fair.c 257.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
/*
 * Completely Fair Scheduling (CFS) Class (SCHED_NORMAL/SCHED_BATCH)
 *
 *  Copyright (C) 2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 *
 *  Interactivity improvements by Mike Galbraith
 *  (C) 2007 Mike Galbraith <efault@gmx.de>
 *
 *  Various enhancements by Dmitry Adamushko.
 *  (C) 2007 Dmitry Adamushko <dmitry.adamushko@gmail.com>
 *
 *  Group scheduling enhancements by Srivatsa Vaddagiri
 *  Copyright IBM Corporation, 2007
 *  Author: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
 *
 *  Scaled math optimizations by Thomas Gleixner
 *  Copyright (C) 2007, Thomas Gleixner <tglx@linutronix.de>
18 19
 *
 *  Adaptive scheduling granularity, math enhancements by Peter Zijlstra
20
 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
21 22
 */

23
#include <linux/sched/mm.h>
24 25
#include <linux/sched/topology.h>

26
#include <linux/latencytop.h>
27
#include <linux/cpumask.h>
28
#include <linux/cpuidle.h>
29 30 31
#include <linux/slab.h>
#include <linux/profile.h>
#include <linux/interrupt.h>
32
#include <linux/mempolicy.h>
33
#include <linux/migrate.h>
34
#include <linux/task_work.h>
35 36 37 38

#include <trace/events/sched.h>

#include "sched.h"
A
Arjan van de Ven 已提交
39

40
/*
41
 * Targeted preemption latency for CPU-bound tasks:
42
 *
43
 * NOTE: this latency value is not the same as the concept of
I
Ingo Molnar 已提交
44 45 46
 * 'timeslice length' - timeslices in CFS are of variable length
 * and have no persistent notion like in traditional, time-slice
 * based scheduling concepts.
47
 *
I
Ingo Molnar 已提交
48 49
 * (to see the precise effective timeslice length of your workload,
 *  run vmstat and monitor the context-switches (cs) field)
50 51
 *
 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
52
 */
53 54
unsigned int sysctl_sched_latency			= 6000000ULL;
unsigned int normalized_sysctl_sched_latency		= 6000000ULL;
55

56 57 58 59
/*
 * The initial- and re-scaling of tunables is configurable
 *
 * Options are:
60 61 62 63 64 65
 *
 *   SCHED_TUNABLESCALING_NONE - unscaled, always *1
 *   SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
 *   SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
 *
 * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
66
 */
67
enum sched_tunable_scaling sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG;
68

69
/*
70
 * Minimal preemption granularity for CPU-bound tasks:
71
 *
72
 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
73
 */
74 75
unsigned int sysctl_sched_min_granularity		= 750000ULL;
unsigned int normalized_sysctl_sched_min_granularity	= 750000ULL;
76 77

/*
78
 * This value is kept at sysctl_sched_latency/sysctl_sched_min_granularity
79
 */
80
static unsigned int sched_nr_latency = 8;
81 82

/*
83
 * After fork, child runs first. If set to 0 (default) then
84
 * parent will (try to) run first.
85
 */
86
unsigned int sysctl_sched_child_runs_first __read_mostly;
87 88 89 90 91 92 93

/*
 * SCHED_OTHER wake-up granularity.
 *
 * This option delays the preemption effects of decoupled workloads
 * and reduces their over-scheduling. Synchronous workloads will still
 * have immediate wakeup/sleep latencies.
94 95
 *
 * (default: 1 msec * (1 + ilog(ncpus)), units: nanoseconds)
96
 */
97 98
unsigned int sysctl_sched_wakeup_granularity		= 1000000UL;
unsigned int normalized_sysctl_sched_wakeup_granularity	= 1000000UL;
99

