580 581

	return rb_entry(last, struct sched_entity, run_node);
582 583
}

584 585 586 587
/**************************************************************
 * Scheduling class statistics methods:
 */

588
int sched_proc_update_handler(struct ctl_table *table, int write,
589
		void __user *buffer, size_t *lenp,
590 591
		loff_t *ppos)
{
592
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
593
	int factor = get_update_sysctl_factor();
594 595 596 597 598 599 600

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

601 602 603 604 605 606 607
#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

608 609 610
	return 0;
}
#endif
611

612
/*
613
 * delta /= w
614 615 616 617
 */
static inline unsigned long
calc_delta_fair(unsigned long delta, struct sched_entity *se)
{
618 619
	if (unlikely(se->load.weight != NICE_0_LOAD))
		delta = calc_delta_mine(delta, NICE_0_LOAD, &se->load);
620 621 622 623

	return delta;
}

624 625 626
/*
 * The idea is to set a period in which each task runs once.
 *
627
 * When there are too many tasks (sched_nr_latency) we have to stretch
628 629 630 631
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
632 633 634
static u64 __sched_period(unsigned long nr_running)
{
	u64 period = sysctl_sched_latency;
635
	unsigned long nr_latency = sched_nr_latency;
636 637

	if (unlikely(nr_running > nr_latency)) {
638
		period = sysctl_sched_min_granularity;
639 640 641 642 643 644
		period *= nr_running;
	}

	return period;
}

645 646 647 648
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
649
 * s = p*P[w/rw]
650
 */
P
Peter Zijlstra 已提交
651
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
652
{
M
Mike Galbraith 已提交
653
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
654

M
Mike Galbraith 已提交
655
	for_each_sched_entity(se) {
L
Lin Ming 已提交
656
		struct load_weight *load;
657
		struct load_weight lw;
L
Lin Ming 已提交
658 659 660

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
661

M
Mike Galbraith 已提交
662
		if (unlikely(!se->on_rq)) {
663
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
664 665 666 667 668 669 670

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
		slice = calc_delta_mine(slice, se->load.weight, load);
	}
	return slice;
671 672
}

673
/*
A
Andrei Epure 已提交
674
 * We calculate the vruntime slice of a to-be-inserted task.
675
 *
676
 * vs = s/w
677
 */
678
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
679
{
680
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
681 682
}

683
#ifdef CONFIG_SMP
684 685
static unsigned long task_h_load(struct task_struct *p);

686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704
static inline void __update_task_entity_contrib(struct sched_entity *se);

/* Give new task start runnable values to heavy its load in infant time */
void init_task_runnable_average(struct task_struct *p)
{
	u32 slice;

	p->se.avg.decay_count = 0;
	slice = sched_slice(task_cfs_rq(p), &p->se) >> 10;
	p->se.avg.runnable_avg_sum = slice;
	p->se.avg.runnable_avg_period = slice;
	__update_task_entity_contrib(&p->se);
}
#else
void init_task_runnable_average(struct task_struct *p)
{
}
#endif

705 706 707 708 709
/*
 * Update the current task's runtime statistics. Skip current tasks that
 * are not in our scheduling class.
 */
static inline void
I
Ingo Molnar 已提交
710 711
__update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
	      unsigned long delta_exec)
712
{
713
	unsigned long delta_exec_weighted;
714

715 716
	schedstat_set(curr->statistics.exec_max,
		      max((u64)delta_exec, curr->statistics.exec_max));
717 718

	curr->sum_exec_runtime += delta_exec;
719
	schedstat_add(cfs_rq, exec_clock, delta_exec);
720
	delta_exec_weighted = calc_delta_fair(delta_exec, curr);
721

I
Ingo Molnar 已提交
722
	curr->vruntime += delta_exec_weighted;
723
	update_min_vruntime(cfs_rq);
724 725
}

726
static void update_curr(struct cfs_rq *cfs_rq)
727
{
728
	struct sched_entity *curr = cfs_rq->curr;
729
	u64 now = rq_clock_task(rq_of(cfs_rq));
730 731 732 733 734 735 736 737 738 739
	unsigned long delta_exec;

	if (unlikely(!curr))
		return;

	/*
	 * Get the amount of time the current task was running
	 * since the last time we changed load (this cannot
	 * overflow on 32 bits):
	 */
I
Ingo Molnar 已提交
740
	delta_exec = (unsigned long)(now - curr->exec_start);
P
Peter Zijlstra 已提交
741 742
	if (!delta_exec)
		return;
743

I
Ingo Molnar 已提交
744 745
	__update_curr(cfs_rq, curr, delta_exec);
	curr->exec_start = now;
746 747 748 749

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

750
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
751
		cpuacct_charge(curtask, delta_exec);
752
		account_group_exec_runtime(curtask, delta_exec);
753
	}
754 755

	account_cfs_rq_runtime(cfs_rq, delta_exec);
756 757 758
}

static inline void
759
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
760
{
761
	schedstat_set(se->statistics.wait_start, rq_clock(rq_of(cfs_rq)));
762 763 764 765 766
}

/*
 * Task is being enqueued - update stats:
 */
767
static void update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
768 769 770 771 772
{
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
773
	if (se != cfs_rq->curr)
774
		update_stats_wait_start(cfs_rq, se);
775 776 777
}

static void
778
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
779
{
780
	schedstat_set(se->statistics.wait_max, max(se->statistics.wait_max,
781
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start));
782 783
	schedstat_set(se->statistics.wait_count, se->statistics.wait_count + 1);
	schedstat_set(se->statistics.wait_sum, se->statistics.wait_sum +
784
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
785 786 787
#ifdef CONFIG_SCHEDSTATS
	if (entity_is_task(se)) {
		trace_sched_stat_wait(task_of(se),
788
			rq_clock(rq_of(cfs_rq)) - se->statistics.wait_start);
789 790
	}
#endif
791
	schedstat_set(se->statistics.wait_start, 0);
792 793 794
}

static inline void
795
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
796 797 798 799 800
{
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
801
	if (se != cfs_rq->curr)
802
		update_stats_wait_end(cfs_rq, se);
803 804 805 806 807 808
}

/*
 * We are picking a new current task - update its stats:
 */
static inline void
809
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
810 811 812 813
{
	/*
	 * We are starting a new run period:
	 */
814
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
815 816 817 818 819 820
}

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

821 822
#ifdef CONFIG_NUMA_BALANCING
/*
823 824 825
 * 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.
826
 */
827 828 829
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
unsigned int sysctl_numa_balancing_scan_period_reset = 60000;
830 831 832

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
833

834 835 836
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
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)
{
	unsigned int scan, floor;
	unsigned int windows = 1;

	if (sysctl_numa_balancing_scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / sysctl_numa_balancing_scan_size;
	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);
}

882 883 884 885 886 887 888
/*
 * Once a preferred node is selected the scheduler balancer will prefer moving
 * a task to that node for sysctl_numa_balancing_settle_count number of PTE
 * scans. This will give the process the chance to accumulate more faults on
 * the preferred node but still allow the scheduler to move the task again if
 * the nodes CPUs are overloaded.
 */
889
unsigned int sysctl_numa_balancing_settle_count __read_mostly = 4;
890

891 892 893 894 895 896 897 898 899 900 901 902
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));
}

903 904 905 906 907
struct numa_group {
	atomic_t refcount;

	spinlock_t lock; /* nr_tasks, tasks */
	int nr_tasks;
908
	pid_t gid;
909 910 911
	struct list_head task_list;

	struct rcu_head rcu;
912
	atomic_long_t total_faults;
913 914 915
	atomic_long_t faults[0];
};

916 917 918 919 920
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

921 922 923 924 925 926 927 928 929 930 931 932 933 934
static inline int task_faults_idx(int nid, int priv)
{
	return 2 * nid + priv;
}

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

	return p->numa_faults[task_faults_idx(nid, 0)] +
		p->numa_faults[task_faults_idx(nid, 1)];
}

935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

	return atomic_long_read(&p->numa_group->faults[2*nid]) +
	       atomic_long_read(&p->numa_group->faults[2*nid+1]);
}

/*
 * 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.
 */
static inline unsigned long task_weight(struct task_struct *p, int nid)
{
	unsigned long total_faults;

	if (!p->numa_faults)
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

	return 1000 * task_faults(p, nid) / total_faults;
}

static inline unsigned long group_weight(struct task_struct *p, int nid)
{
	unsigned long total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = atomic_long_read(&p->numa_group->total_faults);

	if (!total_faults)
		return 0;

977
	return 1000 * group_faults(p, nid) / total_faults;
978 979
}

980
static unsigned long weighted_cpuload(const int cpu);
981 982 983 984 985
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
static unsigned long power_of(int cpu);
static long effective_load(struct task_group *tg, int cpu, long wl, long wg);

986
/* Cached statistics for all CPUs within a node */
987
struct numa_stats {
988
	unsigned long nr_running;
989
	unsigned long load;
990 991 992 993 994 995 996

	/* Total compute capacity of CPUs on a node */
	unsigned long power;

	/* Approximate capacity in terms of runnable tasks on a node */
	unsigned long capacity;
	int has_capacity;
997
};
998

999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
	int cpu;

	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);
		ns->power += power_of(cpu);
	}

	ns->load = (ns->load * SCHED_POWER_SCALE) / ns->power;
	ns->capacity = DIV_ROUND_CLOSEST(ns->power, SCHED_POWER_SCALE);
	ns->has_capacity = (ns->nr_running < ns->capacity);
}

1020 1021
struct task_numa_env {
	struct task_struct *p;
1022

1023 1024
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1025

1026
	struct numa_stats src_stats, dst_stats;
1027

1028 1029 1030 1031
	int imbalance_pct, idx;

	struct task_struct *best_task;
	long best_imp;
1032 1033 1034
	int best_cpu;
};

1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
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);
	if (p)
		get_task_struct(p);

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

/*
 * 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
 */
1054 1055
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1056 1057 1058 1059 1060 1061
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
	long dst_load, src_load;
	long load;
1062
	long imp = (groupimp > 0) ? groupimp : taskimp;
1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080

	rcu_read_lock();
	cur = ACCESS_ONCE(dst_rq->curr);
	if (cur->pid == 0) /* idle */
		cur = NULL;

	/*
	 * "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;

1081 1082
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1083
		 * in any group then look only at task weights.
1084
		 */
1085
		if (cur->numa_group == env->p->numa_group) {
1086 1087
			imp = taskimp + task_weight(cur, env->src_nid) -
			      task_weight(cur, env->dst_nid);
1088 1089 1090 1091 1092 1093
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1094
		} else {
1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110
			/*
			 * Compare the group weights. If a task is all by
			 * itself (not part of a group), use the task weight
			 * instead.
			 */
			if (env->p->numa_group)
				imp = groupimp;
			else
				imp = taskimp;

			if (cur->numa_group)
				imp += group_weight(cur, env->src_nid) -
				       group_weight(cur, env->dst_nid);
			else
				imp += task_weight(cur, env->src_nid) -
				       task_weight(cur, env->dst_nid);
1111
		}
1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160
	}

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

	if (!cur) {
		/* Is there capacity at our destination? */
		if (env->src_stats.has_capacity &&
		    !env->dst_stats.has_capacity)
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
	if (src_rq->nr_running == 1 && dst_rq->nr_running == 1)
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
	dst_load = env->dst_stats.load;
	src_load = env->src_stats.load;

	/* XXX missing power terms */
	load = task_h_load(env->p);
	dst_load += load;
	src_load -= load;

	if (cur) {
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
	}

	/* make src_load the smaller */
	if (dst_load < src_load)
		swap(dst_load, src_load);

	if (src_load * env->imbalance_pct < dst_load * 100)
		goto unlock;

assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1161 1162
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1163 1164 1165 1166 1167 1168 1169 1170 1171
{
	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;
1172
		task_numa_compare(env, taskimp, groupimp);
1173 1174 1175
	}
}

1176 1177 1178 1179
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1180

1181
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1182
		.src_nid = task_node(p),
1183 1184 1185 1186 1187 1188

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
		.best_cpu = -1
1189 1190
	};
	struct sched_domain *sd;
1191
	unsigned long taskweight, groupweight;
1192
	int nid, ret;
1193
	long taskimp, groupimp;
1194

1195
	/*
1196 1197 1198 1199 1200 1201
	 * 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.
1202 1203
	 */
	rcu_read_lock();
1204 1205
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
	env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1206 1207
	rcu_read_unlock();

1208 1209
	taskweight = task_weight(p, env.src_nid);
	groupweight = group_weight(p, env.src_nid);
1210
	update_numa_stats(&env.src_stats, env.src_nid);
1211
	env.dst_nid = p->numa_preferred_nid;
1212 1213
	taskimp = task_weight(p, env.dst_nid) - taskweight;
	groupimp = group_weight(p, env.dst_nid) - groupweight;
1214
	update_numa_stats(&env.dst_stats, env.dst_nid);
1215

1216 1217
	/* If the preferred nid has capacity, try to use it. */
	if (env.dst_stats.has_capacity)
1218
		task_numa_find_cpu(&env, taskimp, groupimp);
1219 1220 1221

	/* No space available on the preferred nid. Look elsewhere. */
	if (env.best_cpu == -1) {
1222 1223 1224
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1225

1226
			/* Only consider nodes where both task and groups benefit */
1227 1228 1229
			taskimp = task_weight(p, nid) - taskweight;
			groupimp = group_weight(p, nid) - groupweight;
			if (taskimp < 0 && groupimp < 0)
1230 1231
				continue;

1232 1233
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1234
			task_numa_find_cpu(&env, taskimp, groupimp);
1235 1236 1237
		}
	}

1238 1239 1240 1241
	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;

1242 1243
	sched_setnuma(p, env.dst_nid);

1244 1245 1246 1247 1248 1249 1250 1251
	if (env.best_task == NULL) {
		int ret = migrate_task_to(p, env.best_cpu);
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
	put_task_struct(env.best_task);
	return ret;
1252 1253
}

1254 1255 1256 1257 1258
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
	/* Success if task is already running on preferred CPU */
	p->numa_migrate_retry = 0;
1259 1260 1261 1262 1263 1264 1265 1266
	if (cpu_to_node(task_cpu(p)) == p->numa_preferred_nid) {
		/*
		 * If migration is temporarily disabled due to a task migration
		 * then re-enable it now as the task is running on its
		 * preferred node and memory should migrate locally
		 */
		if (!p->numa_migrate_seq)
			p->numa_migrate_seq++;
1267
		return;
1268
	}
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278

	/* This task has no NUMA fault statistics yet */
	if (unlikely(p->numa_preferred_nid == -1))
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
	if (task_numa_migrate(p) != 0)
		p->numa_migrate_retry = jiffies + HZ*5;
}

1279 1280
static void task_numa_placement(struct task_struct *p)
{
1281 1282
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
1283
	spinlock_t *group_lock = NULL;
1284

1285
	seq = ACCESS_ONCE(p->mm->numa_scan_seq);
1286 1287 1288
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
1289
	p->numa_migrate_seq++;
1290
	p->numa_scan_period_max = task_scan_max(p);
1291

1292 1293 1294 1295 1296 1297
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
		spin_lock(group_lock);
	}

1298 1299
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
1300
		unsigned long faults = 0, group_faults = 0;
1301
		int priv, i;
1302

1303
		for (priv = 0; priv < 2; priv++) {
1304 1305
			long diff;

1306
			i = task_faults_idx(nid, priv);
1307
			diff = -p->numa_faults[i];
1308

1309 1310 1311 1312
			/* Decay existing window, copy faults since last scan */
			p->numa_faults[i] >>= 1;
			p->numa_faults[i] += p->numa_faults_buffer[i];
			p->numa_faults_buffer[i] = 0;
1313 1314

			faults += p->numa_faults[i];
1315
			diff += p->numa_faults[i];
1316
			p->total_numa_faults += diff;
1317 1318 1319
			if (p->numa_group) {
				/* safe because we can only change our own group */
				atomic_long_add(diff, &p->numa_group->faults[i]);
1320 1321
				atomic_long_add(diff, &p->numa_group->total_faults);
				group_faults += atomic_long_read(&p->numa_group->faults[i]);
1322
			}
1323 1324
		}

1325 1326 1327 1328
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
1329 1330 1331 1332 1333 1334 1335

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

1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
	if (p->numa_group) {
		/*
		 * If the preferred task and group nids are different,
		 * iterate over the nodes again to find the best place.
		 */
		if (max_nid != max_group_nid) {
			unsigned long weight, max_weight = 0;

			for_each_online_node(nid) {
				weight = task_weight(p, nid) + group_weight(p, nid);
				if (weight > max_weight) {
					max_weight = weight;
					max_nid = nid;
				}
1350 1351
			}
		}
1352 1353

		spin_unlock(group_lock);
1354 1355
	}

1356
	/* Preferred node as the node with the most faults */
1357
	if (max_faults && max_nid != p->numa_preferred_nid) {
1358
		/* Update the preferred nid and migrate task if possible */
1359
		sched_setnuma(p, max_nid);
1360
		numa_migrate_preferred(p);
1361
	}
1362 1363
}

1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
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);
}

static void double_lock(spinlock_t *l1, spinlock_t *l2)
{
	if (l1 > l2)
		swap(l1, l2);

	spin_lock(l1);
	spin_lock_nested(l2, SINGLE_DEPTH_NESTING);
}

1384 1385
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
{
	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) +
				    2*nr_node_ids*sizeof(atomic_long_t);

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

		atomic_set(&grp->refcount, 1);
		spin_lock_init(&grp->lock);
		INIT_LIST_HEAD(&grp->task_list);
1404
		grp->gid = p->pid;
1405 1406 1407 1408

		for (i = 0; i < 2*nr_node_ids; i++)
			atomic_long_set(&grp->faults[i], p->numa_faults[i]);

1409 1410
		atomic_long_set(&grp->total_faults, p->total_numa_faults);

1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442
		list_add(&p->numa_entry, &grp->task_list);
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
	tsk = ACCESS_ONCE(cpu_rq(cpu)->curr);

	if (!cpupid_match_pid(tsk, cpupid))
		goto unlock;

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
		goto unlock;

	my_grp = p->numa_group;
	if (grp == my_grp)
		goto unlock;

	/*
	 * 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)
		goto unlock;

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

1443 1444 1445 1446 1447 1448 1449
	/* 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;
1450

1451 1452 1453
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

1454 1455
	if (join && !get_numa_group(grp))
		join = false;
1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466

unlock:
	rcu_read_unlock();

	if (!join)
		return;

	for (i = 0; i < 2*nr_node_ids; i++) {
		atomic_long_sub(p->numa_faults[i], &my_grp->faults[i]);
		atomic_long_add(p->numa_faults[i], &grp->faults[i]);
	}
1467 1468
	atomic_long_sub(p->total_numa_faults, &my_grp->total_faults);
	atomic_long_add(p->total_numa_faults, &grp->total_faults);
1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487

	double_lock(&my_grp->lock, &grp->lock);

	list_move(&p->numa_entry, &grp->task_list);
	my_grp->nr_tasks--;
	grp->nr_tasks++;

	spin_unlock(&my_grp->lock);
	spin_unlock(&grp->lock);

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
	int i;
1488
	void *numa_faults = p->numa_faults;
1489 1490 1491 1492 1493

	if (grp) {
		for (i = 0; i < 2*nr_node_ids; i++)
			atomic_long_sub(p->numa_faults[i], &grp->faults[i]);

1494 1495
		atomic_long_sub(p->total_numa_faults, &grp->total_faults);

1496 1497 1498 1499 1500 1501 1502 1503
		spin_lock(&grp->lock);
		list_del(&p->numa_entry);
		grp->nr_tasks--;
		spin_unlock(&grp->lock);
		rcu_assign_pointer(p->numa_group, NULL);
		put_numa_group(grp);
	}

1504 1505 1506
	p->numa_faults = NULL;
	p->numa_faults_buffer = NULL;
	kfree(numa_faults);
1507 1508
}

1509 1510 1511
/*
 * Got a PROT_NONE fault for a page on @node.
 */
1512
void task_numa_fault(int last_cpupid, int node, int pages, int flags)
1513 1514
{
	struct task_struct *p = current;
1515
	bool migrated = flags & TNF_MIGRATED;
1516
	int priv;
1517

1518
	if (!numabalancing_enabled)
1519 1520
		return;

1521 1522 1523 1524
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

1525 1526 1527 1528
	/* Do not worry about placement if exiting */
	if (p->state == TASK_DEAD)
		return;

1529 1530
	/* Allocate buffer to track faults on a per-node basis */
	if (unlikely(!p->numa_faults)) {
1531
		int size = sizeof(*p->numa_faults) * 2 * nr_node_ids;
1532

1533 1534
		/* numa_faults and numa_faults_buffer share the allocation */
		p->numa_faults = kzalloc(size * 2, GFP_KERNEL|__GFP_NOWARN);
1535 1536
		if (!p->numa_faults)
			return;
1537 1538

		BUG_ON(p->numa_faults_buffer);
1539
		p->numa_faults_buffer = p->numa_faults + (2 * nr_node_ids);
1540
		p->total_numa_faults = 0;
1541
	}
1542

1543 1544 1545 1546 1547 1548 1549 1550
	/*
	 * 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);
1551
		if (!priv && !(flags & TNF_NO_GROUP))
1552
			task_numa_group(p, last_cpupid, flags, &priv);
1553 1554
	}

1555
	/*
1556 1557
	 * If pages are properly placed (did not migrate) then scan slower.
	 * This is reset periodically in case of phase changes
1558
	 */
1559 1560 1561 1562 1563 1564 1565 1566
	if (!migrated) {
		/* Initialise if necessary */
		if (!p->numa_scan_period_max)
			p->numa_scan_period_max = task_scan_max(p);

		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period + 10);
	}
1567

1568
	task_numa_placement(p);
1569

1570 1571 1572 1573
	/* Retry task to preferred node migration if it previously failed */
	if (p->numa_migrate_retry && time_after(jiffies, p->numa_migrate_retry))
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
1574 1575 1576
	if (migrated)
		p->numa_pages_migrated += pages;

1577
	p->numa_faults_buffer[task_faults_idx(node, priv)] += pages;
1578 1579
}

1580 1581 1582 1583 1584 1585
static void reset_ptenuma_scan(struct task_struct *p)
{
	ACCESS_ONCE(p->mm->numa_scan_seq)++;
	p->mm->numa_scan_offset = 0;
}

1586 1587 1588 1589 1590 1591 1592 1593 1594
/*
 * 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;
1595
	struct vm_area_struct *vma;
1596
	unsigned long start, end;
1597
	unsigned long nr_pte_updates = 0;
1598
	long pages;
1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613

	WARN_ON_ONCE(p != container_of(work, struct task_struct, numa_work));

	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;

1614 1615 1616 1617 1618 1619 1620
	if (!mm->numa_next_reset || !mm->numa_next_scan) {
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
		mm->numa_next_reset = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_period_reset);
	}

1621 1622 1623 1624 1625 1626 1627 1628
	/*
	 * Reset the scan period if enough time has gone by. Objective is that
	 * scanning will be reduced if pages are properly placed. As tasks
	 * can enter different phases this needs to be re-examined. Lacking
	 * proper tracking of reference behaviour, this blunt hammer is used.
	 */
	migrate = mm->numa_next_reset;
	if (time_after(now, migrate)) {
1629
		p->numa_scan_period = task_scan_min(p);
1630 1631 1632 1633
		next_scan = now + msecs_to_jiffies(sysctl_numa_balancing_scan_period_reset);
		xchg(&mm->numa_next_reset, next_scan);
	}

1634 1635 1636 1637 1638 1639 1640
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

1641 1642 1643 1644
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
		p->numa_scan_period = task_scan_min(p);
	}
1645

1646
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
1647 1648 1649
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

1650 1651 1652 1653 1654 1655
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

1656 1657 1658 1659 1660
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
	if (!pages)
		return;
1661

1662
	down_read(&mm->mmap_sem);
1663
	vma = find_vma(mm, start);
1664 1665
	if (!vma) {
		reset_ptenuma_scan(p);
1666
		start = 0;
1667 1668
		vma = mm->mmap;
	}
1669
	for (; vma; vma = vma->vm_next) {
1670
		if (!vma_migratable(vma) || !vma_policy_mof(p, vma))
1671 1672
			continue;

1673 1674 1675 1676 1677 1678 1679 1680 1681 1682
		/*
		 * 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;

1683 1684 1685 1686
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
1687 1688 1689 1690 1691 1692 1693 1694 1695
			nr_pte_updates += change_prot_numa(vma, start, end);

			/*
			 * Scan sysctl_numa_balancing_scan_size but ensure that
			 * at least one PTE is updated so that unused virtual
			 * address space is quickly skipped.
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
1696

1697 1698 1699 1700
			start = end;
			if (pages <= 0)
				goto out;
		} while (end != vma->vm_end);
1701
	}
1702

1703
out:
1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715
	/*
	 * If the whole process was scanned without updates then no NUMA
	 * hinting faults are being recorded and scan rate should be lower.
	 */
	if (mm->numa_scan_offset == 0 && !nr_pte_updates) {
		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period << 1);

		next_scan = now + msecs_to_jiffies(p->numa_scan_period);
		mm->numa_next_scan = next_scan;
	}

1716
	/*
P
Peter Zijlstra 已提交
1717 1718 1719 1720
	 * 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.
1721 1722
	 */
	if (vma)
1723
		mm->numa_scan_offset = start;
1724 1725 1726
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752
}

/*
 * 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;

	if (now - curr->node_stamp > period) {
1753
		if (!curr->node_stamp)
1754
			curr->numa_scan_period = task_scan_min(curr);
1755
		curr->node_stamp += period;
1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766

		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)
{
}
1767 1768 1769 1770 1771 1772 1773 1774

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)
{
}
1775 1776
#endif /* CONFIG_NUMA_BALANCING */

1777 1778 1779 1780
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
1781
	if (!parent_entity(se))
1782
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
1783
#ifdef CONFIG_SMP
1784 1785 1786 1787 1788 1789
	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);
	}
1790
#endif
1791 1792 1793 1794 1795 1796 1797
	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);
1798
	if (!parent_entity(se))
1799
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
1800 1801
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
1802
		list_del_init(&se->group_node);
1803
	}
1804 1805 1806
	cfs_rq->nr_running--;
}

1807 1808
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
1809 1810 1811 1812 1813 1814 1815 1816 1817
static inline long calc_tg_weight(struct task_group *tg, struct cfs_rq *cfs_rq)
{
	long tg_weight;

	/*
	 * Use this CPU's actual weight instead of the last load_contribution
	 * to gain a more accurate current total weight. See
	 * update_cfs_rq_load_contribution().
	 */
1818
	tg_weight = atomic_long_read(&tg->load_avg);
1819
	tg_weight -= cfs_rq->tg_load_contrib;
1820 1821 1822 1823 1824
	tg_weight += cfs_rq->load.weight;

	return tg_weight;
}

1825
static long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
1826
{
1827
	long tg_weight, load, shares;
1828

1829
	tg_weight = calc_tg_weight(tg, cfs_rq);
1830
	load = cfs_rq->load.weight;
1831 1832

	shares = (tg->shares * load);
1833 1834
	if (tg_weight)
		shares /= tg_weight;
1835 1836 1837 1838 1839 1840 1841 1842 1843

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

	return shares;
}
# else /* CONFIG_SMP */
1844
static inline long calc_cfs_shares(struct cfs_rq *cfs_rq, struct task_group *tg)
1845 1846 1847 1848
{
	return tg->shares;
}
# endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
1849 1850 1851
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
			    unsigned long weight)
{
1852 1853 1854 1855
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
P
Peter Zijlstra 已提交
1856
		account_entity_dequeue(cfs_rq, se);
1857
	}
P
Peter Zijlstra 已提交
1858 1859 1860 1861 1862 1863 1864

	update_load_set(&se->load, weight);

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

1865 1866
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

1867
static void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
1868 1869 1870
{
	struct task_group *tg;
	struct sched_entity *se;
1871
	long shares;
P
Peter Zijlstra 已提交
1872 1873 1874

	tg = cfs_rq->tg;
	se = tg->se[cpu_of(rq_of(cfs_rq))];
1875
	if (!se || throttled_hierarchy(cfs_rq))
P
Peter Zijlstra 已提交
1876
		return;
1877 1878 1879 1880
#ifndef CONFIG_SMP
	if (likely(se->load.weight == tg->shares))
		return;
#endif
1881
	shares = calc_cfs_shares(cfs_rq, tg);
P
Peter Zijlstra 已提交
1882 1883 1884 1885

	reweight_entity(cfs_rq_of(se), se, shares);
}
#else /* CONFIG_FAIR_GROUP_SCHED */
1886
static inline void update_cfs_shares(struct cfs_rq *cfs_rq)
P
Peter Zijlstra 已提交
1887 1888 1889 1890
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

1891
#ifdef CONFIG_SMP
1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919
/*
 * We choose a half-life close to 1 scheduling period.
 * Note: The tables below are dependent on this value.
 */
#define LOAD_AVG_PERIOD 32
#define LOAD_AVG_MAX 47742 /* maximum possible load avg */
#define LOAD_AVG_MAX_N 345 /* number of full periods to produce LOAD_MAX_AVG */

/* 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,
};

1920 1921 1922 1923 1924 1925
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
static __always_inline u64 decay_load(u64 val, u64 n)
{
1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945
	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
	 *    y^n = 1/2^(n/PERIOD) * k^(n%PERIOD)
	 * With a look-up table which covers k^n (n<PERIOD)
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
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
	val *= runnable_avg_yN_inv[local_n];
	/* We don't use SRR here since we always want to round down. */
	return val >> 32;
}

/*
 * 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;

	/* Compute \Sum k^n combining precomputed values for k^i, \Sum k^j */
	do {
		contrib /= 2; /* y^LOAD_AVG_PERIOD = 1/2 */
		contrib += runnable_avg_yN_sum[LOAD_AVG_PERIOD];

		n -= LOAD_AVG_PERIOD;
	} while (n > LOAD_AVG_PERIOD);

	contrib = decay_load(contrib, n);
	return contrib + runnable_avg_yN_sum[n];
1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
}

/*
 * 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}]
 */
static __always_inline int __update_entity_runnable_avg(u64 now,
							struct sched_avg *sa,
							int runnable)
{
2013 2014
	u64 delta, periods;
	u32 runnable_contrib;
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047
	int delta_w, decayed = 0;

	delta = now - sa->last_runnable_update;
	/*
	 * 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) {
		sa->last_runnable_update = now;
		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;
	sa->last_runnable_update = now;

	/* delta_w is the amount already accumulated against our next period */
	delta_w = sa->runnable_avg_period % 1024;
	if (delta + delta_w >= 1024) {
		/* period roll-over */
		decayed = 1;

		/*
		 * 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;
2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067
		if (runnable)
			sa->runnable_avg_sum += delta_w;
		sa->runnable_avg_period += delta_w;

		delta -= delta_w;

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

		sa->runnable_avg_sum = decay_load(sa->runnable_avg_sum,
						  periods + 1);
		sa->runnable_avg_period = decay_load(sa->runnable_avg_period,
						     periods + 1);

		/* Efficiently calculate \sum (1..n_period) 1024*y^i */
		runnable_contrib = __compute_runnable_contrib(periods);
		if (runnable)
			sa->runnable_avg_sum += runnable_contrib;
		sa->runnable_avg_period += runnable_contrib;
2068 2069 2070 2071 2072 2073 2074 2075 2076 2077
	}

	/* Remainder of delta accrued against u_0` */
	if (runnable)
		sa->runnable_avg_sum += delta;
	sa->runnable_avg_period += delta;

	return decayed;
}

2078
/* Synchronize an entity's decay with its parenting cfs_rq.*/
2079
static inline u64 __synchronize_entity_decay(struct sched_entity *se)
2080 2081 2082 2083 2084 2085
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 decays = atomic64_read(&cfs_rq->decay_counter);

	decays -= se->avg.decay_count;
	if (!decays)
2086
		return 0;
2087 2088 2089

	se->avg.load_avg_contrib = decay_load(se->avg.load_avg_contrib, decays);
	se->avg.decay_count = 0;
2090 2091

	return decays;
2092 2093
}

2094 2095 2096 2097 2098
#ifdef CONFIG_FAIR_GROUP_SCHED
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update)
{
	struct task_group *tg = cfs_rq->tg;
2099
	long tg_contrib;
2100 2101 2102 2103

	tg_contrib = cfs_rq->runnable_load_avg + cfs_rq->blocked_load_avg;
	tg_contrib -= cfs_rq->tg_load_contrib;

2104 2105
	if (force_update || abs(tg_contrib) > cfs_rq->tg_load_contrib / 8) {
		atomic_long_add(tg_contrib, &tg->load_avg);
2106 2107 2108
		cfs_rq->tg_load_contrib += tg_contrib;
	}
}
2109

2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130
/*
 * Aggregate cfs_rq runnable averages into an equivalent task_group
 * representation for computing load contributions.
 */
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq)
{
	struct task_group *tg = cfs_rq->tg;
	long contrib;

	/* The fraction of a cpu used by this cfs_rq */
	contrib = div_u64(sa->runnable_avg_sum << NICE_0_SHIFT,
			  sa->runnable_avg_period + 1);
	contrib -= cfs_rq->tg_runnable_contrib;

	if (abs(contrib) > cfs_rq->tg_runnable_contrib / 64) {
		atomic_add(contrib, &tg->runnable_avg);
		cfs_rq->tg_runnable_contrib += contrib;
	}
}

2131 2132 2133 2134
static inline void __update_group_entity_contrib(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = group_cfs_rq(se);
	struct task_group *tg = cfs_rq->tg;
2135 2136
	int runnable_avg;

2137 2138 2139
	u64 contrib;

	contrib = cfs_rq->tg_load_contrib * tg->shares;
2140 2141
	se->avg.load_avg_contrib = div_u64(contrib,
				     atomic_long_read(&tg->load_avg) + 1);
2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170

	/*
	 * For group entities we need to compute a correction term in the case
	 * that they are consuming <1 cpu so that we would contribute the same
	 * load as a task of equal weight.
	 *
	 * Explicitly co-ordinating this measurement would be expensive, but
	 * fortunately the sum of each cpus contribution forms a usable
	 * lower-bound on the true value.
	 *
	 * Consider the aggregate of 2 contributions.  Either they are disjoint
	 * (and the sum represents true value) or they are disjoint and we are
	 * understating by the aggregate of their overlap.
	 *
	 * Extending this to N cpus, for a given overlap, the maximum amount we
	 * understand is then n_i(n_i+1)/2 * w_i where n_i is the number of
	 * cpus that overlap for this interval and w_i is the interval width.
	 *
	 * On a small machine; the first term is well-bounded which bounds the
	 * total error since w_i is a subset of the period.  Whereas on a
	 * larger machine, while this first term can be larger, if w_i is the
	 * of consequential size guaranteed to see n_i*w_i quickly converge to
	 * our upper bound of 1-cpu.
	 */
	runnable_avg = atomic_read(&tg->runnable_avg);
	if (runnable_avg < NICE_0_LOAD) {
		se->avg.load_avg_contrib *= runnable_avg;
		se->avg.load_avg_contrib >>= NICE_0_SHIFT;
	}
2171
}
2172 2173 2174
#else
static inline void __update_cfs_rq_tg_load_contrib(struct cfs_rq *cfs_rq,
						 int force_update) {}
2175 2176
static inline void __update_tg_runnable_avg(struct sched_avg *sa,
						  struct cfs_rq *cfs_rq) {}
2177
static inline void __update_group_entity_contrib(struct sched_entity *se) {}
2178 2179
#endif

2180 2181 2182 2183 2184 2185 2186 2187 2188 2189
static inline void __update_task_entity_contrib(struct sched_entity *se)
{
	u32 contrib;

	/* avoid overflowing a 32-bit type w/ SCHED_LOAD_SCALE */
	contrib = se->avg.runnable_avg_sum * scale_load_down(se->load.weight);
	contrib /= (se->avg.runnable_avg_period + 1);
	se->avg.load_avg_contrib = scale_load(contrib);
}

2190 2191 2192 2193 2194
/* Compute the current contribution to load_avg by se, return any delta */
static long __update_entity_load_avg_contrib(struct sched_entity *se)
{
	long old_contrib = se->avg.load_avg_contrib;