100
const_debug unsigned int sysctl_sched_migration_cost	= 500000UL;
101

T
Tim Chen 已提交
102 103 104 105 106 107 108 109 110 111
#ifdef CONFIG_SMP
/*
 * For asym packing, by default the lower numbered cpu has higher priority.
 */
int __weak arch_asym_cpu_priority(int cpu)
{
	return -cpu;
}
#endif

112 113 114 115 116 117 118 119 120
#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.
 *
121 122 123
 * (default: 5 msec, units: microseconds)
 */
unsigned int sysctl_sched_cfs_bandwidth_slice		= 5000UL;
124 125
#endif

126 127
/*
 * The margin used when comparing utilization with CPU capacity:
128
 * util * margin < capacity * 1024
129 130
 *
 * (default: ~20%)
131
 */
132
unsigned int capacity_margin				= 1280;
133

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
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;
}

152 153 154 155 156 157 158 159 160
/*
 * 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:
 */
161
static unsigned int get_update_sysctl_factor(void)
162
{
163
	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
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
	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();
}

199
#define WMULT_CONST	(~0U)
200 201
#define WMULT_SHIFT	32

202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
static void __update_inv_weight(struct load_weight *lw)
{
	unsigned long w;

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

	w = scale_load_down(lw->weight);

	if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST))
		lw->inv_weight = 1;
	else if (unlikely(!w))
		lw->inv_weight = WMULT_CONST;
	else
		lw->inv_weight = WMULT_CONST / w;
}
218 219

/*
220 221 222 223
 * delta_exec * weight / lw.weight
 *   OR
 * (delta_exec * (weight * lw->inv_weight)) >> WMULT_SHIFT
 *
224
 * Either weight := NICE_0_LOAD and lw \e sched_prio_to_wmult[], in which case
225 226 227 228 229
 * we're guaranteed shift stays positive because inv_weight is guaranteed to
 * fit 32 bits, and NICE_0_LOAD gives another 10 bits; therefore shift >= 22.
 *
 * Or, weight =< lw.weight (because lw.weight is the runqueue weight), thus
 * weight/lw.weight <= 1, and therefore our shift will also be positive.
230
 */
231
static u64 __calc_delta(u64 delta_exec, unsigned long weight, struct load_weight *lw)
232
{
233 234
	u64 fact = scale_load_down(weight);
	int shift = WMULT_SHIFT;
235

236
	__update_inv_weight(lw);
237

238 239 240 241 242
	if (unlikely(fact >> 32)) {
		while (fact >> 32) {
			fact >>= 1;
			shift--;
		}
243 244
	}

245 246
	/* hint to use a 32x32->64 mul */
	fact = (u64)(u32)fact * lw->inv_weight;
247

248 249 250 251
	while (fact >> 32) {
		fact >>= 1;
		shift--;
	}
252

253
	return mul_u64_u32_shr(delta_exec, fact, shift);
254 255 256 257
}


const struct sched_class fair_sched_class;
258

259 260 261 262
/**************************************************************
 * CFS operations on generic schedulable entities:
 */

263
#ifdef CONFIG_FAIR_GROUP_SCHED
264

265
/* cpu runqueue to which this cfs_rq is attached */
266 267
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
268
	return cfs_rq->rq;
269 270
}

271 272
/* An entity is a task if it doesn't "own" a runqueue */
#define entity_is_task(se)	(!se->my_q)
273

274 275
static inline struct task_struct *task_of(struct sched_entity *se)
{
276
	SCHED_WARN_ON(!entity_is_task(se));
277 278 279
	return container_of(se, struct task_struct, se);
}

P
Peter Zijlstra 已提交
280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
/* 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;
}

301 302 303
static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
{
	if (!cfs_rq->on_list) {
304 305
		struct rq *rq = rq_of(cfs_rq);
		int cpu = cpu_of(rq);
306 307 308
		/*
		 * Ensure we either appear before our parent (if already
		 * enqueued) or force our parent to appear after us when it is
309 310 311 312 313
		 * enqueued. The fact that we always enqueue bottom-up
		 * reduces this to two cases and a special case for the root
		 * cfs_rq. Furthermore, it also means that we will always reset
		 * tmp_alone_branch either when the branch is connected
		 * to a tree or when we reach the beg of the tree
314 315
		 */
		if (cfs_rq->tg->parent &&
316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
		    cfs_rq->tg->parent->cfs_rq[cpu]->on_list) {
			/*
			 * If parent is already on the list, we add the child
			 * just before. Thanks to circular linked property of
			 * the list, this means to put the child at the tail
			 * of the list that starts by parent.
			 */
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
				&(cfs_rq->tg->parent->cfs_rq[cpu]->leaf_cfs_rq_list));
			/*
			 * The branch is now connected to its tree so we can
			 * reset tmp_alone_branch to the beginning of the
			 * list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else if (!cfs_rq->tg->parent) {
			/*
			 * cfs rq without parent should be put
			 * at the tail of the list.
			 */
336
			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
				&rq->leaf_cfs_rq_list);
			/*
			 * We have reach the beg of a tree so we can reset
			 * tmp_alone_branch to the beginning of the list.
			 */
			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
		} else {
			/*
			 * The parent has not already been added so we want to
			 * make sure that it will be put after us.
			 * tmp_alone_branch points to the beg of the branch
			 * where we will add parent.
			 */
			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
				rq->tmp_alone_branch);
			/*
			 * update tmp_alone_branch to points to the new beg
			 * of the branch
			 */
			rq->tmp_alone_branch = &cfs_rq->leaf_cfs_rq_list;
357
		}
358 359 360 361 362 363 364 365 366 367 368 369 370

		cfs_rq->on_list = 1;
	}
}

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

P
Peter Zijlstra 已提交
371
/* Iterate thr' all leaf cfs_rq's on a runqueue */
372 373 374
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)			\
	list_for_each_entry_safe(cfs_rq, pos, &rq->leaf_cfs_rq_list,	\
				 leaf_cfs_rq_list)
P
Peter Zijlstra 已提交
375 376

/* Do the two (enqueued) entities belong to the same group ? */
P
Peter Zijlstra 已提交
377
static inline struct cfs_rq *
P
Peter Zijlstra 已提交
378 379 380
is_same_group(struct sched_entity *se, struct sched_entity *pse)
{
	if (se->cfs_rq == pse->cfs_rq)
P
Peter Zijlstra 已提交
381
		return se->cfs_rq;
P
Peter Zijlstra 已提交
382

P
Peter Zijlstra 已提交
383
	return NULL;
P
Peter Zijlstra 已提交
384 385 386 387 388 389 390
}

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

391 392 393 394 395 396 397 398 399 400 401 402 403
static void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
	int se_depth, pse_depth;

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

	/* First walk up until both entities are at same depth */
P
Peter Zijlstra 已提交
404 405
	se_depth = (*se)->depth;
	pse_depth = (*pse)->depth;
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422

	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);
	}
}

423 424 425 426 427 428
#else	/* !CONFIG_FAIR_GROUP_SCHED */

static inline struct task_struct *task_of(struct sched_entity *se)
{
	return container_of(se, struct task_struct, se);
}
429

430 431 432
static inline struct rq *rq_of(struct cfs_rq *cfs_rq)
{
	return container_of(cfs_rq, struct rq, cfs);
433 434 435 436
}

#define entity_is_task(se)	1

P
Peter Zijlstra 已提交
437 438
#define for_each_sched_entity(se) \
		for (; se; se = NULL)
439

P
Peter Zijlstra 已提交
440
static inline struct cfs_rq *task_cfs_rq(struct task_struct *p)
441
{
P
Peter Zijlstra 已提交
442
	return &task_rq(p)->cfs;
443 444
}

P
Peter Zijlstra 已提交
445 446 447 448 449 450 451 452 453 454 455 456 457 458
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;
}

459 460 461 462 463 464 465 466
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)
{
}