2195 2196 2197
	if (entity_is_task(se)) {
		__update_task_entity_contrib(se);
	} else {
2198
		__update_tg_runnable_avg(&se->avg, group_cfs_rq(se));
2199 2200
		__update_group_entity_contrib(se);
	}
2201 2202 2203 2204

	return se->avg.load_avg_contrib - old_contrib;
}

2205 2206 2207 2208 2209 2210 2211 2212 2213
static inline void subtract_blocked_load_contrib(struct cfs_rq *cfs_rq,
						 long load_contrib)
{
	if (likely(load_contrib < cfs_rq->blocked_load_avg))
		cfs_rq->blocked_load_avg -= load_contrib;
	else
		cfs_rq->blocked_load_avg = 0;
}

2214 2215
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);

2216
/* Update a sched_entity's runnable average */
2217 2218
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq)
2219
{
2220 2221
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	long contrib_delta;
2222
	u64 now;
2223

2224 2225 2226 2227 2228 2229 2230 2231 2232 2233
	/*
	 * For a group entity we need to use their owned cfs_rq_clock_task() in
	 * case they are the parent of a throttled hierarchy.
	 */
	if (entity_is_task(se))
		now = cfs_rq_clock_task(cfs_rq);
	else
		now = cfs_rq_clock_task(group_cfs_rq(se));

	if (!__update_entity_runnable_avg(now, &se->avg, se->on_rq))
2234 2235 2236
		return;

	contrib_delta = __update_entity_load_avg_contrib(se);
2237 2238 2239 2240

	if (!update_cfs_rq)
		return;

2241 2242
	if (se->on_rq)
		cfs_rq->runnable_load_avg += contrib_delta;
2243 2244 2245 2246 2247 2248 2249 2250
	else
		subtract_blocked_load_contrib(cfs_rq, -contrib_delta);
}

/*
 * Decay the load contributed by all blocked children and account this so that
 * their contribution may appropriately discounted when they wake up.
 */
2251
static void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq, int force_update)
2252
{
2253
	u64 now = cfs_rq_clock_task(cfs_rq) >> 20;
2254 2255 2256
	u64 decays;

	decays = now - cfs_rq->last_decay;
2257
	if (!decays && !force_update)
2258 2259
		return;

2260 2261 2262
	if (atomic_long_read(&cfs_rq->removed_load)) {
		unsigned long removed_load;
		removed_load = atomic_long_xchg(&cfs_rq->removed_load, 0);
2263 2264
		subtract_blocked_load_contrib(cfs_rq, removed_load);
	}
2265

2266 2267 2268 2269 2270 2271
	if (decays) {
		cfs_rq->blocked_load_avg = decay_load(cfs_rq->blocked_load_avg,
						      decays);
		atomic64_add(decays, &cfs_rq->decay_counter);
		cfs_rq->last_decay = now;
	}
2272 2273

	__update_cfs_rq_tg_load_contrib(cfs_rq, force_update);
2274
}
2275 2276 2277

static inline void update_rq_runnable_avg(struct rq *rq, int runnable)
{
2278
	__update_entity_runnable_avg(rq_clock_task(rq), &rq->avg, runnable);
2279
	__update_tg_runnable_avg(&rq->avg, &rq->cfs);
2280
}
2281 2282 2283

/* Add the load generated by se into cfs_rq's child load-average */
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2284 2285
						  struct sched_entity *se,
						  int wakeup)
2286
{
2287 2288 2289 2290
	/*
	 * We track migrations using entity decay_count <= 0, on a wake-up
	 * migration we use a negative decay count to track the remote decays
	 * accumulated while sleeping.
2291 2292 2293 2294
	 *
	 * Newly forked tasks are enqueued with se->avg.decay_count == 0, they
	 * are seen by enqueue_entity_load_avg() as a migration with an already
	 * constructed load_avg_contrib.
2295 2296
	 */
	if (unlikely(se->avg.decay_count <= 0)) {
2297
		se->avg.last_runnable_update = rq_clock_task(rq_of(cfs_rq));
2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312
		if (se->avg.decay_count) {
			/*
			 * In a wake-up migration we have to approximate the
			 * time sleeping.  This is because we can't synchronize
			 * clock_task between the two cpus, and it is not
			 * guaranteed to be read-safe.  Instead, we can
			 * approximate this using our carried decays, which are
			 * explicitly atomically readable.
			 */
			se->avg.last_runnable_update -= (-se->avg.decay_count)
							<< 20;
			update_entity_load_avg(se, 0);
			/* Indicate that we're now synchronized and on-rq */
			se->avg.decay_count = 0;
		}
2313 2314
		wakeup = 0;
	} else {
2315 2316 2317 2318 2319 2320 2321
		/*
		 * Task re-woke on same cpu (or else migrate_task_rq_fair()
		 * would have made count negative); we must be careful to avoid
		 * double-accounting blocked time after synchronizing decays.
		 */
		se->avg.last_runnable_update += __synchronize_entity_decay(se)
							<< 20;
2322 2323
	}

2324 2325
	/* migrated tasks did not contribute to our blocked load */
	if (wakeup) {
2326
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
2327 2328
		update_entity_load_avg(se, 0);
	}
2329

2330
	cfs_rq->runnable_load_avg += se->avg.load_avg_contrib;
2331 2332
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !wakeup);
2333 2334
}

2335 2336 2337 2338 2339
/*
 * Remove se's load from this cfs_rq child load-average, if the entity is
 * transitioning to a blocked state we track its projected decay using
 * blocked_load_avg.
 */
2340
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2341 2342
						  struct sched_entity *se,
						  int sleep)
2343
{
2344
	update_entity_load_avg(se, 1);
2345 2346
	/* we force update consideration on load-balancer moves */
	update_cfs_rq_blocked_load(cfs_rq, !sleep);
2347

2348
	cfs_rq->runnable_load_avg -= se->avg.load_avg_contrib;
2349 2350 2351 2352
	if (sleep) {
		cfs_rq->blocked_load_avg += se->avg.load_avg_contrib;
		se->avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
	} /* migrations, e.g. sleep=0 leave decay_count == 0 */
2353
}
2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374

/*
 * Update the rq's load with the elapsed running time before entering
 * idle. if the last scheduled task is not a CFS task, idle_enter will
 * be the only way to update the runnable statistic.
 */
void idle_enter_fair(struct rq *this_rq)
{
	update_rq_runnable_avg(this_rq, 1);
}

/*
 * Update the rq's load with the elapsed idle time before a task is
 * scheduled. if the newly scheduled task is not a CFS task, idle_exit will
 * be the only way to update the runnable statistic.
 */
void idle_exit_fair(struct rq *this_rq)
{
	update_rq_runnable_avg(this_rq, 0);
}

2375
#else
2376 2377
static inline void update_entity_load_avg(struct sched_entity *se,
					  int update_cfs_rq) {}
2378
static inline void update_rq_runnable_avg(struct rq *rq, int runnable) {}
2379
static inline void enqueue_entity_load_avg(struct cfs_rq *cfs_rq,
2380 2381
					   struct sched_entity *se,
					   int wakeup) {}
2382
static inline void dequeue_entity_load_avg(struct cfs_rq *cfs_rq,
2383 2384
					   struct sched_entity *se,
					   int sleep) {}
2385 2386
static inline void update_cfs_rq_blocked_load(struct cfs_rq *cfs_rq,
					      int force_update) {}
2387 2388
#endif

2389
static void enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
2390 2391
{
#ifdef CONFIG_SCHEDSTATS
2392 2393 2394 2395 2396
	struct task_struct *tsk = NULL;

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

2397
	if (se->statistics.sleep_start) {
2398
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.sleep_start;
2399 2400 2401 2402

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

2403 2404
		if (unlikely(delta > se->statistics.sleep_max))
			se->statistics.sleep_max = delta;
2405

2406
		se->statistics.sleep_start = 0;
2407
		se->statistics.sum_sleep_runtime += delta;
A
Arjan van de Ven 已提交
2408

2409
		if (tsk) {
2410
			account_scheduler_latency(tsk, delta >> 10, 1);
2411 2412
			trace_sched_stat_sleep(tsk, delta);
		}
2413
	}
2414
	if (se->statistics.block_start) {
2415
		u64 delta = rq_clock(rq_of(cfs_rq)) - se->statistics.block_start;
2416 2417 2418 2419

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

2420 2421
		if (unlikely(delta > se->statistics.block_max))
			se->statistics.block_max = delta;
2422

2423
		se->statistics.block_start = 0;
2424
		se->statistics.sum_sleep_runtime += delta;
I
Ingo Molnar 已提交
2425

2426
		if (tsk) {
2427
			if (tsk->in_iowait) {
2428 2429
				se->statistics.iowait_sum += delta;
				se->statistics.iowait_count++;
2430
				trace_sched_stat_iowait(tsk, delta);
2431 2432
			}

2433 2434
			trace_sched_stat_blocked(tsk, delta);

2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445
			/*
			 * 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);
I
Ingo Molnar 已提交
2446
		}
2447 2448 2449 2450
	}
#endif
}

P
Peter Zijlstra 已提交
2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463
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)
		schedstat_inc(cfs_rq, nr_spread_over);
#endif
}

2464 2465 2466
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
2467
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
2468

2469 2470 2471 2472 2473 2474
	/*
	 * 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
2475
	if (initial && sched_feat(START_DEBIT))
2476
		vruntime += sched_vslice(cfs_rq, se);
2477

2478
	/* sleeps up to a single latency don't count. */
2479
	if (!initial) {
2480
		unsigned long thresh = sysctl_sched_latency;
2481

2482 2483 2484 2485 2486 2487
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
2488

2489
		vruntime -= thresh;
2490 2491
	}

2492
	/* ensure we never gain time by being placed backwards. */
2493
	se->vruntime = max_vruntime(se->vruntime, vruntime);
2494 2495
}

2496 2497
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

2498
static void
2499
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2500
{
2501 2502
	/*
	 * Update the normalized vruntime before updating min_vruntime
2503
	 * through calling update_curr().
2504
	 */
2505
	if (!(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_WAKING))
2506 2507
		se->vruntime += cfs_rq->min_vruntime;

2508
	/*
2509
	 * Update run-time statistics of the 'current'.
2510
	 */
2511
	update_curr(cfs_rq);
2512
	enqueue_entity_load_avg(cfs_rq, se, flags & ENQUEUE_WAKEUP);
2513 2514
	account_entity_enqueue(cfs_rq, se);
	update_cfs_shares(cfs_rq);
2515

2516
	if (flags & ENQUEUE_WAKEUP) {
2517
		place_entity(cfs_rq, se, 0);
2518
		enqueue_sleeper(cfs_rq, se);
I
Ingo Molnar 已提交
2519
	}
2520

2521
	update_stats_enqueue(cfs_rq, se);
P
Peter Zijlstra 已提交
2522
	check_spread(cfs_rq, se);
2523 2524
	if (se != cfs_rq->curr)
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
2525
	se->on_rq = 1;
2526

2527
	if (cfs_rq->nr_running == 1) {
2528
		list_add_leaf_cfs_rq(cfs_rq);
2529 2530
		check_enqueue_throttle(cfs_rq);
	}
2531 2532
}

2533
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
2534
{
2535 2536 2537 2538 2539 2540 2541 2542
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		if (cfs_rq->last == se)
			cfs_rq->last = NULL;
		else
			break;
	}
}
P
Peter Zijlstra 已提交
2543

2544 2545 2546 2547 2548 2549 2550 2551 2552
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		if (cfs_rq->next == se)
			cfs_rq->next = NULL;
		else
			break;
	}
P
Peter Zijlstra 已提交
2553 2554
}

2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		if (cfs_rq->skip == se)
			cfs_rq->skip = NULL;
		else
			break;
	}
}

P
Peter Zijlstra 已提交
2566 2567
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2568 2569 2570 2571 2572
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
2573 2574 2575

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
2576 2577
}

2578
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
2579

2580
static void
2581
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
2582
{
2583 2584 2585 2586
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
2587
	dequeue_entity_load_avg(cfs_rq, se, flags & DEQUEUE_SLEEP);
2588

2589
	update_stats_dequeue(cfs_rq, se);
2590
	if (flags & DEQUEUE_SLEEP) {
P
Peter Zijlstra 已提交
2591
#ifdef CONFIG_SCHEDSTATS
2592 2593 2594 2595
		if (entity_is_task(se)) {
			struct task_struct *tsk = task_of(se);

			if (tsk->state & TASK_INTERRUPTIBLE)
2596
				se->statistics.sleep_start = rq_clock(rq_of(cfs_rq));
2597
			if (tsk->state & TASK_UNINTERRUPTIBLE)
2598
				se->statistics.block_start = rq_clock(rq_of(cfs_rq));
2599
		}
2600
#endif
P
Peter Zijlstra 已提交
2601 2602
	}

P
Peter Zijlstra 已提交
2603
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
2604

2605
	if (se != cfs_rq->curr)
2606
		__dequeue_entity(cfs_rq, se);
2607
	se->on_rq = 0;
2608
	account_entity_dequeue(cfs_rq, se);
2609 2610 2611 2612 2613 2614

	/*
	 * Normalize the entity after updating the min_vruntime because the
	 * update can refer to the ->curr item and we need to reflect this
	 * movement in our normalized position.
	 */
2615
	if (!(flags & DEQUEUE_SLEEP))
2616
		se->vruntime -= cfs_rq->min_vruntime;
2617

2618 2619 2620
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

2621
	update_min_vruntime(cfs_rq);
2622
	update_cfs_shares(cfs_rq);
2623 2624 2625 2626 2627
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
2628
static void
I
Ingo Molnar 已提交
2629
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
2630
{
2631
	unsigned long ideal_runtime, delta_exec;
2632 2633
	struct sched_entity *se;
	s64 delta;
2634

P
Peter Zijlstra 已提交
2635
	ideal_runtime = sched_slice(cfs_rq, curr);
2636
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
2637
	if (delta_exec > ideal_runtime) {
2638
		resched_task(rq_of(cfs_rq)->curr);
2639 2640 2641 2642 2643
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654
		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;

2655 2656
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
2657

2658 2659
	if (delta < 0)
		return;
2660

2661 2662
	if (delta > ideal_runtime)
		resched_task(rq_of(cfs_rq)->curr);
2663 2664
}

2665
static void
2666
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
2667
{
2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678
	/* '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.
		 */
		update_stats_wait_end(cfs_rq, se);
		__dequeue_entity(cfs_rq, se);
	}

2679
	update_stats_curr_start(cfs_rq, se);
2680
	cfs_rq->curr = se;
I
Ingo Molnar 已提交
2681 2682 2683 2684 2685 2686
#ifdef CONFIG_SCHEDSTATS
	/*
	 * 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):
	 */
2687
	if (rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
2688
		se->statistics.slice_max = max(se->statistics.slice_max,
I
Ingo Molnar 已提交
2689 2690 2691
			se->sum_exec_runtime - se->prev_sum_exec_runtime);
	}
#endif
2692
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
2693 2694
}

2695 2696 2697
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

2698 2699 2700 2701 2702 2703 2704
/*
 * 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
 */
2705
static struct sched_entity *pick_next_entity(struct cfs_rq *cfs_rq)
2706
{
2707
	struct sched_entity *se = __pick_first_entity(cfs_rq);
2708
	struct sched_entity *left = se;
2709

2710 2711 2712 2713 2714 2715 2716 2717 2718
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
		struct sched_entity *second = __pick_next_entity(se);
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
2719

2720 2721 2722 2723 2724 2725
	/*
	 * 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;

2726 2727 2728 2729 2730 2731
	/*
	 * 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;

2732
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
2733 2734

	return se;
2735 2736
}

2737 2738
static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq);

2739
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
2740 2741 2742 2743 2744 2745
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
2746
		update_curr(cfs_rq);
2747

2748 2749 2750
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

P
Peter Zijlstra 已提交
2751
	check_spread(cfs_rq, prev);
2752
	if (prev->on_rq) {
2753
		update_stats_wait_start(cfs_rq, prev);
2754 2755
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
2756
		/* in !on_rq case, update occurred at dequeue */
2757
		update_entity_load_avg(prev, 1);
2758
	}
2759
	cfs_rq->curr = NULL;
2760 2761
}

P
Peter Zijlstra 已提交
2762 2763
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
2764 2765
{
	/*
2766
	 * Update run-time statistics of the 'current'.
2767
	 */
2768
	update_curr(cfs_rq);
2769

2770 2771 2772
	/*
	 * Ensure that runnable average is periodically updated.
	 */
2773
	update_entity_load_avg(curr, 1);
2774
	update_cfs_rq_blocked_load(cfs_rq, 1);
2775
	update_cfs_shares(cfs_rq);
2776

P
Peter Zijlstra 已提交
2777 2778 2779 2780 2781
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
2782 2783 2784 2785
	if (queued) {
		resched_task(rq_of(cfs_rq)->curr);
		return;
	}
P
Peter Zijlstra 已提交
2786 2787 2788 2789 2790 2791 2792 2793
	/*
	 * 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
2794
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
2795
		check_preempt_tick(cfs_rq, curr);
2796 2797
}

2798 2799 2800 2801 2802 2803

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

#ifdef CONFIG_CFS_BANDWIDTH
2804 2805

#ifdef HAVE_JUMP_LABEL
2806
static struct static_key __cfs_bandwidth_used;
2807 2808 2809

static inline bool cfs_bandwidth_used(void)
{
2810
	return static_key_false(&__cfs_bandwidth_used);
2811 2812 2813 2814 2815 2816
}

void account_cfs_bandwidth_used(int enabled, int was_enabled)
{
	/* only need to count groups transitioning between enabled/!enabled */
	if (enabled && !was_enabled)
2817
		static_key_slow_inc(&__cfs_bandwidth_used);
2818
	else if (!enabled && was_enabled)
2819
		static_key_slow_dec(&__cfs_bandwidth_used);
2820 2821 2822 2823 2824 2825 2826 2827 2828 2829
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

void account_cfs_bandwidth_used(int enabled, int was_enabled) {}
#endif /* HAVE_JUMP_LABEL */

2830 2831 2832 2833 2834 2835 2836 2837
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
2838 2839 2840 2841 2842 2843

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

P
Paul Turner 已提交
2844 2845 2846 2847 2848 2849 2850
/*
 * 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
 */
2851
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862
{
	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);
}

2863 2864 2865 2866 2867
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

2868 2869 2870 2871 2872 2873
/* 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))
		return cfs_rq->throttled_clock_task;

2874
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
2875 2876
}

2877 2878
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
2879 2880 2881
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
2882
	u64 amount = 0, min_amount, expires;
2883 2884 2885 2886 2887 2888 2889