467 468
#define for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos)	\
		for (cfs_rq = &rq->cfs, pos = NULL; cfs_rq; cfs_rq = pos)
P
Peter Zijlstra 已提交
469 470 471 472 473 474

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

475 476 477 478 479
static inline void
find_matching_se(struct sched_entity **se, struct sched_entity **pse)
{
}

P
Peter Zijlstra 已提交
480 481
#endif	/* CONFIG_FAIR_GROUP_SCHED */

482
static __always_inline
483
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec);
484 485 486 487 488

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

489
static inline u64 max_vruntime(u64 max_vruntime, u64 vruntime)
490
{
491
	s64 delta = (s64)(vruntime - max_vruntime);
492
	if (delta > 0)
493
		max_vruntime = vruntime;
494

495
	return max_vruntime;
496 497
}

498
static inline u64 min_vruntime(u64 min_vruntime, u64 vruntime)
P
Peter Zijlstra 已提交
499 500 501 502 503 504 505 506
{
	s64 delta = (s64)(vruntime - min_vruntime);
	if (delta < 0)
		min_vruntime = vruntime;

	return min_vruntime;
}

507 508 509 510 511 512
static inline int entity_before(struct sched_entity *a,
				struct sched_entity *b)
{
	return (s64)(a->vruntime - b->vruntime) < 0;
}

513 514
static void update_min_vruntime(struct cfs_rq *cfs_rq)
{
515
	struct sched_entity *curr = cfs_rq->curr;
516
	struct rb_node *leftmost = rb_first_cached(&cfs_rq->tasks_timeline);
517

518 519
	u64 vruntime = cfs_rq->min_vruntime;

520 521 522 523 524 525
	if (curr) {
		if (curr->on_rq)
			vruntime = curr->vruntime;
		else
			curr = NULL;
	}
526

527 528 529
	if (leftmost) { /* non-empty tree */
		struct sched_entity *se;
		se = rb_entry(leftmost, struct sched_entity, run_node);
530

531
		if (!curr)
532 533 534 535 536
			vruntime = se->vruntime;
		else
			vruntime = min_vruntime(vruntime, se->vruntime);
	}

537
	/* ensure we never gain time by being placed backwards. */
538
	cfs_rq->min_vruntime = max_vruntime(cfs_rq->min_vruntime, vruntime);
539 540 541 542
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
543 544
}

545 546 547
/*
 * Enqueue an entity into the rb-tree:
 */
548
static void __enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
549
{
550
	struct rb_node **link = &cfs_rq->tasks_timeline.rb_root.rb_node;
551 552
	struct rb_node *parent = NULL;
	struct sched_entity *entry;
553
	bool leftmost = true;
554 555 556 557 558 559 560 561 562 563 564

	/*
	 * 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.
		 */
565
		if (entity_before(se, entry)) {
566 567 568
			link = &parent->rb_left;
		} else {
			link = &parent->rb_right;
569
			leftmost = false;
570 571 572 573
		}
	}

	rb_link_node(&se->run_node, parent, link);
574 575
	rb_insert_color_cached(&se->run_node,
			       &cfs_rq->tasks_timeline, leftmost);
576 577
}

578
static void __dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
579
{
580
	rb_erase_cached(&se->run_node, &cfs_rq->tasks_timeline);
581 582
}

583
struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq)
584
{
585
	struct rb_node *left = rb_first_cached(&cfs_rq->tasks_timeline);
586 587 588 589 590

	if (!left)
		return NULL;

	return rb_entry(left, struct sched_entity, run_node);
591 592
}

593 594 595 596 597 598 599 600 601 602 603
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
604
struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
605
{
606
	struct rb_node *last = rb_last(&cfs_rq->tasks_timeline.rb_root);
607

608 609
	if (!last)
		return NULL;
610 611

	return rb_entry(last, struct sched_entity, run_node);
612 613
}

614 615 616 617
/**************************************************************
 * Scheduling class statistics methods:
 */

618
int sched_proc_update_handler(struct ctl_table *table, int write,
619
		void __user *buffer, size_t *lenp,
620 621
		loff_t *ppos)
{
622
	int ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
623
	unsigned int factor = get_update_sysctl_factor();
624 625 626 627 628 629 630

	if (ret || !write)
		return ret;

	sched_nr_latency = DIV_ROUND_UP(sysctl_sched_latency,
					sysctl_sched_min_granularity);

631 632 633 634 635 636 637
#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

638 639 640
	return 0;
}
#endif
641

642
/*
643
 * delta /= w
644
 */
645
static inline u64 calc_delta_fair(u64 delta, struct sched_entity *se)
646
{
647
	if (unlikely(se->load.weight != NICE_0_LOAD))
648
		delta = __calc_delta(delta, NICE_0_LOAD, &se->load);
649 650 651 652

	return delta;
}

653 654 655
/*
 * The idea is to set a period in which each task runs once.
 *
656
 * When there are too many tasks (sched_nr_latency) we have to stretch
657 658 659 660
 * this period because otherwise the slices get too small.
 *
 * p = (nr <= nl) ? l : l*nr/nl
 */
661 662
static u64 __sched_period(unsigned long nr_running)
{
663 664 665 666
	if (unlikely(nr_running > sched_nr_latency))
		return nr_running * sysctl_sched_min_granularity;
	else
		return sysctl_sched_latency;
667 668
}

669 670 671 672
/*
 * We calculate the wall-time slice from the period by taking a part
 * proportional to the weight.
 *
673
 * s = p*P[w/rw]
674
 */
P
Peter Zijlstra 已提交
675
static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
676
{
M
Mike Galbraith 已提交
677
	u64 slice = __sched_period(cfs_rq->nr_running + !se->on_rq);
678

M
Mike Galbraith 已提交
679
	for_each_sched_entity(se) {
L
Lin Ming 已提交
680
		struct load_weight *load;
681
		struct load_weight lw;
L
Lin Ming 已提交
682 683 684

		cfs_rq = cfs_rq_of(se);
		load = &cfs_rq->load;
685

M
Mike Galbraith 已提交
686
		if (unlikely(!se->on_rq)) {
687
			lw = cfs_rq->load;
M
Mike Galbraith 已提交
688 689 690 691

			update_load_add(&lw, se->load.weight);
			load = &lw;
		}
692
		slice = __calc_delta(slice, se->load.weight, load);
M
Mike Galbraith 已提交
693 694
	}
	return slice;
695 696
}

697
/*
A
Andrei Epure 已提交
698
 * We calculate the vruntime slice of a to-be-inserted task.
699
 *
700
 * vs = s/w
701
 */
702
static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
P
Peter Zijlstra 已提交
703
{
704
	return calc_delta_fair(sched_slice(cfs_rq, se), se);
705 706
}

707
#ifdef CONFIG_SMP
708 709 710

#include "sched-pelt.h"

711
static int select_idle_sibling(struct task_struct *p, int prev_cpu, int cpu);
712 713
static unsigned long task_h_load(struct task_struct *p);

714 715
/* Give new sched_entity start runnable values to heavy its load in infant time */
void init_entity_runnable_average(struct sched_entity *se)
716
{
717
	struct sched_avg *sa = &se->avg;
718

719 720
	memset(sa, 0, sizeof(*sa));

721 722 723 724 725 726 727
	/*
	 * Tasks are intialized with full load to be seen as heavy tasks until
	 * they get a chance to stabilize to their real load level.
	 * Group entities are intialized with zero load to reflect the fact that
	 * nothing has been attached to the task group yet.
	 */
	if (entity_is_task(se))
728 729
		sa->runnable_load_avg = sa->load_avg = scale_load_down(se->load.weight);

730 731
	se->runnable_weight = se->load.weight;