	/* 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;
2890
	else {
P
Paul Turner 已提交
2891 2892 2893 2894 2895 2896 2897 2898
		/*
		 * If the bandwidth pool has become inactive, then at least one
		 * period must have elapsed since the last consumption.
		 * Refresh the global state and ensure bandwidth timer becomes
		 * active.
		 */
		if (!cfs_b->timer_active) {
			__refill_cfs_bandwidth_runtime(cfs_b);
2899
			__start_cfs_bandwidth(cfs_b);
P
Paul Turner 已提交
2900
		}
2901 2902 2903 2904 2905 2906

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
2907
	}
P
Paul Turner 已提交
2908
	expires = cfs_b->runtime_expires;
2909 2910 2911
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
2912 2913 2914 2915 2916 2917 2918
	/*
	 * 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;
2919 2920

	return cfs_rq->runtime_remaining > 0;
2921 2922
}

P
Paul Turner 已提交
2923 2924 2925 2926 2927
/*
 * 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)
2928
{
P
Paul Turner 已提交
2929 2930 2931
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
2932
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
2933 2934
		return;

P
Paul Turner 已提交
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
	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
	 * whether the global deadline has advanced.
	 */

	if ((s64)(cfs_rq->runtime_expires - cfs_b->runtime_expires) >= 0) {
		/* 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;
	}
}

static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
				     unsigned long delta_exec)
{
	/* dock delta_exec before expiring quota (as it could span periods) */
2960
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
2961 2962 2963
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
2964 2965
		return;

2966 2967 2968 2969 2970 2971
	/*
	 * 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))
		resched_task(rq_of(cfs_rq)->curr);
2972 2973
}

2974 2975
static __always_inline
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, unsigned long delta_exec)
2976
{
2977
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
2978 2979 2980 2981 2982
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

2983 2984
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
2985
	return cfs_bandwidth_used() && cfs_rq->throttled;
2986 2987
}

2988 2989 2990
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
2991
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019
}

/*
 * 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--;
#ifdef CONFIG_SMP
	if (!cfs_rq->throttle_count) {
3020
		/* adjust cfs_rq_clock_task() */
3021
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
3022
					     cfs_rq->throttled_clock_task;
3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033
	}
#endif

	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)];

3034 3035
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
3036
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
3037 3038 3039 3040 3041
	cfs_rq->throttle_count++;

	return 0;
}

3042
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
3043 3044 3045 3046 3047 3048 3049 3050
{
	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;

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

3051
	/* freeze hierarchy runnable averages while throttled */
3052 3053 3054
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074

	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)
		rq->nr_running -= task_delta;

	cfs_rq->throttled = 1;
3075
	cfs_rq->throttled_clock = rq_clock(rq);
3076 3077 3078 3079 3080
	raw_spin_lock(&cfs_b->lock);
	list_add_tail_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
	raw_spin_unlock(&cfs_b->lock);
}

3081
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
3082 3083 3084 3085 3086 3087 3088
{
	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;

3089
	se = cfs_rq->tg->se[cpu_of(rq)];
3090 3091

	cfs_rq->throttled = 0;
3092 3093 3094

	update_rq_clock(rq);

3095
	raw_spin_lock(&cfs_b->lock);
3096
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
3097 3098 3099
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

3100 3101 3102
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165
	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)
		rq->nr_running += task_delta;

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
		resched_task(rq->curr);
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
	u64 runtime = remaining;

	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();

	return remaining;
}

3166 3167 3168 3169 3170 3171 3172 3173
/*
 * 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)
{
3174 3175
	u64 runtime, runtime_expires;
	int idle = 1, throttled;
3176 3177 3178 3179 3180 3181

	raw_spin_lock(&cfs_b->lock);
	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
		goto out_unlock;

3182 3183 3184
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
	/* idle depends on !throttled (for the case of a large deficit) */
	idle = cfs_b->idle && !throttled;
3185
	cfs_b->nr_periods += overrun;
3186

P
Paul Turner 已提交
3187 3188 3189 3190 3191 3192
	/* if we're going inactive then everything else can be deferred */
	if (idle)
		goto out_unlock;

	__refill_cfs_bandwidth_runtime(cfs_b);

3193 3194 3195 3196 3197 3198
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
		goto out_unlock;
	}

3199 3200 3201
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225
	/*
	 * There are throttled entities so we must first use the new bandwidth
	 * to unthrottle them before making it generally available.  This
	 * ensures that all existing debts will be paid before a new cfs_rq is
	 * allowed to run.
	 */
	runtime = cfs_b->runtime;
	runtime_expires = cfs_b->runtime_expires;
	cfs_b->runtime = 0;

	/*
	 * 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.
	 */
	while (throttled && runtime > 0) {
		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);
	}
3226

3227 3228 3229 3230 3231 3232 3233 3234 3235
	/* return (any) remaining runtime */
	cfs_b->runtime = runtime;
	/*
	 * 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;
3236 3237 3238 3239 3240 3241 3242
out_unlock:
	if (idle)
		cfs_b->timer_active = 0;
	raw_spin_unlock(&cfs_b->lock);

	return idle;
}
3243

3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307
/* 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;

/* are we near the end of the current quota period? */
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;

	start_bandwidth_timer(&cfs_b->slack_timer,
				ns_to_ktime(cfs_bandwidth_slack_period));
}

/* 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)
{
3308 3309 3310
	if (!cfs_bandwidth_used())
		return;

3311
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348
		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 */
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration))
		return;

	raw_spin_lock(&cfs_b->lock);
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice) {
		runtime = cfs_b->runtime;
		cfs_b->runtime = 0;
	}
	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)
		cfs_b->runtime = runtime;
	raw_spin_unlock(&cfs_b->lock);
}

3349 3350 3351 3352 3353 3354 3355
/*
 * 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)
{
3356 3357 3358
	if (!cfs_bandwidth_used())
		return;

3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375
	/* 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);
}

/* conditionally throttle active cfs_rq's from put_prev_entity() */
static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
{
3376 3377 3378
	if (!cfs_bandwidth_used())
		return;

3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
		return;

	/*
	 * 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))
		return;

	throttle_cfs_rq(cfs_rq);
}
3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471

static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, slack_timer);
	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);
	ktime_t now;
	int overrun;
	int idle = 0;

	for (;;) {
		now = hrtimer_cb_get_time(timer);
		overrun = hrtimer_forward(timer, now, cfs_b->period);

		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}

	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);
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	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);
}

/* requires cfs_b->lock, may release to reprogram timer */
void __start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	/*
	 * The timer may be active because we're trying to set a new bandwidth
	 * period or because we're racing with the tear-down path
	 * (timer_active==0 becomes visible before the hrtimer call-back
	 * terminates).  In either case we ensure that it's re-programmed
	 */
	while (unlikely(hrtimer_active(&cfs_b->period_timer))) {
		raw_spin_unlock(&cfs_b->lock);
		/* ensure cfs_b->lock is available while we wait */
		hrtimer_cancel(&cfs_b->period_timer);

		raw_spin_lock(&cfs_b->lock);
		/* if someone else restarted the timer then we're done */
		if (cfs_b->timer_active)
			return;
	}

	cfs_b->timer_active = 1;
	start_bandwidth_timer(&cfs_b->period_timer, cfs_b->period);
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

3472
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492
{
	struct cfs_rq *cfs_rq;

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

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
		cfs_rq->runtime_remaining = cfs_b->quota;
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
}

#else /* CONFIG_CFS_BANDWIDTH */
3493 3494
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
3495
	return rq_clock_task(rq_of(cfs_rq));
3496 3497 3498 3499
}

static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq,
				     unsigned long delta_exec) {}
3500 3501
static void check_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
3502
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
3503 3504 3505 3506 3507

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518

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;
}
3519 3520 3521 3522 3523

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) {}
3524 3525
#endif

3526 3527 3528 3529 3530
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) {}
3531
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
3532 3533 3534

#endif /* CONFIG_CFS_BANDWIDTH */

3535 3536 3537 3538
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
3539 3540 3541 3542 3543 3544 3545 3546
#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);

	WARN_ON(task_rq(p) != rq);

3547
	if (cfs_rq->nr_running > 1) {
P
Peter Zijlstra 已提交
3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561
		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)
				resched_task(p);
			return;
		}

		/*
		 * Don't schedule slices shorter than 10000ns, that just
		 * doesn't make sense. Rely on vruntime for fairness.
		 */
3562
		if (rq->curr != p)
3563
			delta = max_t(s64, 10000LL, delta);
P
Peter Zijlstra 已提交
3564

3565
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
3566 3567
	}
}
3568 3569 3570 3571 3572 3573 3574 3575 3576 3577

/*
 * 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;

3578
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
3579 3580 3581 3582 3583
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
3584
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
3585 3586 3587 3588
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
3589 3590 3591 3592

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
3593 3594
#endif

3595 3596 3597 3598 3599
/*
 * 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:
 */
3600
static void
3601
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3602 3603
{
	struct cfs_rq *cfs_rq;
3604
	struct sched_entity *se = &p->se;
3605 3606

	for_each_sched_entity(se) {
3607
		if (se->on_rq)
3608 3609
			break;
		cfs_rq = cfs_rq_of(se);
3610
		enqueue_entity(cfs_rq, se, flags);
3611 3612 3613 3614 3615 3616 3617 3618 3619

		/*
		 * 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.
		*/
		if (cfs_rq_throttled(cfs_rq))
			break;
3620
		cfs_rq->h_nr_running++;
3621

3622
		flags = ENQUEUE_WAKEUP;
3623
	}
P
Peter Zijlstra 已提交
3624

P
Peter Zijlstra 已提交
3625
	for_each_sched_entity(se) {
3626
		cfs_rq = cfs_rq_of(se);
3627
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
3628

3629 3630 3631
		if (cfs_rq_throttled(cfs_rq))
			break;

3632
		update_cfs_shares(cfs_rq);
3633
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
3634 3635
	}

3636 3637
	if (!se) {
		update_rq_runnable_avg(rq, rq->nr_running);
3638
		inc_nr_running(rq);
3639
	}
3640
	hrtick_update(rq);
3641 3642
}

3643 3644
static void set_next_buddy(struct sched_entity *se);

3645 3646 3647 3648 3649
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
3650
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
3651 3652
{
	struct cfs_rq *cfs_rq;
3653
	struct sched_entity *se = &p->se;
3654
	int task_sleep = flags & DEQUEUE_SLEEP;
3655 3656 3657

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
3658
		dequeue_entity(cfs_rq, se, flags);
3659 3660 3661 3662 3663 3664 3665 3666 3667

		/*
		 * 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;
3668
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
3669

3670
		/* Don't dequeue parent if it has other entities besides us */
3671 3672 3673 3674 3675 3676 3677
		if (cfs_rq->load.weight) {
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
			if (task_sleep && parent_entity(se))
				set_next_buddy(parent_entity(se));
3678 3679 3680

			/* avoid re-evaluating load for this entity */
			se = parent_entity(se);
3681
			break;
3682
		}
3683
		flags |= DEQUEUE_SLEEP;
3684
	}
P
Peter Zijlstra 已提交
3685

P
Peter Zijlstra 已提交
3686
	for_each_sched_entity(se) {
3687
		cfs_rq = cfs_rq_of(se);
3688
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
3689

3690 3691 3692
		if (cfs_rq_throttled(cfs_rq))
			break;

3693
		update_cfs_shares(cfs_rq);
3694
		update_entity_load_avg(se, 1);
P
Peter Zijlstra 已提交
3695 3696
	}

3697
	if (!se) {
3698
		dec_nr_running(rq);
3699 3700
		update_rq_runnable_avg(rq, 1);
	}
3701
	hrtick_update(rq);
3702 3703
}

3704
#ifdef CONFIG_SMP
3705 3706 3707
/* Used instead of source_load when we know the type == 0 */
static unsigned long weighted_cpuload(const int cpu)
{
3708
	return cpu_rq(cpu)->cfs.runnable_load_avg;
3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752
}

/*
 * 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);
}

static unsigned long power_of(int cpu)
{
	return cpu_rq(cpu)->cpu_power;
}

static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long nr_running = ACCESS_ONCE(rq->nr_running);
3753
	unsigned long load_avg = rq->cfs.runnable_load_avg;
3754 3755

	if (nr_running)
3756
		return load_avg / nr_running;
3757 3758 3759 3760

	return 0;
}

3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777
static void record_wakee(struct task_struct *p)
{
	/*
	 * Rough decay (wiping) for cost saving, don't worry
	 * about the boundary, really active task won't care
	 * about the loss.
	 */
	if (jiffies > current->wakee_flip_decay_ts + HZ) {
		current->wakee_flips = 0;
		current->wakee_flip_decay_ts = jiffies;
	}

	if (current->last_wakee != p) {
		current->last_wakee = p;
		current->wakee_flips++;
	}
}
3778

3779
static void task_waking_fair(struct task_struct *p)
3780 3781 3782
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
3783 3784 3785 3786
	u64 min_vruntime;

#ifndef CONFIG_64BIT
	u64 min_vruntime_copy;
3787

3788 3789 3790 3791 3792 3793 3794 3795
	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
3796

3797
	se->vruntime -= min_vruntime;
3798
	record_wakee(p);
3799 3800
}

3801
#ifdef CONFIG_FAIR_GROUP_SCHED
3802 3803 3804 3805 3806 3807
/*
 * 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.
3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850
 *
 * 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.
3851
 */
P
Peter Zijlstra 已提交
3852
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
3853
{
P
Peter Zijlstra 已提交
3854
	struct sched_entity *se = tg->se[cpu];
3855

3856
	if (!tg->parent || !wl)	/* the trivial, non-cgroup case */
3857 3858
		return wl;

P
Peter Zijlstra 已提交
3859
	for_each_sched_entity(se) {
3860
		long w, W;
P
Peter Zijlstra 已提交
3861

3862
		tg = se->my_q->tg;
3863

3864 3865 3866 3867
		/*
		 * W = @wg + \Sum rw_j
		 */
		W = wg + calc_tg_weight(tg, se->my_q);
P
Peter Zijlstra 已提交
3868

3869 3870 3871 3872
		/*
		 * w = rw_i + @wl
		 */
		w = se->my_q->load.weight + wl;
3873

3874 3875 3876 3877 3878
		/*
		 * wl = S * s'_i; see (2)
		 */
		if (W > 0 && w < W)
			wl = (w * tg->shares) / W;
3879 3880
		else
			wl = tg->shares;
3881

3882 3883 3884 3885 3886
		/*
		 * 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().
		 */
3887 3888
		if (wl < MIN_SHARES)
			wl = MIN_SHARES;
3889 3890 3891 3892

		/*
		 * wl = dw_i = S * (s'_i - s_i); see (3)
		 */
3893
		wl -= se->load.weight;
3894 3895 3896 3897 3898 3899 3900 3901

		/*
		 * 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
3902 3903
		wg = 0;
	}
3904

P
Peter Zijlstra 已提交
3905
	return wl;
3906 3907
}
#else
P
Peter Zijlstra 已提交
3908

3909
static long effective_load(struct task_group *tg, int cpu, long wl, long wg)
P
Peter Zijlstra 已提交
3910
{
3911
	return wl;
3912
}
P
Peter Zijlstra 已提交
3913

3914 3915
#endif

3916 3917
static int wake_wide(struct task_struct *p)
{
3918
	int factor = this_cpu_read(sd_llc_size);
3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937

	/*
	 * Yeah, it's the switching-frequency, could means many wakee or
	 * rapidly switch, use factor here will just help to automatically
	 * adjust the loose-degree, so bigger node will lead to more pull.
	 */
	if (p->wakee_flips > factor) {
		/*
		 * wakee is somewhat hot, it needs certain amount of cpu
		 * resource, so if waker is far more hot, prefer to leave
		 * it alone.
		 */
		if (current->wakee_flips > (factor * p->wakee_flips))
			return 1;
	}

	return 0;
}

3938
static int wake_affine(struct sched_domain *sd, struct task_struct *p, int sync)
3939
{
3940
	s64 this_load, load;
3941
	int idx, this_cpu, prev_cpu;
3942
	unsigned long tl_per_task;
3943
	struct task_group *tg;
3944
	unsigned long weight;
3945
	int balanced;
3946

3947 3948 3949 3950 3951 3952 3953
	/*
	 * If we wake multiple tasks be careful to not bounce
	 * ourselves around too much.
	 */
	if (wake_wide(p))
		return 0;

3954 3955 3956 3957 3958
	idx	  = sd->wake_idx;
	this_cpu  = smp_processor_id();
	prev_cpu  = task_cpu(p);
	load	  = source_load(prev_cpu, idx);
	this_load = target_load(this_cpu, idx);
3959

3960 3961 3962 3963 3964
	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current CPU:
	 */
3965 3966 3967 3968
	if (sync) {
		tg = task_group(current);
		weight = current->se.load.weight;

3969
		this_load += effective_load(tg, this_cpu, -weight, -weight);
3970 3971
		load += effective_load(tg, prev_cpu, 0, -weight);
	}
3972

3973 3974
	tg = task_group(p);
	weight = p->se.load.weight;
3975

3976 3977
	/*
	 * In low-load situations, where prev_cpu is idle and this_cpu is idle
3978 3979 3980
	 * 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.
3981 3982 3983 3984
	 *
	 * Otherwise check if either cpus are near enough in load to allow this
	 * task to be woken on this_cpu.
	 */
3985 3986
	if (this_load > 0) {
		s64 this_eff_load, prev_eff_load;
3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999

		this_eff_load = 100;
		this_eff_load *= power_of(prev_cpu);
		this_eff_load *= this_load +
			effective_load(tg, this_cpu, weight, weight);

		prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
		prev_eff_load *= power_of(this_cpu);
		prev_eff_load *= load + effective_load(tg, prev_cpu, 0, weight);

		balanced = this_eff_load <= prev_eff_load;
	} else
		balanced = true;
4000

4001
	/*
I
Ingo Molnar 已提交
4002 4003 4004
	 * If the currently running task will sleep within
	 * a reasonable amount of time then attract this newly
	 * woken task:
4005
	 */
4006 4007
	if (sync && balanced)
		return 1;
4008

4009
	schedstat_inc(p, se.statistics.nr_wakeups_affine_attempts);
4010 4011
	tl_per_task = cpu_avg_load_per_task(this_cpu);