732
	/* when this task enqueue'ed, it will contribute to its cfs_rq's load_avg */
733
}
734

735
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
736
static void attach_entity_cfs_rq(struct sched_entity *se);
737

738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
/*
 * With new tasks being created, their initial util_avgs are extrapolated
 * based on the cfs_rq's current util_avg:
 *
 *   util_avg = cfs_rq->util_avg / (cfs_rq->load_avg + 1) * se.load.weight
 *
 * However, in many cases, the above util_avg does not give a desired
 * value. Moreover, the sum of the util_avgs may be divergent, such
 * as when the series is a harmonic series.
 *
 * To solve this problem, we also cap the util_avg of successive tasks to
 * only 1/2 of the left utilization budget:
 *
 *   util_avg_cap = (1024 - cfs_rq->avg.util_avg) / 2^n
 *
 * where n denotes the nth task.
 *
 * For example, a simplest series from the beginning would be like:
 *
 *  task  util_avg: 512, 256, 128,  64,  32,   16,    8, ...
 * cfs_rq util_avg: 512, 768, 896, 960, 992, 1008, 1016, ...
 *
 * Finally, that extrapolated util_avg is clamped to the cap (util_avg_cap)
 * if util_avg > util_avg_cap.
 */
void post_init_entity_util_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	struct sched_avg *sa = &se->avg;
767
	long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
768 769 770 771 772 773 774 775 776 777 778 779

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

			if (sa->util_avg > cap)
				sa->util_avg = cap;
		} else {
			sa->util_avg = cap;
		}
	}
780 781 782 783 784 785 786

	if (entity_is_task(se)) {
		struct task_struct *p = task_of(se);
		if (p->sched_class != &fair_sched_class) {
			/*
			 * For !fair tasks do:
			 *
787
			update_cfs_rq_load_avg(now, cfs_rq);
788 789 790 791 792 793
			attach_entity_load_avg(cfs_rq, se);
			switched_from_fair(rq, p);
			 *
			 * such that the next switched_to_fair() has the
			 * expected state.
			 */
794
			se->avg.last_update_time = cfs_rq_clock_task(cfs_rq);
795 796 797 798
			return;
		}
	}

799
	attach_entity_cfs_rq(se);
800 801
}

802
#else /* !CONFIG_SMP */
803
void init_entity_runnable_average(struct sched_entity *se)
804 805
{
}
806 807 808
void post_init_entity_util_avg(struct sched_entity *se)
{
}
809 810 811
static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
{
}
812
#endif /* CONFIG_SMP */
813

814
/*
815
 * Update the current task's runtime statistics.
816
 */
817
static void update_curr(struct cfs_rq *cfs_rq)
818
{
819
	struct sched_entity *curr = cfs_rq->curr;
820
	u64 now = rq_clock_task(rq_of(cfs_rq));
821
	u64 delta_exec;
822 823 824 825

	if (unlikely(!curr))
		return;

826 827
	delta_exec = now - curr->exec_start;
	if (unlikely((s64)delta_exec <= 0))
P
Peter Zijlstra 已提交
828
		return;
829

I
Ingo Molnar 已提交
830
	curr->exec_start = now;
831

832 833 834 835
	schedstat_set(curr->statistics.exec_max,
		      max(delta_exec, curr->statistics.exec_max));

	curr->sum_exec_runtime += delta_exec;
836
	schedstat_add(cfs_rq->exec_clock, delta_exec);
837 838 839 840

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

841 842 843
	if (entity_is_task(curr)) {
		struct task_struct *curtask = task_of(curr);

844
		trace_sched_stat_runtime(curtask, delta_exec, curr->vruntime);
845
		cpuacct_charge(curtask, delta_exec);
846
		account_group_exec_runtime(curtask, delta_exec);
847
	}
848 849

	account_cfs_rq_runtime(cfs_rq, delta_exec);
850 851
}

852 853 854 855 856
static void update_curr_fair(struct rq *rq)
{
	update_curr(cfs_rq_of(&rq->curr->se));
}

857
static inline void
858
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
859
{
860 861 862 863 864 865 866
	u64 wait_start, prev_wait_start;

	if (!schedstat_enabled())
		return;

	wait_start = rq_clock(rq_of(cfs_rq));
	prev_wait_start = schedstat_val(se->statistics.wait_start);
867 868

	if (entity_is_task(se) && task_on_rq_migrating(task_of(se)) &&
869 870
	    likely(wait_start > prev_wait_start))
		wait_start -= prev_wait_start;
871

872
	schedstat_set(se->statistics.wait_start, wait_start);
873 874
}

875
static inline void
876 877 878
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *p;
879 880
	u64 delta;

881 882 883 884
	if (!schedstat_enabled())
		return;

	delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
885 886 887 888 889 890 891 892 893

	if (entity_is_task(se)) {
		p = task_of(se);
		if (task_on_rq_migrating(p)) {
			/*
			 * Preserve migrating task's wait time so wait_start
			 * time stamp can be adjusted to accumulate wait time
			 * prior to migration.
			 */
894
			schedstat_set(se->statistics.wait_start, delta);
895 896 897 898 899
			return;
		}
		trace_sched_stat_wait(p, delta);
	}

900 901 902 903 904
	schedstat_set(se->statistics.wait_max,
		      max(schedstat_val(se->statistics.wait_max), delta));
	schedstat_inc(se->statistics.wait_count);
	schedstat_add(se->statistics.wait_sum, delta);
	schedstat_set(se->statistics.wait_start, 0);
905 906
}

907
static inline void
908 909 910
update_stats_enqueue_sleeper(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	struct task_struct *tsk = NULL;
911 912 913 914 915 916 917
	u64 sleep_start, block_start;

	if (!schedstat_enabled())
		return;

	sleep_start = schedstat_val(se->statistics.sleep_start);
	block_start = schedstat_val(se->statistics.block_start);
918 919 920 921

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

922 923
	if (sleep_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - sleep_start;
924 925 926 927

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

928 929
		if (unlikely(delta > schedstat_val(se->statistics.sleep_max)))
			schedstat_set(se->statistics.sleep_max, delta);
930

931 932
		schedstat_set(se->statistics.sleep_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
933 934 935 936 937 938

		if (tsk) {
			account_scheduler_latency(tsk, delta >> 10, 1);
			trace_sched_stat_sleep(tsk, delta);
		}
	}
939 940
	if (block_start) {
		u64 delta = rq_clock(rq_of(cfs_rq)) - block_start;
941 942 943 944

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

945 946
		if (unlikely(delta > schedstat_val(se->statistics.block_max)))
			schedstat_set(se->statistics.block_max, delta);
947

948 949
		schedstat_set(se->statistics.block_start, 0);
		schedstat_add(se->statistics.sum_sleep_runtime, delta);
950 951 952

		if (tsk) {
			if (tsk->in_iowait) {
953 954
				schedstat_add(se->statistics.iowait_sum, delta);
				schedstat_inc(se->statistics.iowait_count);
955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
				trace_sched_stat_iowait(tsk, delta);
			}

			trace_sched_stat_blocked(tsk, delta);

			/*
			 * Blocking time is in units of nanosecs, so shift by
			 * 20 to get a milliseconds-range estimation of the
			 * amount of time that the task spent sleeping:
			 */
			if (unlikely(prof_on == SLEEP_PROFILING)) {
				profile_hits(SLEEP_PROFILING,
						(void *)get_wchan(tsk),
						delta >> 20);
			}
			account_scheduler_latency(tsk, delta >> 10, 0);
		}
	}
973 974
}

975 976 977
/*
 * Task is being enqueued - update stats:
 */
978
static inline void
979
update_stats_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
980
{
981 982 983
	if (!schedstat_enabled())
		return;