4012 4013 4014
	if (balanced ||
	    (this_load <= load &&
	     this_load + target_load(prev_cpu, idx) <= tl_per_task)) {
4015 4016 4017 4018 4019
		/*
		 * This domain has SD_WAKE_AFFINE and
		 * p is cache cold in this domain, and
		 * there is no bad imbalance.
		 */
4020
		schedstat_inc(sd, ttwu_move_affine);
4021
		schedstat_inc(p, se.statistics.nr_wakeups_affine);
4022 4023 4024 4025 4026 4027

		return 1;
	}
	return 0;
}

4028 4029 4030 4031 4032
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
4033
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
4034
		  int this_cpu, int load_idx)
4035
{
4036
	struct sched_group *idlest = NULL, *group = sd->groups;
4037 4038
	unsigned long min_load = ULONG_MAX, this_load = 0;
	int imbalance = 100 + (sd->imbalance_pct-100)/2;
4039

4040 4041 4042 4043
	do {
		unsigned long load, avg_load;
		int local_group;
		int i;
4044

4045 4046
		/* Skip over this group if it has no CPUs allowed */
		if (!cpumask_intersects(sched_group_cpus(group),
4047
					tsk_cpus_allowed(p)))
4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066
			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;
		}

		/* Adjust by relative CPU power of the group */
4067
		avg_load = (avg_load * SCHED_POWER_SCALE) / group->sgp->power;
4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092

		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;
	int idlest = -1;
	int i;

	/* Traverse only the allowed CPUs */
4093
	for_each_cpu_and(i, sched_group_cpus(group), tsk_cpus_allowed(p)) {
4094 4095 4096 4097 4098
		load = weighted_cpuload(i);

		if (load < min_load || (load == min_load && i == this_cpu)) {
			min_load = load;
			idlest = i;
4099 4100 4101
		}
	}

4102 4103
	return idlest;
}
4104

4105 4106 4107
/*
 * Try and locate an idle CPU in the sched_domain.
 */
4108
static int select_idle_sibling(struct task_struct *p, int target)
4109
{
4110
	struct sched_domain *sd;
4111
	struct sched_group *sg;
4112
	int i = task_cpu(p);
4113

4114 4115
	if (idle_cpu(target))
		return target;
4116 4117

	/*
4118
	 * If the prevous cpu is cache affine and idle, don't be stupid.
4119
	 */
4120 4121
	if (i != target && cpus_share_cache(i, target) && idle_cpu(i))
		return i;
4122 4123

	/*
4124
	 * Otherwise, iterate the domains and find an elegible idle cpu.
4125
	 */
4126
	sd = rcu_dereference(per_cpu(sd_llc, target));
4127
	for_each_lower_domain(sd) {
4128 4129 4130 4131 4132 4133 4134
		sg = sd->groups;
		do {
			if (!cpumask_intersects(sched_group_cpus(sg),
						tsk_cpus_allowed(p)))
				goto next;

			for_each_cpu(i, sched_group_cpus(sg)) {
4135
				if (i == target || !idle_cpu(i))
4136 4137
					goto next;
			}
4138

4139 4140 4141 4142 4143 4144 4145 4146
			target = cpumask_first_and(sched_group_cpus(sg),
					tsk_cpus_allowed(p));
			goto done;
next:
			sg = sg->next;
		} while (sg != sd->groups);
	}
done:
4147 4148 4149
	return target;
}

4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160
/*
 * sched_balance_self: balance the current task (running on cpu) in domains
 * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and
 * SD_BALANCE_EXEC.
 *
 * Balance, ie. select the least loaded group.
 *
 * Returns the target CPU number, or the same CPU if no balancing is needed.
 *
 * preempt must be disabled.
 */
4161
static int
4162
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
4163
{
4164
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
4165 4166
	int cpu = smp_processor_id();
	int new_cpu = cpu;
4167
	int want_affine = 0;
4168
	int sync = wake_flags & WF_SYNC;
4169

4170
	if (p->nr_cpus_allowed == 1)
4171 4172
		return prev_cpu;

4173
	if (sd_flag & SD_BALANCE_WAKE) {
4174
		if (cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
4175 4176 4177
			want_affine = 1;
		new_cpu = prev_cpu;
	}
4178

4179
	rcu_read_lock();
4180
	for_each_domain(cpu, tmp) {
4181 4182 4183
		if (!(tmp->flags & SD_LOAD_BALANCE))
			continue;

4184
		/*
4185 4186
		 * If both cpu and prev_cpu are part of this domain,
		 * cpu is a valid SD_WAKE_AFFINE target.
4187
		 */
4188 4189 4190
		if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
		    cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
			affine_sd = tmp;
4191
			break;
4192
		}
4193

4194
		if (tmp->flags & sd_flag)
4195 4196 4197
			sd = tmp;
	}

4198
	if (affine_sd) {
4199
		if (cpu != prev_cpu && wake_affine(affine_sd, p, sync))
4200 4201 4202 4203
			prev_cpu = cpu;

		new_cpu = select_idle_sibling(p, prev_cpu);
		goto unlock;
4204
	}
4205

4206
	while (sd) {
4207
		int load_idx = sd->forkexec_idx;
4208
		struct sched_group *group;
4209
		int weight;
4210

4211
		if (!(sd->flags & sd_flag)) {
4212 4213 4214
			sd = sd->child;
			continue;
		}
4215

4216 4217
		if (sd_flag & SD_BALANCE_WAKE)
			load_idx = sd->wake_idx;
4218

4219
		group = find_idlest_group(sd, p, cpu, load_idx);
4220 4221 4222 4223
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
4224

4225
		new_cpu = find_idlest_cpu(group, p, cpu);
4226 4227 4228 4229
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
4230
		}
4231 4232 4233

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
4234
		weight = sd->span_weight;
4235 4236
		sd = NULL;
		for_each_domain(cpu, tmp) {
4237
			if (weight <= tmp->span_weight)
4238
				break;
4239
			if (tmp->flags & sd_flag)
4240 4241 4242
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
4243
	}
4244 4245
unlock:
	rcu_read_unlock();
4246

4247
	return new_cpu;
4248
}
4249 4250 4251 4252 4253 4254 4255 4256 4257 4258

/*
 * 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
 * previous cpu.  However, the caller only guarantees p->pi_lock is held; no
 * other assumptions, including the state of rq->lock, should be made.
 */
static void
migrate_task_rq_fair(struct task_struct *p, int next_cpu)
{
4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	/*
	 * Load tracking: accumulate removed load so that it can be processed
	 * when we next update owning cfs_rq under rq->lock.  Tasks contribute
	 * to blocked load iff they have a positive decay-count.  It can never
	 * be negative here since on-rq tasks have decay-count == 0.
	 */
	if (se->avg.decay_count) {
		se->avg.decay_count = -__synchronize_entity_decay(se);
4270 4271
		atomic_long_add(se->avg.load_avg_contrib,
						&cfs_rq->removed_load);
4272
	}
4273
}
4274 4275
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
4276 4277
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
4278 4279 4280 4281
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
4282 4283
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
4284 4285 4286 4287 4288 4289 4290 4291 4292
	 *
	 * 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.
4293
	 */
4294
	return calc_delta_fair(gran, se);
4295 4296
}

4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318
/*
 * 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
4319
	gran = wakeup_gran(curr, se);
4320 4321 4322 4323 4324 4325
	if (vdiff > gran)
		return 1;

	return 0;
}

4326 4327
static void set_last_buddy(struct sched_entity *se)
{
4328 4329 4330 4331 4332
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->last = se;
4333 4334 4335 4336
}

static void set_next_buddy(struct sched_entity *se)
{
4337 4338 4339 4340 4341
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

	for_each_sched_entity(se)
		cfs_rq_of(se)->next = se;
4342 4343
}

4344 4345
static void set_skip_buddy(struct sched_entity *se)
{
4346 4347
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
4348 4349
}

4350 4351 4352
/*
 * Preempt the current task with a newly woken task if needed:
 */
4353
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
4354 4355
{
	struct task_struct *curr = rq->curr;
4356
	struct sched_entity *se = &curr->se, *pse = &p->se;
4357
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
4358
	int scale = cfs_rq->nr_running >= sched_nr_latency;
4359
	int next_buddy_marked = 0;
4360

I
Ingo Molnar 已提交
4361 4362 4363
	if (unlikely(se == pse))
		return;

4364
	/*
4365
	 * This is possible from callers such as move_task(), in which we
4366 4367 4368 4369 4370 4371 4372
	 * 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;

4373
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
4374
		set_next_buddy(pse);
4375 4376
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
4377

4378 4379 4380
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
4381 4382 4383 4384 4385 4386
	 *
	 * 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.
4387 4388 4389 4390
	 */
	if (test_tsk_need_resched(curr))
		return;

4391 4392 4393 4394 4395
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

4396
	/*
4397 4398
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
4399
	 */
4400
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
4401
		return;
4402

4403
	find_matching_se(&se, &pse);
4404
	update_curr(cfs_rq_of(se));
4405
	BUG_ON(!pse);
4406 4407 4408 4409 4410 4411 4412
	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);
4413
		goto preempt;
4414
	}
4415

4416
	return;
4417

4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433
preempt:
	resched_task(curr);
	/*
	 * 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);
4434 4435
}

4436
static struct task_struct *pick_next_task_fair(struct rq *rq)
4437
{
P
Peter Zijlstra 已提交
4438
	struct task_struct *p;
4439 4440 4441
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;

4442
	if (!cfs_rq->nr_running)
4443 4444 4445
		return NULL;

	do {
4446
		se = pick_next_entity(cfs_rq);
4447
		set_next_entity(cfs_rq, se);
4448 4449 4450
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
4451
	p = task_of(se);
4452 4453
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
4454 4455

	return p;
4456 4457 4458 4459 4460
}

/*
 * Account for a descheduled task:
 */
4461
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
4462 4463 4464 4465 4466 4467
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
4468
		put_prev_entity(cfs_rq, se);
4469 4470 4471
	}
}

4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496
/*
 * 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);
4497 4498 4499 4500 4501 4502
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
		 rq->skip_clock_update = 1;
4503 4504 4505 4506 4507
	}

	set_skip_buddy(se);
}

4508 4509 4510 4511
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

4512 4513
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
4514 4515 4516 4517 4518 4519 4520 4521 4522 4523
		return false;

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

	yield_task_fair(rq);

	return true;
}

4524
#ifdef CONFIG_SMP
4525
/**************************************************
P
Peter Zijlstra 已提交
4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641
 * 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
 * is derived from the nice value as per prio_to_weight[].
 *
 * 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)
 *
 * P_i is the cpu power (or compute capacity) of cpu i, typically it is the
 * 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):
 *
 *   imb_i,j = max{ avg(W/P), W_i/P_i } - min{ avg(W/P), W_j/P_j }    (4)
 *
 * 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:
 *
 *             log_2 n     
 *   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.]
 */ 
4642

4643 4644
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

4645 4646
enum fbq_type { regular, remote, all };

4647
#define LBF_ALL_PINNED	0x01
4648
#define LBF_NEED_BREAK	0x02
4649 4650
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
4651 4652 4653 4654 4655

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
4656
	int			src_cpu;
4657 4658 4659 4660

	int			dst_cpu;
	struct rq		*dst_rq;

4661 4662
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
4663
	enum cpu_idle_type	idle;
4664
	long			imbalance;
4665 4666 4667
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

4668
	unsigned int		flags;
4669 4670 4671 4672

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
4673 4674

	enum fbq_type		fbq_type;
4675 4676
};

4677
/*
4678
 * move_task - move a task from one runqueue to another runqueue.
4679 4680
 * Both runqueues must be locked.
 */
4681
static void move_task(struct task_struct *p, struct lb_env *env)
4682
{
4683 4684 4685 4686
	deactivate_task(env->src_rq, p, 0);
	set_task_cpu(p, env->dst_cpu);
	activate_task(env->dst_rq, p, 0);
	check_preempt_curr(env->dst_rq, p, 0);
4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700
#ifdef CONFIG_NUMA_BALANCING
	if (p->numa_preferred_nid != -1) {
		int src_nid = cpu_to_node(env->src_cpu);
		int dst_nid = cpu_to_node(env->dst_cpu);

		/*
		 * If the load balancer has moved the task then limit
		 * migrations from taking place in the short term in
		 * case this is a short-lived migration.
		 */
		if (src_nid != dst_nid && dst_nid != p->numa_preferred_nid)
			p->numa_migrate_seq = 0;
	}
#endif
4701 4702
}

4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734
/*
 * Is this task likely cache-hot:
 */
static int
task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
{
	s64 delta;

	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
	if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running &&
			(&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;

	delta = now - p->se.exec_start;

	return delta < (s64)sysctl_sched_migration_cost;
}

4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748
#ifdef CONFIG_NUMA_BALANCING
/* Returns true if the destination node has incurred more faults */
static bool migrate_improves_locality(struct task_struct *p, struct lb_env *env)
{
	int src_nid, dst_nid;

	if (!sched_feat(NUMA_FAVOUR_HIGHER) || !p->numa_faults ||
	    !(env->sd->flags & SD_NUMA)) {
		return false;
	}

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

4749
	if (src_nid == dst_nid)
4750 4751
		return false;

4752 4753 4754 4755
	/* Always encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
		return true;

4756 4757 4758
	/* If both task and group weight improve, this move is a winner. */
	if (task_weight(p, dst_nid) > task_weight(p, src_nid) &&
	    group_weight(p, dst_nid) > group_weight(p, src_nid))
4759 4760 4761 4762
		return true;

	return false;
}
4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777


static bool migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
{
	int src_nid, dst_nid;

	if (!sched_feat(NUMA) || !sched_feat(NUMA_RESIST_LOWER))
		return false;

	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
		return false;

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

4778
	if (src_nid == dst_nid)
4779 4780
		return false;

4781 4782 4783 4784
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid)
		return true;

4785 4786 4787
	/* If either task or group weight get worse, don't do it. */
	if (task_weight(p, dst_nid) < task_weight(p, src_nid) ||
	    group_weight(p, dst_nid) < group_weight(p, src_nid))
4788 4789 4790 4791 4792
		return true;

	return false;
}

4793 4794 4795 4796 4797 4798
#else
static inline bool migrate_improves_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
4799 4800 4801 4802 4803 4804

static inline bool migrate_degrades_locality(struct task_struct *p,
					     struct lb_env *env)
{
	return false;
}
4805 4806
#endif

4807 4808 4809 4810
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
4811
int can_migrate_task(struct task_struct *p, struct lb_env *env)
4812 4813 4814 4815
{
	int tsk_cache_hot = 0;
	/*
	 * We do not migrate tasks that are:
4816
	 * 1) throttled_lb_pair, or
4817
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
4818 4819
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
4820
	 */
4821 4822 4823
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

4824
	if (!cpumask_test_cpu(env->dst_cpu, tsk_cpus_allowed(p))) {
4825
		int cpu;
4826

4827
		schedstat_inc(p, se.statistics.nr_failed_migrations_affine);
4828

4829 4830
		env->flags |= LBF_SOME_PINNED;

4831 4832 4833 4834 4835 4836 4837 4838
		/*
		 * 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.
		 */
4839
		if (!env->dst_grpmask || (env->flags & LBF_DST_PINNED))
4840 4841
			return 0;

4842 4843 4844
		/* 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))) {
4845
				env->flags |= LBF_DST_PINNED;
4846 4847 4848
				env->new_dst_cpu = cpu;
				break;
			}
4849
		}
4850

4851 4852
		return 0;
	}
4853 4854

	/* Record that we found atleast one task that could run on dst_cpu */
4855
	env->flags &= ~LBF_ALL_PINNED;
4856

4857
	if (task_running(env->src_rq, p)) {
4858
		schedstat_inc(p, se.statistics.nr_failed_migrations_running);
4859 4860 4861 4862 4863
		return 0;
	}

	/*
	 * Aggressive migration if:
4864 4865 4866
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
4867
	 */
4868
	tsk_cache_hot = task_hot(p, rq_clock_task(env->src_rq), env->sd);
4869 4870
	if (!tsk_cache_hot)
		tsk_cache_hot = migrate_degrades_locality(p, env);
4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881

	if (migrate_improves_locality(p, env)) {
#ifdef CONFIG_SCHEDSTATS
		if (tsk_cache_hot) {
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
			schedstat_inc(p, se.statistics.nr_forced_migrations);
		}
#endif
		return 1;
	}

4882
	if (!tsk_cache_hot ||
4883
		env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
Z
Zhang Hang 已提交
4884

4885
		if (tsk_cache_hot) {
4886
			schedstat_inc(env->sd, lb_hot_gained[env->idle]);
4887
			schedstat_inc(p, se.statistics.nr_forced_migrations);
4888
		}
Z
Zhang Hang 已提交
4889

4890 4891 4892
		return 1;
	}

Z
Zhang Hang 已提交
4893 4894
	schedstat_inc(p, se.statistics.nr_failed_migrations_hot);
	return 0;
4895 4896
}

4897 4898 4899 4900 4901 4902 4903
/*
 * move_one_task tries to move exactly one task from busiest to this_rq, as
 * part of active balancing operations within "domain".
 * Returns 1 if successful and 0 otherwise.
 *
 * Called with both runqueues locked.
 */
4904
static int move_one_task(struct lb_env *env)
4905 4906 4907
{
	struct task_struct *p, *n;

4908 4909 4910
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
4911

4912 4913 4914 4915 4916 4917 4918 4919
		move_task(p, env);
		/*
		 * Right now, this is only the second place move_task()
		 * is called, so we can safely collect move_task()
		 * stats here rather than inside move_task().
		 */
		schedstat_inc(env->sd, lb_gained[env->idle]);
		return 1;
4920 4921 4922 4923
	}
	return 0;
}

4924 4925
static const unsigned int sched_nr_migrate_break = 32;

4926
/*
4927
 * move_tasks tries to move up to imbalance weighted load from busiest to
4928 4929 4930 4931 4932 4933
 * this_rq, as part of a balancing operation within domain "sd".
 * Returns 1 if successful and 0 otherwise.
 *
 * Called with both runqueues locked.
 */
static int move_tasks(struct lb_env *env)
4934
{
4935 4936
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
4937 4938
	unsigned long load;
	int pulled = 0;
4939

4940
	if (env->imbalance <= 0)
4941
		return 0;
4942

4943 4944
	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
4945

4946 4947
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
4948
		if (env->loop > env->loop_max)
4949
			break;
4950 4951

		/* take a breather every nr_migrate tasks */
4952
		if (env->loop > env->loop_break) {
4953
			env->loop_break += sched_nr_migrate_break;
4954
			env->flags |= LBF_NEED_BREAK;
4955
			break;
4956
		}
4957

4958
		if (!can_migrate_task(p, env))
4959 4960 4961
			goto next;

		load = task_h_load(p);
4962

4963
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
4964 4965
			goto next;

4966
		if ((load / 2) > env->imbalance)
4967
			goto next;
4968

4969
		move_task(p, env);
4970
		pulled++;
4971
		env->imbalance -= load;
4972 4973

#ifdef CONFIG_PREEMPT
4974 4975 4976 4977 4978
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
		 * kernels will stop after the first task is pulled to minimize
		 * the critical section.
		 */
4979
		if (env->idle == CPU_NEWLY_IDLE)
4980
			break;
4981 4982
#endif

4983 4984 4985 4986
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
4987
		if (env->imbalance <= 0)
4988
			break;
4989 4990 4991

		continue;
next:
4992
		list_move_tail(&p->se.group_node, tasks);
4993
	}
4994

4995
	/*
4996 4997 4998
	 * Right now, this is one of only two places move_task() is called,
	 * so we can safely collect move_task() stats here rather than
	 * inside move_task().
4999
	 */
5000
	schedstat_add(env->sd, lb_gained[env->idle], pulled);
5001

5002
	return pulled;
5003 5004
}

P
Peter Zijlstra 已提交
5005
#ifdef CONFIG_FAIR_GROUP_SCHED
5006 5007 5008
/*
 * update tg->load_weight by folding this cpu's load_avg
 */
5009
static void __update_blocked_averages_cpu(struct task_group *tg, int cpu)
5010
{
5011 5012
	struct sched_entity *se = tg->se[cpu];
	struct cfs_rq *cfs_rq = tg->cfs_rq[cpu];
5013

5014 5015 5016
	/* throttled entities do not contribute to load */
	if (throttled_hierarchy(cfs_rq))
		return;
5017

5018
	update_cfs_rq_blocked_load(cfs_rq, 1);
5019

5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033
	if (se) {
		update_entity_load_avg(se, 1);
		/*
		 * We pivot on our runnable average having decayed to zero for
		 * list removal.  This generally implies that all our children
		 * have also been removed (modulo rounding error or bandwidth
		 * control); however, such cases are rare and we can fix these
		 * at enqueue.
		 *
		 * TODO: fix up out-of-order children on enqueue.
		 */
		if (!se->avg.runnable_avg_sum && !cfs_rq->nr_running)
			list_del_leaf_cfs_rq(cfs_rq);
	} else {
5034
		struct rq *rq = rq_of(cfs_rq);
5035 5036
		update_rq_runnable_avg(rq, rq->nr_running);
	}
5037 5038
}

5039
static void update_blocked_averages(int cpu)
5040 5041
{
	struct rq *rq = cpu_rq(cpu);
5042 5043
	struct cfs_rq *cfs_rq;
	unsigned long flags;
5044

5045 5046
	raw_spin_lock_irqsave(&rq->lock, flags);
	update_rq_clock(rq);
5047 5048 5049 5050
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
5051
	for_each_leaf_cfs_rq(rq, cfs_rq) {
5052 5053 5054 5055 5056 5057
		/*
		 * Note: We may want to consider periodically releasing
		 * rq->lock about these updates so that creating many task
		 * groups does not result in continually extending hold time.
		 */
		__update_blocked_averages_cpu(cfs_rq->tg, rq->cpu);
5058
	}
5059 5060

	raw_spin_unlock_irqrestore(&rq->lock, flags);
5061 5062
}

5063
/*
5064
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
5065 5066 5067
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
5068
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
5069
{
5070 5071
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
5072
	unsigned long now = jiffies;
5073
	unsigned long load;
5074

5075
	if (cfs_rq->last_h_load_update == now)
5076 5077
		return;

5078 5079 5080 5081 5082 5083 5084
	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;
	}
5085

5086
	if (!se) {
5087
		cfs_rq->h_load = cfs_rq->runnable_load_avg;
5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
		load = div64_ul(load * se->avg.load_avg_contrib,
				cfs_rq->runnable_load_avg + 1);
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
5099 5100
}

5101
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
5102
{
5103
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
5104

5105
	update_cfs_rq_h_load(cfs_rq);
5106 5107
	return div64_ul(p->se.avg.load_avg_contrib * cfs_rq->h_load,
			cfs_rq->runnable_load_avg + 1);
P
Peter Zijlstra 已提交
5108 5109
}
#else
5110
static inline void update_blocked_averages(int cpu)
5111 5112 5113
{
}

5114
static unsigned long task_h_load(struct task_struct *p)
5115
{
5116
	return p->se.avg.load_avg_contrib;
5117
}
P
Peter Zijlstra 已提交
5118
#endif
5119 5120 5121 5122 5123 5124 5125 5126 5127

/********** Helpers for find_busiest_group ************************/
/*
 * 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
5128
	unsigned long load_per_task;
5129
	unsigned long group_power;
5130 5131 5132 5133
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int group_capacity;
	unsigned int idle_cpus;
	unsigned int group_weight;
5134
	int group_imb; /* Is there an imbalance in the group ? */
5135
	int group_has_capacity; /* Is there extra capacity in the group? */
5136 5137 5138 5139
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
5140 5141
};

J
Joonsoo Kim 已提交
5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153
/*
 * 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 */
	unsigned long total_pwr;	/* Total power of all groups in sd */
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
5154
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
5155 5156
};

5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175
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,
		.total_pwr = 0UL,
		.busiest_stat = {
			.avg_load = 0UL,
		},
	};
}

5176 5177 5178 5179
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
 * @idle: The Idle status of the CPU for whose sd load_icx is obtained.
5180 5181
 *
 * Return: The load index.
5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203
 */
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;
}

5204
static unsigned long default_scale_freq_power(struct sched_domain *sd, int cpu)
5205
{
5206
	return SCHED_POWER_SCALE;
5207 5208 5209 5210 5211 5212 5213
}

unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
{
	return default_scale_freq_power(sd, cpu);
}

5214
static unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu)
5215
{
5216
	unsigned long weight = sd->span_weight;
5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228
	unsigned long smt_gain = sd->smt_gain;

	smt_gain /= weight;

	return smt_gain;
}

unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu)
{
	return default_scale_smt_power(sd, cpu);
}

5229
static unsigned long scale_rt_power(int cpu)
5230 5231
{
	struct rq *rq = cpu_rq(cpu);
5232
	u64 total, available, age_stamp, avg;
5233

5234 5235 5236 5237 5238 5239 5240
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
	age_stamp = ACCESS_ONCE(rq->age_stamp);
	avg = ACCESS_ONCE(rq->rt_avg);

5241
	total = sched_avg_period() + (rq_clock(rq) - age_stamp);
5242

5243
	if (unlikely(total < avg)) {
5244 5245 5246
		/* Ensures that power won't end up being negative */
		available = 0;
	} else {
5247
		available = total - avg;
5248
	}
5249

5250 5251
	if (unlikely((s64)total < SCHED_POWER_SCALE))
		total = SCHED_POWER_SCALE;
5252

5253
	total >>= SCHED_POWER_SHIFT;
5254 5255 5256 5257 5258 5259

	return div_u64(available, total);
}

static void update_cpu_power(struct sched_domain *sd, int cpu)
{
5260
	unsigned long weight = sd->span_weight;
5261
	unsigned long power = SCHED_POWER_SCALE;
5262 5263 5264 5265 5266 5267 5268 5269
	struct sched_group *sdg = sd->groups;

	if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
		if (sched_feat(ARCH_POWER))
			power *= arch_scale_smt_power(sd, cpu);
		else
			power *= default_scale_smt_power(sd, cpu);

5270
		power >>= SCHED_POWER_SHIFT;
5271 5272
	}

5273
	sdg->sgp->power_orig = power;
5274 5275 5276 5277 5278 5279

	if (sched_feat(ARCH_POWER))
		power *= arch_scale_freq_power(sd, cpu);
	else
		power *= default_scale_freq_power(sd, cpu);

5280
	power >>= SCHED_POWER_SHIFT;
5281

5282
	power *= scale_rt_power(cpu);
5283
	power >>= SCHED_POWER_SHIFT;
5284 5285 5286 5287

	if (!power)
		power = 1;

5288
	cpu_rq(cpu)->cpu_power = power;
5289
	sdg->sgp->power = power;
5290 5291
}

5292
void update_group_power(struct sched_domain *sd, int cpu)
5293 5294 5295
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
5296
	unsigned long power, power_orig;
5297 5298 5299 5300 5301
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
	sdg->sgp->next_update = jiffies + interval;
5302 5303 5304 5305 5306 5307

	if (!child) {
		update_cpu_power(sd, cpu);
		return;
	}

5308
	power_orig = power = 0;
5309

P
Peter Zijlstra 已提交
5310 5311 5312 5313 5314 5315
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

5316 5317 5318 5319 5320 5321
		for_each_cpu(cpu, sched_group_cpus(sdg)) {
			struct sched_group *sg = cpu_rq(cpu)->sd->groups;

			power_orig += sg->sgp->power_orig;
			power += sg->sgp->power;
		}
P
Peter Zijlstra 已提交
5322 5323 5324 5325 5326 5327 5328 5329
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
		 */ 

		group = child->groups;
		do {
5330
			power_orig += group->sgp->power_orig;
P
Peter Zijlstra 已提交
5331 5332 5333 5334
			power += group->sgp->power;
			group = group->next;
		} while (group != child->groups);
	}
5335

5336 5337
	sdg->sgp->power_orig = power_orig;
	sdg->sgp->power = power;
5338 5339
}

5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350
/*
 * Try and fix up capacity for tiny siblings, this is needed when
 * things like SD_ASYM_PACKING need f_b_g to select another sibling
 * which on its own isn't powerful enough.
 *
 * See update_sd_pick_busiest() and check_asym_packing().
 */
static inline int
fix_small_capacity(struct sched_domain *sd, struct sched_group *group)
{
	/*
5351
	 * Only siblings can have significantly less than SCHED_POWER_SCALE
5352
	 */
P
Peter Zijlstra 已提交
5353
	if (!(sd->flags & SD_SHARE_CPUPOWER))
5354 5355 5356 5357 5358
		return 0;

	/*
	 * If ~90% of the cpu_power is still there, we're good.
	 */
5359
	if (group->sgp->power * 32 > group->sgp->power_orig * 29)
5360 5361 5362 5363 5364
		return 1;

	return 0;
}

5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380
/*
 * 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
5381 5382
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
5383 5384 5385
 *
 * When this is so detected; this group becomes a candidate for busiest; see
 * update_sd_pick_busiest(). And calculcate_imbalance() and
5386
 * find_busiest_group() avoid some of the usual balance conditions to allow it
5387 5388 5389 5390 5391 5392 5393
 * 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.
 */

5394
static inline int sg_imbalanced(struct sched_group *group)
5395
{
5396
	return group->sgp->imbalance;
5397 5398
}

5399 5400 5401
/*
 * Compute the group capacity.
 *
5402 5403 5404
 * Avoid the issue where N*frac(smt_power) >= 1 creates 'phantom' cores by
 * first dividing out the smt factor and computing the actual number of cores
 * and limit power unit capacity with that.
5405 5406 5407
 */
static inline int sg_capacity(struct lb_env *env, struct sched_group *group)
{
5408 5409 5410 5411 5412 5413
	unsigned int capacity, smt, cpus;
	unsigned int power, power_orig;

	power = group->sgp->power;
	power_orig = group->sgp->power_orig;
	cpus = group->group_weight;
5414

5415 5416 5417
	/* smt := ceil(cpus / power), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_POWER_SCALE * cpus, power_orig);
	capacity = cpus / smt; /* cores */
5418

5419
	capacity = min_t(unsigned, capacity, DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE));
5420 5421 5422 5423 5424 5425
	if (!capacity)
		capacity = fix_small_capacity(env->sd, group);

	return capacity;
}

5426 5427
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
5428
 * @env: The load balancing environment.
5429 5430 5431 5432 5433
 * @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.
 */
5434 5435
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
5436
			int local_group, struct sg_lb_stats *sgs)
5437
{
5438 5439
	unsigned long nr_running;
	unsigned long load;
5440
	int i;
5441

5442 5443
	memset(sgs, 0, sizeof(*sgs));

5444
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
5445 5446
		struct rq *rq = cpu_rq(i);

5447 5448
		nr_running = rq->nr_running;

5449
		/* Bias balancing toward cpus of our domain */
5450
		if (local_group)
5451
			load = target_load(i, load_idx);
5452
		else
5453 5454 5455
			load = source_load(i, load_idx);

		sgs->group_load += load;
5456
		sgs->sum_nr_running += nr_running;
5457 5458 5459 5460
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
5461
		sgs->sum_weighted_load += weighted_cpuload(i);
5462 5463
		if (idle_cpu(i))
			sgs->idle_cpus++;
5464 5465 5466
	}

	/* Adjust by relative CPU power of the group */
5467 5468
	sgs->group_power = group->sgp->power;
	sgs->avg_load = (sgs->group_load*SCHED_POWER_SCALE) / sgs->group_power;
5469

5470
	if (sgs->sum_nr_running)
5471
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
5472

5473
	sgs->group_weight = group->group_weight;
5474

5475 5476 5477
	sgs->group_imb = sg_imbalanced(group);
	sgs->group_capacity = sg_capacity(env, group);

5478 5479
	if (sgs->group_capacity > sgs->sum_nr_running)
		sgs->group_has_capacity = 1;
5480 5481
}

5482 5483
/**
 * update_sd_pick_busiest - return 1 on busiest group
5484
 * @env: The load balancing environment.
5485 5486
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
5487
 * @sgs: sched_group statistics
5488 5489 5490
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
5491 5492 5493
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
5494
 */
5495
static bool update_sd_pick_busiest(struct lb_env *env,
5496 5497
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
5498
				   struct sg_lb_stats *sgs)
5499
{
J
Joonsoo Kim 已提交
5500
	if (sgs->avg_load <= sds->busiest_stat.avg_load)
5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513
		return false;

	if (sgs->sum_nr_running > sgs->group_capacity)
		return true;

	if (sgs->group_imb)
		return true;

	/*
	 * 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.
	 */
5514 5515
	if ((env->sd->flags & SD_ASYM_PACKING) && sgs->sum_nr_running &&
	    env->dst_cpu < group_first_cpu(sg)) {
5516 5517 5518 5519 5520 5521 5522 5523 5524 5525
		if (!sds->busiest)
			return true;

		if (group_first_cpu(sds->busiest) > group_first_cpu(sg))
			return true;
	}

	return false;
}

5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555
#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 */

5556
/**
5557
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
5558
 * @env: The load balancing environment.
5559 5560 5561
 * @balance: Should we balance.
 * @sds: variable to hold the statistics for this sched_domain.
 */
5562
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
5563
{
5564 5565
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
J
Joonsoo Kim 已提交
5566
	struct sg_lb_stats tmp_sgs;
5567 5568 5569 5570 5571
	int load_idx, prefer_sibling = 0;

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

5572
	load_idx = get_sd_load_idx(env->sd, env->idle);
5573 5574

	do {
J
Joonsoo Kim 已提交
5575
		struct sg_lb_stats *sgs = &tmp_sgs;
5576 5577
		int local_group;

5578
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_cpus(sg));
J
Joonsoo Kim 已提交
5579 5580 5581
		if (local_group) {
			sds->local = sg;
			sgs = &sds->local_stat;
5582 5583 5584 5585

			if (env->idle != CPU_NEWLY_IDLE ||
			    time_after_eq(jiffies, sg->sgp->next_update))
				update_group_power(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
5586
		}
5587

J
Joonsoo Kim 已提交
5588
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs);
5589

5590 5591 5592
		if (local_group)
			goto next_group;

5593 5594
		/*
		 * In case the child domain prefers tasks go to siblings
5595
		 * first, lower the sg capacity to one so that we'll try
5596 5597 5598 5599 5600 5601
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
		 * these excess tasks, i.e. nr_running < group_capacity. 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).
5602
		 */
5603 5604
		if (prefer_sibling && sds->local &&
		    sds->local_stat.group_has_capacity)
5605
			sgs->group_capacity = min(sgs->group_capacity, 1U);
5606

5607
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
5608
			sds->busiest = sg;
J
Joonsoo Kim 已提交
5609
			sds->busiest_stat = *sgs;
5610 5611
		}

5612 5613 5614 5615 5616
next_group:
		/* Now, start updating sd_lb_stats */
		sds->total_load += sgs->group_load;
		sds->total_pwr += sgs->group_power;

5617
		sg = sg->next;
5618
	} while (sg != env->sd->groups);
5619 5620 5621

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640
}

/**
 * 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.
 *
5641
 * Return: 1 when packing is required and a task should be moved to
5642 5643
 * this CPU.  The amount of the imbalance is returned in *imbalance.
 *
5644
 * @env: The load balancing environment.
5645 5646
 * @sds: Statistics of the sched_domain which is to be packed
 */
5647
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
5648 5649 5650
{
	int busiest_cpu;

5651
	if (!(env->sd->flags & SD_ASYM_PACKING))
5652 5653 5654 5655 5656 5657
		return 0;

	if (!sds->busiest)
		return 0;

	busiest_cpu = group_first_cpu(sds->busiest);
5658
	if (env->dst_cpu > busiest_cpu)
5659 5660
		return 0;

5661
	env->imbalance = DIV_ROUND_CLOSEST(
5662 5663
		sds->busiest_stat.avg_load * sds->busiest_stat.group_power,
		SCHED_POWER_SCALE);
5664

5665
	return 1;
5666 5667 5668 5669 5670 5671
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
5672
 * @env: The load balancing environment.
5673 5674
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
5675 5676
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
5677 5678 5679
{
	unsigned long tmp, pwr_now = 0, pwr_move = 0;
	unsigned int imbn = 2;
5680
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
5681
	struct sg_lb_stats *local, *busiest;
5682

J
Joonsoo Kim 已提交
5683 5684
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
5685

J
Joonsoo Kim 已提交
5686 5687 5688 5689
	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;
5690

J
Joonsoo Kim 已提交
5691 5692
	scaled_busy_load_per_task =
		(busiest->load_per_task * SCHED_POWER_SCALE) /
5693
		busiest->group_power;
J
Joonsoo Kim 已提交
5694

5695 5696
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
5697
		env->imbalance = busiest->load_per_task;
5698 5699 5700 5701 5702 5703 5704 5705 5706
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
	 * however we may be able to increase total CPU power used by
	 * moving them.
	 */

5707
	pwr_now += busiest->group_power *
J
Joonsoo Kim 已提交
5708
			min(busiest->load_per_task, busiest->avg_load);
5709
	pwr_now += local->group_power *
J
Joonsoo Kim 已提交
5710
			min(local->load_per_task, local->avg_load);
5711
	pwr_now /= SCHED_POWER_SCALE;
5712 5713

	/* Amount of load we'd subtract */
J
Joonsoo Kim 已提交
5714
	tmp = (busiest->load_per_task * SCHED_POWER_SCALE) /
5715
		busiest->group_power;
J
Joonsoo Kim 已提交
5716
	if (busiest->avg_load > tmp) {
5717
		pwr_move += busiest->group_power *
J
Joonsoo Kim 已提交
5718 5719 5720
			    min(busiest->load_per_task,
				busiest->avg_load - tmp);
	}
5721 5722

	/* Amount of load we'd add */
5723
	if (busiest->avg_load * busiest->group_power <
J
Joonsoo Kim 已提交
5724
	    busiest->load_per_task * SCHED_POWER_SCALE) {
5725 5726
		tmp = (busiest->avg_load * busiest->group_power) /
		      local->group_power;
J
Joonsoo Kim 已提交
5727 5728
	} else {
		tmp = (busiest->load_per_task * SCHED_POWER_SCALE) /
5729
		      local->group_power;
J
Joonsoo Kim 已提交
5730
	}
5731 5732
	pwr_move += local->group_power *
		    min(local->load_per_task, local->avg_load + tmp);
5733
	pwr_move /= SCHED_POWER_SCALE;
5734 5735 5736

	/* Move if we gain throughput */
	if (pwr_move > pwr_now)
J
Joonsoo Kim 已提交
5737
		env->imbalance = busiest->load_per_task;
5738 5739 5740 5741 5742
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
5743
 * @env: load balance environment
5744 5745
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
5746
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
5747
{
5748
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
5749 5750 5751 5752
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
5753

J
Joonsoo Kim 已提交
5754
	if (busiest->group_imb) {
5755 5756 5757 5758
		/*
		 * 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
5759 5760
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
5761 5762
	}

5763 5764 5765 5766 5767
	/*
	 * In the presence of smp nice balancing, certain scenarios can have
	 * max load less than avg load(as we skip the groups at or below
	 * its cpu_power, while calculating max_load..)
	 */
5768 5769
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
5770 5771
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
5772 5773
	}

J
Joonsoo Kim 已提交
5774
	if (!busiest->group_imb) {
5775 5776
		/*
		 * Don't want to pull so many tasks that a group would go idle.
5777 5778
		 * Except of course for the group_imb case, since then we might
		 * have to drop below capacity to reach cpu-load equilibrium.
5779
		 */
J
Joonsoo Kim 已提交
5780 5781
		load_above_capacity =
			(busiest->sum_nr_running - busiest->group_capacity);
5782

5783
		load_above_capacity *= (SCHED_LOAD_SCALE * SCHED_POWER_SCALE);
5784
		load_above_capacity /= busiest->group_power;
5785 5786 5787 5788 5789 5790 5791 5792 5793 5794
	}

	/*
	 * 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,
	 * we also don't want to reduce the group load below the group capacity
	 * (so that we can implement power-savings policies etc). Thus we look
	 * for the minimum possible imbalance.
	 */
5795
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
5796 5797

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
5798
	env->imbalance = min(
5799 5800
		max_pull * busiest->group_power,
		(sds->avg_load - local->avg_load) * local->group_power
J
Joonsoo Kim 已提交
5801
	) / SCHED_POWER_SCALE;
5802 5803 5804

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
5805
	 * there is no guarantee that any tasks will be moved so we'll have
5806 5807 5808
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
5809
	if (env->imbalance < busiest->load_per_task)
5810
		return fix_small_imbalance(env, sds);
5811
}
5812

5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
 * if there is an imbalance. If there isn't an imbalance, and
 * the user has opted for power-savings, it returns a group whose
 * CPUs can be put to idle by rebalancing those tasks elsewhere, if
 * such a group exists.
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
5825
 * @env: The load balancing environment.
5826
 *
5827
 * Return:	- The busiest group if imbalance exists.
5828 5829 5830 5831
 *		- If no imbalance and user has opted for power-savings balance,
 *		   return the least loaded group whose CPUs can be
 *		   put to idle by rebalancing its tasks onto our group.
 */
J
Joonsoo Kim 已提交
5832
static struct sched_group *find_busiest_group(struct lb_env *env)
5833
{
J
Joonsoo Kim 已提交
5834
	struct sg_lb_stats *local, *busiest;
5835 5836
	struct sd_lb_stats sds;

5837
	init_sd_lb_stats(&sds);
5838 5839 5840 5841 5842

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
5843
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
5844 5845
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
5846

5847 5848
	if ((env->idle == CPU_IDLE || env->idle == CPU_NEWLY_IDLE) &&
	    check_asym_packing(env, &sds))
5849 5850
		return sds.busiest;

5851
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
5852
	if (!sds.busiest || busiest->sum_nr_running == 0)
5853 5854
		goto out_balanced;

5855
	sds.avg_load = (SCHED_POWER_SCALE * sds.total_load) / sds.total_pwr;
5856

P
Peter Zijlstra 已提交
5857 5858
	/*
	 * If the busiest group is imbalanced the below checks don't
5859
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
5860 5861
	 * isn't true due to cpus_allowed constraints and the like.
	 */
J
Joonsoo Kim 已提交
5862
	if (busiest->group_imb)
P
Peter Zijlstra 已提交
5863 5864
		goto force_balance;

5865
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
J
Joonsoo Kim 已提交
5866 5867
	if (env->idle == CPU_NEWLY_IDLE && local->group_has_capacity &&
	    !busiest->group_has_capacity)
5868 5869
		goto force_balance;

5870 5871 5872 5873
	/*
	 * If the local group is more busy than the selected busiest group
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
5874
	if (local->avg_load >= busiest->avg_load)
5875 5876
		goto out_balanced;

5877 5878 5879 5880
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
5881
	if (local->avg_load >= sds.avg_load)
5882 5883
		goto out_balanced;

5884
	if (env->idle == CPU_IDLE) {
5885 5886 5887 5888 5889 5890
		/*
		 * This cpu is idle. If the busiest group load doesn't
		 * have more tasks than the number of available cpu's and
		 * there is no imbalance between this and busiest group
		 * wrt to idle cpu's, it is balanced.
		 */
J
Joonsoo Kim 已提交
5891 5892
		if ((local->idle_cpus < busiest->idle_cpus) &&
		    busiest->sum_nr_running <= busiest->group_weight)
5893
			goto out_balanced;
5894 5895 5896 5897 5898
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
5899 5900
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
5901
			goto out_balanced;
5902
	}
5903

5904
force_balance:
5905
	/* Looks like there is an imbalance. Compute it */
5906
	calculate_imbalance(env, &sds);
5907 5908 5909
	return sds.busiest;

out_balanced:
5910
	env->imbalance = 0;
5911 5912 5913 5914 5915 5916
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
5917
static struct rq *find_busiest_queue(struct lb_env *env,
5918
				     struct sched_group *group)
5919 5920
{
	struct rq *busiest = NULL, *rq;
5921
	unsigned long busiest_load = 0, busiest_power = 1;
5922 5923
	int i;

5924
	for_each_cpu_and(i, sched_group_cpus(group), env->cpus) {
5925 5926 5927 5928 5929
		unsigned long power, capacity, wl;
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
5930

5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954
		/*
		 * 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;

		power = power_of(i);
		capacity = DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE);
5955
		if (!capacity)
5956
			capacity = fix_small_capacity(env->sd, group);
5957

5958
		wl = weighted_cpuload(i);
5959

5960 5961 5962 5963
		/*
		 * When comparing with imbalance, use weighted_cpuload()
		 * which is not scaled with the cpu power.
		 */
5964
		if (capacity && rq->nr_running == 1 && wl > env->imbalance)
5965 5966
			continue;

5967 5968 5969 5970 5971
		/*
		 * For the load comparisons with the other cpu's, consider
		 * the weighted_cpuload() scaled with the cpu power, so that
		 * the load can be moved away from the cpu that is potentially
		 * running at a lower capacity.
5972 5973 5974 5975 5976
		 *
		 * Thus we're looking for max(wl_i / power_i), crosswise
		 * multiplication to rid ourselves of the division works out
		 * to: wl_i * power_j > wl_j * power_i;  where j is our
		 * previous maximum.
5977
		 */
5978 5979 5980
		if (wl * busiest_power > busiest_load * power) {
			busiest_load = wl;
			busiest_power = power;
5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994
			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

/* Working cpumask for load_balance and load_balance_newidle. */
5995
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
5996

5997
static int need_active_balance(struct lb_env *env)
5998
{
5999 6000 6001
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
6002 6003 6004 6005 6006 6007

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
		 * higher numbered CPUs in order to pack all tasks in the
		 * lowest numbered CPUs.
		 */
6008
		if ((sd->flags & SD_ASYM_PACKING) && env->src_cpu > env->dst_cpu)
6009
			return 1;
6010 6011 6012 6013 6014
	}

	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

6015 6016
static int active_load_balance_cpu_stop(void *data);

6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047
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.
	 */
6048
	return balance_cpu == env->dst_cpu;
6049 6050
}

6051 6052 6053 6054 6055 6056
/*
 * 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,
6057
			int *continue_balancing)
6058
{
6059
	int ld_moved, cur_ld_moved, active_balance = 0;
6060
	struct sched_domain *sd_parent = sd->parent;
6061 6062 6063
	struct sched_group *group;
	struct rq *busiest;
	unsigned long flags;
6064
	struct cpumask *cpus = __get_cpu_var(load_balance_mask);
6065

6066 6067
	struct lb_env env = {
		.sd		= sd,
6068 6069
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
6070
		.dst_grpmask    = sched_group_cpus(sd->groups),
6071
		.idle		= idle,
6072
		.loop_break	= sched_nr_migrate_break,
6073
		.cpus		= cpus,
6074
		.fbq_type	= all,
6075 6076
	};

6077 6078 6079 6080
	/*
	 * For NEWLY_IDLE load_balancing, we don't need to consider
	 * other cpus in our group
	 */
6081
	if (idle == CPU_NEWLY_IDLE)
6082 6083
		env.dst_grpmask = NULL;

6084 6085 6086 6087 6088
	cpumask_copy(cpus, cpu_active_mask);

	schedstat_inc(sd, lb_count[idle]);

redo:
6089 6090
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
6091
		goto out_balanced;
6092
	}
6093

6094
	group = find_busiest_group(&env);
6095 6096 6097 6098 6099
	if (!group) {
		schedstat_inc(sd, lb_nobusyg[idle]);
		goto out_balanced;
	}

6100
	busiest = find_busiest_queue(&env, group);
6101 6102 6103 6104 6105
	if (!busiest) {
		schedstat_inc(sd, lb_nobusyq[idle]);
		goto out_balanced;
	}

6106
	BUG_ON(busiest == env.dst_rq);
6107

6108
	schedstat_add(sd, lb_imbalance[idle], env.imbalance);
6109 6110 6111 6112 6113 6114 6115 6116 6117

	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.
		 */
6118
		env.flags |= LBF_ALL_PINNED;
6119 6120 6121
		env.src_cpu   = busiest->cpu;
		env.src_rq    = busiest;
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
6122

6123
more_balance:
6124
		local_irq_save(flags);
6125
		double_rq_lock(env.dst_rq, busiest);
6126 6127 6128 6129 6130 6131 6132

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
		cur_ld_moved = move_tasks(&env);
		ld_moved += cur_ld_moved;
6133
		double_rq_unlock(env.dst_rq, busiest);
6134 6135 6136 6137 6138
		local_irq_restore(flags);

		/*
		 * some other cpu did the load balance for us.
		 */
6139 6140 6141
		if (cur_ld_moved && env.dst_cpu != smp_processor_id())
			resched_cpu(env.dst_cpu);

6142 6143 6144 6145 6146
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165
		/*
		 * 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.
		 */
6166
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
6167

6168 6169 6170
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

6171
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
6172
			env.dst_cpu	 = env.new_dst_cpu;
6173
			env.flags	&= ~LBF_DST_PINNED;
6174 6175
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
6176

6177 6178 6179 6180 6181 6182
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
6183

6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
			int *group_imbalance = &sd_parent->groups->sgp->imbalance;

			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0) {
				*group_imbalance = 1;
			} else if (*group_imbalance)
				*group_imbalance = 0;
		}

6196
		/* All tasks on this runqueue were pinned by CPU affinity */
6197
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
6198
			cpumask_clear_cpu(cpu_of(busiest), cpus);
6199 6200 6201
			if (!cpumask_empty(cpus)) {
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
6202
				goto redo;
6203
			}
6204 6205 6206 6207 6208 6209
			goto out_balanced;
		}
	}

	if (!ld_moved) {
		schedstat_inc(sd, lb_failed[idle]);
6210 6211 6212 6213 6214 6215 6216 6217
		/*
		 * 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++;
6218

6219
		if (need_active_balance(&env)) {
6220 6221
			raw_spin_lock_irqsave(&busiest->lock, flags);

6222 6223 6224
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
6225 6226
			 */
			if (!cpumask_test_cpu(this_cpu,
6227
					tsk_cpus_allowed(busiest->curr))) {
6228 6229
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
6230
				env.flags |= LBF_ALL_PINNED;
6231 6232 6233
				goto out_one_pinned;
			}

6234 6235 6236 6237 6238
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
6239 6240 6241 6242 6243 6244
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
6245

6246
			if (active_balance) {
6247 6248 6249
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
6250
			}
6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283

			/*
			 * We've kicked active balancing, reset the failure
			 * counter.
			 */
			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
		 * move_tasks).
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
	schedstat_inc(sd, lb_balanced[idle]);

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
6284
	if (((env.flags & LBF_ALL_PINNED) &&
6285
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
6286 6287 6288
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

6289
	ld_moved = 0;
6290 6291 6292 6293 6294 6295 6296 6297
out:
	return ld_moved;
}

/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
6298
void idle_balance(int this_cpu, struct rq *this_rq)
6299 6300 6301 6302
{
	struct sched_domain *sd;
	int pulled_task = 0;
	unsigned long next_balance = jiffies + HZ;
6303
	u64 curr_cost = 0;
6304

6305
	this_rq->idle_stamp = rq_clock(this_rq);
6306 6307 6308 6309

	if (this_rq->avg_idle < sysctl_sched_migration_cost)
		return;

6310 6311 6312 6313 6314
	/*
	 * Drop the rq->lock, but keep IRQ/preempt disabled.
	 */
	raw_spin_unlock(&this_rq->lock);

6315
	update_blocked_averages(this_cpu);
6316
	rcu_read_lock();
6317 6318
	for_each_domain(this_cpu, sd) {
		unsigned long interval;
6319
		int continue_balancing = 1;
6320
		u64 t0, domain_cost;
6321 6322 6323 6324

		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

6325 6326 6327
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost)
			break;

6328
		if (sd->flags & SD_BALANCE_NEWIDLE) {
6329 6330
			t0 = sched_clock_cpu(this_cpu);

6331
			/* If we've pulled tasks over stop searching: */
6332
			pulled_task = load_balance(this_cpu, this_rq,
6333 6334
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
6335 6336 6337 6338 6339 6340

			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;
6341
		}
6342 6343 6344 6345

		interval = msecs_to_jiffies(sd->balance_interval);
		if (time_after(next_balance, sd->last_balance + interval))
			next_balance = sd->last_balance + interval;
N
Nikhil Rao 已提交
6346 6347
		if (pulled_task) {
			this_rq->idle_stamp = 0;
6348
			break;
N
Nikhil Rao 已提交
6349
		}
6350
	}
6351
	rcu_read_unlock();
6352 6353 6354

	raw_spin_lock(&this_rq->lock);

6355 6356 6357 6358 6359 6360 6361
	if (pulled_task || time_after(jiffies, this_rq->next_balance)) {
		/*
		 * We are going idle. next_balance may be set based on
		 * a busy processor. So reset next_balance.
		 */
		this_rq->next_balance = next_balance;
	}
6362 6363 6364

	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;
6365 6366 6367
}

/*
6368 6369 6370 6371
 * 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.
6372
 */
6373
static int active_load_balance_cpu_stop(void *data)
6374
{
6375 6376
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
6377
	int target_cpu = busiest_rq->push_cpu;
6378
	struct rq *target_rq = cpu_rq(target_cpu);
6379
	struct sched_domain *sd;
6380 6381 6382 6383 6384 6385 6386

	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;
6387 6388 6389

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
6390
		goto out_unlock;
6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402

	/*
	 * 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);

	/* move a task from busiest_rq to target_rq */
	double_lock_balance(busiest_rq, target_rq);

	/* Search for an sd spanning us and the target CPU. */
6403
	rcu_read_lock();
6404 6405 6406 6407 6408 6409 6410
	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)) {
6411 6412
		struct lb_env env = {
			.sd		= sd,
6413 6414 6415 6416
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
6417 6418 6419
			.idle		= CPU_IDLE,
		};

6420 6421
		schedstat_inc(sd, alb_count);

6422
		if (move_one_task(&env))
6423 6424 6425 6426
			schedstat_inc(sd, alb_pushed);
		else
			schedstat_inc(sd, alb_failed);
	}
6427
	rcu_read_unlock();
6428
	double_unlock_balance(busiest_rq, target_rq);
6429 6430 6431 6432
out_unlock:
	busiest_rq->active_balance = 0;
	raw_spin_unlock_irq(&busiest_rq->lock);
	return 0;
6433 6434
}

6435
#ifdef CONFIG_NO_HZ_COMMON
6436 6437 6438 6439 6440 6441
/*
 * 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.
 */
6442
static struct {
6443
	cpumask_var_t idle_cpus_mask;
6444
	atomic_t nr_cpus;
6445 6446
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
6447

6448
static inline int find_new_ilb(int call_cpu)
6449
{
6450
	int ilb = cpumask_first(nohz.idle_cpus_mask);
6451

6452 6453 6454 6455
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
6456 6457
}

6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468
/*
 * 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).
 */
static void nohz_balancer_kick(int cpu)
{
	int ilb_cpu;

	nohz.next_balance++;

6469
	ilb_cpu = find_new_ilb(cpu);
6470

6471 6472
	if (ilb_cpu >= nr_cpu_ids)
		return;
6473

6474
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
6475 6476 6477 6478 6479 6480 6481 6482
		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);
6483 6484 6485
	return;
}

6486
static inline void nohz_balance_exit_idle(int cpu)
6487 6488 6489 6490 6491 6492 6493 6494
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
		cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
		atomic_dec(&nohz.nr_cpus);
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

6495 6496 6497 6498 6499
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;

	rcu_read_lock();
N
Nathan Zimmer 已提交
6500
	sd = rcu_dereference_check_sched_domain(this_rq()->sd);
V
Vincent Guittot 已提交
6501 6502 6503 6504 6505 6506

	if (!sd || !sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 0;

	for (; sd; sd = sd->parent)
6507
		atomic_inc(&sd->groups->sgp->nr_busy_cpus);
V
Vincent Guittot 已提交
6508
unlock:
6509 6510 6511 6512 6513 6514 6515 6516
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;

	rcu_read_lock();
N
Nathan Zimmer 已提交
6517
	sd = rcu_dereference_check_sched_domain(this_rq()->sd);
V
Vincent Guittot 已提交
6518 6519 6520 6521 6522 6523

	if (!sd || sd->nohz_idle)
		goto unlock;
	sd->nohz_idle = 1;

	for (; sd; sd = sd->parent)
6524
		atomic_dec(&sd->groups->sgp->nr_busy_cpus);
V
Vincent Guittot 已提交
6525
unlock:
6526 6527 6528
	rcu_read_unlock();
}

6529
/*
6530
 * This routine will record that the cpu is going idle with tick stopped.
6531
 * This info will be used in performing idle load balancing in the future.
6532
 */
6533
void nohz_balance_enter_idle(int cpu)
6534
{
6535 6536 6537 6538 6539 6540
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

6541 6542
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
6543

6544 6545 6546
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
6547
}
6548

6549
static int sched_ilb_notifier(struct notifier_block *nfb,
6550 6551 6552 6553
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DYING:
6554
		nohz_balance_exit_idle(smp_processor_id());
6555 6556 6557 6558 6559
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}
6560 6561 6562 6563
#endif

static DEFINE_SPINLOCK(balancing);

6564 6565 6566 6567
/*
 * 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.
 */
6568
void update_max_interval(void)
6569 6570 6571 6572
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

6573 6574 6575 6576
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
6577
 * Balancing parameters are set up in init_sched_domains.
6578 6579 6580
 */
static void rebalance_domains(int cpu, enum cpu_idle_type idle)
{
6581
	int continue_balancing = 1;
6582 6583
	struct rq *rq = cpu_rq(cpu);
	unsigned long interval;
6584
	struct sched_domain *sd;
6585 6586 6587
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
6588 6589
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
6590

6591
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
6592

6593
	rcu_read_lock();
6594
	for_each_domain(cpu, sd) {
6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606
		/*
		 * 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;

6607 6608 6609
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620
		/*
		 * 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;
		}

6621 6622 6623 6624 6625 6626
		interval = sd->balance_interval;
		if (idle != CPU_IDLE)
			interval *= sd->busy_factor;

		/* scale ms to jiffies */
		interval = msecs_to_jiffies(interval);
6627
		interval = clamp(interval, 1UL, max_load_balance_interval);
6628 6629 6630 6631 6632 6633 6634 6635 6636

		need_serialize = sd->flags & SD_SERIALIZE;

		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
6637
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
6638
				/*
6639
				 * The LBF_DST_PINNED logic could have changed
6640 6641
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
6642
				 */
6643
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
6644 6645 6646 6647 6648 6649 6650 6651 6652 6653
			}
			sd->last_balance = jiffies;
		}
		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;
		}
6654 6655
	}
	if (need_decay) {
6656
		/*
6657 6658
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
6659
		 */
6660 6661
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
6662
	}
6663
	rcu_read_unlock();
6664 6665 6666 6667 6668 6669 6670 6671 6672 6673

	/*
	 * next_balance will be updated only when there is a need.
	 * When the cpu is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance))
		rq->next_balance = next_balance;
}

6674
#ifdef CONFIG_NO_HZ_COMMON
6675
/*
6676
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
6677 6678
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
6679 6680 6681 6682 6683 6684
static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle)
{
	struct rq *this_rq = cpu_rq(this_cpu);
	struct rq *rq;
	int balance_cpu;

6685 6686 6687
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
6688 6689

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
6690
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
6691 6692 6693 6694 6695 6696 6697
			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.
		 */
6698
		if (need_resched())
6699 6700
			break;

V
Vincent Guittot 已提交
6701 6702 6703 6704 6705 6706
		rq = cpu_rq(balance_cpu);

		raw_spin_lock_irq(&rq->lock);
		update_rq_clock(rq);
		update_idle_cpu_load(rq);
		raw_spin_unlock_irq(&rq->lock);
6707 6708 6709 6710 6711 6712 6713

		rebalance_domains(balance_cpu, CPU_IDLE);

		if (time_after(this_rq->next_balance, rq->next_balance))
			this_rq->next_balance = rq->next_balance;
	}
	nohz.next_balance = this_rq->next_balance;
6714 6715
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
6716 6717 6718
}

/*
6719 6720 6721 6722 6723 6724 6725
 * Current heuristic for kicking the idle load balancer in the presence
 * of an idle cpu is the system.
 *   - This rq has more than one task.
 *   - At any scheduler domain level, this cpu's scheduler group has multiple
 *     busy cpu's exceeding the group's power.
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
6726 6727 6728 6729
 */
static inline int nohz_kick_needed(struct rq *rq, int cpu)
{
	unsigned long now = jiffies;
6730
	struct sched_domain *sd;
6731

6732
	if (unlikely(idle_cpu(cpu)))
6733 6734
		return 0;

6735 6736 6737 6738
       /*
	* 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.
	*/
6739
	set_cpu_sd_state_busy();
6740
	nohz_balance_exit_idle(cpu);
6741 6742 6743 6744 6745 6746 6747

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
		return 0;
6748 6749

	if (time_before(now, nohz.next_balance))
6750 6751
		return 0;

6752 6753
	if (rq->nr_running >= 2)
		goto need_kick;
6754

6755
	rcu_read_lock();
6756 6757 6758 6759
	for_each_domain(cpu, sd) {
		struct sched_group *sg = sd->groups;
		struct sched_group_power *sgp = sg->sgp;
		int nr_busy = atomic_read(&sgp->nr_busy_cpus);
6760

6761
		if (sd->flags & SD_SHARE_PKG_RESOURCES && nr_busy > 1)
6762
			goto need_kick_unlock;
6763 6764 6765 6766

		if (sd->flags & SD_ASYM_PACKING && nr_busy != sg->group_weight
		    && (cpumask_first_and(nohz.idle_cpus_mask,
					  sched_domain_span(sd)) < cpu))
6767
			goto need_kick_unlock;
6768 6769 6770

		if (!(sd->flags & (SD_SHARE_PKG_RESOURCES | SD_ASYM_PACKING)))
			break;
6771
	}
6772
	rcu_read_unlock();
6773
	return 0;
6774 6775 6776

need_kick_unlock:
	rcu_read_unlock();
6777 6778
need_kick:
	return 1;
6779 6780 6781 6782 6783 6784 6785 6786 6787
}
#else
static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle) { }
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
6788 6789 6790 6791
static void run_rebalance_domains(struct softirq_action *h)
{
	int this_cpu = smp_processor_id();
	struct rq *this_rq = cpu_rq(this_cpu);
6792
	enum cpu_idle_type idle = this_rq->idle_balance ?
6793 6794 6795 6796 6797
						CPU_IDLE : CPU_NOT_IDLE;

	rebalance_domains(this_cpu, idle);

	/*
6798
	 * If this cpu has a pending nohz_balance_kick, then do the
6799 6800 6801
	 * balancing on behalf of the other idle cpus whose ticks are
	 * stopped.
	 */
6802
	nohz_idle_balance(this_cpu, idle);
6803 6804 6805 6806
}

static inline int on_null_domain(int cpu)
{
6807
	return !rcu_dereference_sched(cpu_rq(cpu)->sd);
6808 6809 6810 6811 6812
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
6813
void trigger_load_balance(struct rq *rq, int cpu)
6814 6815 6816 6817 6818
{
	/* Don't need to rebalance while attached to NULL domain */
	if (time_after_eq(jiffies, rq->next_balance) &&
	    likely(!on_null_domain(cpu)))
		raise_softirq(SCHED_SOFTIRQ);
6819
#ifdef CONFIG_NO_HZ_COMMON
6820
	if (nohz_kick_needed(rq, cpu) && likely(!on_null_domain(cpu)))
6821 6822
		nohz_balancer_kick(cpu);
#endif
6823 6824
}

6825 6826 6827 6828 6829 6830 6831 6832
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
6833 6834 6835

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
6836 6837
}

6838
#endif /* CONFIG_SMP */
6839

6840 6841 6842
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
6843
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
6844 6845 6846 6847 6848 6849
{
	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
6850
		entity_tick(cfs_rq, se, queued);
6851
	}
6852

6853
	if (numabalancing_enabled)
6854
		task_tick_numa(rq, curr);
6855

6856
	update_rq_runnable_avg(rq, 1);
6857 6858 6859
}

/*
P
Peter Zijlstra 已提交
6860 6861 6862
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
6863
 */
P
Peter Zijlstra 已提交
6864
static void task_fork_fair(struct task_struct *p)
6865
{
6866 6867
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
6868
	int this_cpu = smp_processor_id();
P
Peter Zijlstra 已提交
6869 6870 6871
	struct rq *rq = this_rq();
	unsigned long flags;

6872
	raw_spin_lock_irqsave(&rq->lock, flags);
6873

6874 6875
	update_rq_clock(rq);

6876 6877 6878
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;

6879 6880 6881 6882 6883 6884 6885 6886 6887
	/*
	 * Not only the cpu but also the task_group of the parent might have
	 * been changed after parent->se.parent,cfs_rq were copied to
	 * child->se.parent,cfs_rq. So call __set_task_cpu() to make those
	 * of child point to valid ones.
	 */
	rcu_read_lock();
	__set_task_cpu(p, this_cpu);
	rcu_read_unlock();
6888

6889
	update_curr(cfs_rq);
P
Peter Zijlstra 已提交
6890

6891 6892
	if (curr)
		se->vruntime = curr->vruntime;
6893
	place_entity(cfs_rq, se, 1);
6894

P
Peter Zijlstra 已提交
6895
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
6896
		/*
6897 6898 6899
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
6900
		swap(curr->vruntime, se->vruntime);
6901
		resched_task(rq->curr);
6902
	}
6903

6904 6905
	se->vruntime -= cfs_rq->min_vruntime;

6906
	raw_spin_unlock_irqrestore(&rq->lock, flags);
6907 6908
}

6909 6910 6911 6912
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
6913 6914
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
6915
{
P
Peter Zijlstra 已提交
6916 6917 6918
	if (!p->se.on_rq)
		return;

6919 6920 6921 6922 6923
	/*
	 * 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
6924
	if (rq->curr == p) {
6925 6926 6927
		if (p->prio > oldprio)
			resched_task(rq->curr);
	} else
6928
		check_preempt_curr(rq, p, 0);
6929 6930
}

P
Peter Zijlstra 已提交
6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	/*
	 * Ensure the task's vruntime is normalized, so that when its
	 * switched back to the fair class the enqueue_entity(.flags=0) will
	 * do the right thing.
	 *
	 * If it was on_rq, then the dequeue_entity(.flags=0) will already
	 * have normalized the vruntime, if it was !on_rq, then only when
	 * the task is sleeping will it still have non-normalized vruntime.
	 */
	if (!se->on_rq && p->state != TASK_RUNNING) {
		/*
		 * 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;
	}
6953

6954
#ifdef CONFIG_SMP
6955 6956 6957 6958 6959
	/*
	* Remove our load from contribution when we leave sched_fair
	* and ensure we don't carry in an old decay_count if we
	* switch back.
	*/
6960 6961 6962
	if (se->avg.decay_count) {
		__synchronize_entity_decay(se);
		subtract_blocked_load_contrib(cfs_rq, se->avg.load_avg_contrib);
6963 6964
	}
#endif
P
Peter Zijlstra 已提交
6965 6966
}

6967 6968 6969
/*
 * We switched to the sched_fair class.
 */
P
Peter Zijlstra 已提交
6970
static void switched_to_fair(struct rq *rq, struct task_struct *p)
6971
{
P
Peter Zijlstra 已提交
6972 6973 6974
	if (!p->se.on_rq)
		return;

6975 6976 6977 6978 6979
	/*
	 * 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.
	 */
P
Peter Zijlstra 已提交
6980
	if (rq->curr == p)
6981 6982
		resched_task(rq->curr);
	else
6983
		check_preempt_curr(rq, p, 0);
6984 6985
}

6986 6987 6988 6989 6990 6991 6992 6993 6994
/* 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;

6995 6996 6997 6998 6999 7000 7001
	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);
	}
7002 7003
}

7004 7005 7006 7007 7008 7009 7010
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
7011
#ifdef CONFIG_SMP
7012
	atomic64_set(&cfs_rq->decay_counter, 1);
7013
	atomic_long_set(&cfs_rq->removed_load, 0);
7014
#endif
7015 7016
}

P
Peter Zijlstra 已提交
7017
#ifdef CONFIG_FAIR_GROUP_SCHED
7018
static void task_move_group_fair(struct task_struct *p, int on_rq)
P
Peter Zijlstra 已提交
7019
{
7020
	struct cfs_rq *cfs_rq;
7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033
	/*
	 * If the task was not on the rq at the time of this cgroup movement
	 * it must have been asleep, sleeping tasks keep their ->vruntime
	 * absolute on their old rq until wakeup (needed for the fair sleeper
	 * bonus in place_entity()).
	 *
	 * If it was on the rq, we've just 'preempted' it, which does convert
	 * ->vruntime to a relative base.
	 *
	 * Make sure both cases convert their relative position when migrating
	 * to another cgroup's rq. This does somewhat interfere with the
	 * fair sleeper stuff for the first placement, but who cares.
	 */
7034 7035 7036 7037 7038 7039
	/*
	 * When !on_rq, vruntime of the task has usually NOT been normalized.
	 * But there are some cases where it has already been normalized:
	 *
	 * - Moving a forked child which is waiting for being woken up by
	 *   wake_up_new_task().
7040 7041
	 * - Moving a task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
7042 7043 7044 7045
	 *
	 * To prevent boost or penalty in the new cfs_rq caused by delta
	 * min_vruntime between the two cfs_rqs, we skip vruntime adjustment.
	 */
7046
	if (!on_rq && (!p->se.sum_exec_runtime || p->state == TASK_WAKING))
7047 7048
		on_rq = 1;

7049 7050 7051
	if (!on_rq)
		p->se.vruntime -= cfs_rq_of(&p->se)->min_vruntime;
	set_task_rq(p, task_cpu(p));
7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064
	if (!on_rq) {
		cfs_rq = cfs_rq_of(&p->se);
		p->se.vruntime += cfs_rq->min_vruntime;
#ifdef CONFIG_SMP
		/*
		 * migrate_task_rq_fair() will have removed our previous
		 * contribution, but we must synchronize for ongoing future
		 * decay.
		 */
		p->se.avg.decay_count = atomic64_read(&cfs_rq->decay_counter);
		cfs_rq->blocked_load_avg += p->se.avg.load_avg_contrib;
#endif
	}
P
Peter Zijlstra 已提交
7065
}
7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131 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 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194

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]);
		if (tg->se)
			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 cfs_rq *cfs_rq;
	struct sched_entity *se;
	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]);
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

void unregister_fair_sched_group(struct task_group *tg, int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

	/*
	* 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)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
	list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

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;

	if (!parent)
		se->cfs_rq = &rq->cfs;
	else
		se->cfs_rq = parent->my_q;

	se->my_q = cfs_rq;
	update_load_set(&se->load, 0);
	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);
7195 7196 7197

		/* Possible calls to update_curr() need rq clock */
		update_rq_clock(rq);
7198
		for_each_sched_entity(se)
7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219
			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;
}

void unregister_fair_sched_group(struct task_group *tg, int cpu) { }

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
7220

7221
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
7222 7223 7224 7225 7226 7227 7228 7229 7230
{
	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)
7231
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
7232 7233 7234 7235

	return rr_interval;
}

7236 7237 7238
/*
 * All the scheduling class methods:
 */
7239
const struct sched_class fair_sched_class = {
7240
	.next			= &idle_sched_class,
7241 7242 7243
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
7244
	.yield_to_task		= yield_to_task_fair,
7245

I
Ingo Molnar 已提交
7246
	.check_preempt_curr	= check_preempt_wakeup,
7247 7248 7249 7250

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

7251
#ifdef CONFIG_SMP
L
Li Zefan 已提交
7252
	.select_task_rq		= select_task_rq_fair,
7253
	.migrate_task_rq	= migrate_task_rq_fair,
7254

7255 7256
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
7257 7258

	.task_waking		= task_waking_fair,
7259
#endif
7260

7261
	.set_curr_task          = set_curr_task_fair,
7262
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
7263
	.task_fork		= task_fork_fair,
7264 7265

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
7266
	.switched_from		= switched_from_fair,
7267
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
7268

7269 7270
	.get_rr_interval	= get_rr_interval_fair,

P
Peter Zijlstra 已提交
7271
#ifdef CONFIG_FAIR_GROUP_SCHED
7272
	.task_move_group	= task_move_group_fair,
P
Peter Zijlstra 已提交
7273
#endif
7274 7275 7276
};

#ifdef CONFIG_SCHED_DEBUG
7277
void print_cfs_stats(struct seq_file *m, int cpu)
7278 7279 7280
{
	struct cfs_rq *cfs_rq;

7281
	rcu_read_lock();
7282
	for_each_leaf_cfs_rq(cpu_rq(cpu), cfs_rq)
7283
		print_cfs_rq(m, cpu, cfs_rq);
7284
	rcu_read_unlock();
7285 7286
}
#endif
7287 7288 7289 7290 7291 7292

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

7293
#ifdef CONFIG_NO_HZ_COMMON
7294
	nohz.next_balance = jiffies;
7295
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
7296
	cpu_notifier(sched_ilb_notifier, 0);
7297 7298 7299 7300
#endif
#endif /* SMP */

}