984 985 986 987
	/*
	 * Are we enqueueing a waiting task? (for current tasks
	 * a dequeue/enqueue event is a NOP)
	 */
988
	if (se != cfs_rq->curr)
989
		update_stats_wait_start(cfs_rq, se);
990 991 992

	if (flags & ENQUEUE_WAKEUP)
		update_stats_enqueue_sleeper(cfs_rq, se);
993 994 995
}

static inline void
996
update_stats_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
997
{
998 999 1000 1001

	if (!schedstat_enabled())
		return;

1002 1003 1004 1005
	/*
	 * Mark the end of the wait period if dequeueing a
	 * waiting task:
	 */
1006
	if (se != cfs_rq->curr)
1007
		update_stats_wait_end(cfs_rq, se);
1008

1009 1010
	if ((flags & DEQUEUE_SLEEP) && entity_is_task(se)) {
		struct task_struct *tsk = task_of(se);
1011

1012 1013 1014 1015 1016 1017
		if (tsk->state & TASK_INTERRUPTIBLE)
			schedstat_set(se->statistics.sleep_start,
				      rq_clock(rq_of(cfs_rq)));
		if (tsk->state & TASK_UNINTERRUPTIBLE)
			schedstat_set(se->statistics.block_start,
				      rq_clock(rq_of(cfs_rq)));
1018 1019 1020
	}
}

1021 1022 1023 1024
/*
 * We are picking a new current task - update its stats:
 */
static inline void
1025
update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
1026 1027 1028 1029
{
	/*
	 * We are starting a new run period:
	 */
1030
	se->exec_start = rq_clock_task(rq_of(cfs_rq));
1031 1032 1033 1034 1035 1036
}

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

1037 1038
#ifdef CONFIG_NUMA_BALANCING
/*
1039 1040 1041
 * 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.
1042
 */
1043 1044
unsigned int sysctl_numa_balancing_scan_period_min = 1000;
unsigned int sysctl_numa_balancing_scan_period_max = 60000;
1045 1046 1047

/* Portion of address space to scan in MB */
unsigned int sysctl_numa_balancing_scan_size = 256;
1048

1049 1050 1051
/* Scan @scan_size MB every @scan_period after an initial @scan_delay in ms */
unsigned int sysctl_numa_balancing_scan_delay = 1000;

1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
struct numa_group {
	atomic_t refcount;

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

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

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

1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098
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)
{
1099
	unsigned int scan_size = READ_ONCE(sysctl_numa_balancing_scan_size);
1100 1101 1102
	unsigned int scan, floor;
	unsigned int windows = 1;

1103 1104
	if (scan_size < MAX_SCAN_WINDOW)
		windows = MAX_SCAN_WINDOW / scan_size;
1105 1106 1107 1108 1109 1110
	floor = 1000 / windows;

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

1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
static unsigned int task_scan_start(struct task_struct *p)
{
	unsigned long smin = task_scan_min(p);
	unsigned long period = smin;

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

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

	return max(smin, period);
}

1130 1131
static unsigned int task_scan_max(struct task_struct *p)
{
1132 1133
	unsigned long smin = task_scan_min(p);
	unsigned long smax;
1134 1135 1136

	/* Watch for min being lower than max due to floor calculations */
	smax = sysctl_numa_balancing_scan_period_max / task_nr_scan_windows(p);
1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151

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

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

		smax = max(smax, period);
	}

1152 1153 1154
	return max(smin, smax);
}

1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
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));
}

1167 1168 1169 1170 1171 1172 1173 1174 1175
/* Shared or private faults. */
#define NR_NUMA_HINT_FAULT_TYPES 2

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

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

1176 1177 1178 1179 1180
pid_t task_numa_group_id(struct task_struct *p)
{
	return p->numa_group ? p->numa_group->gid : 0;
}

1181 1182 1183 1184 1185 1186 1187
/*
 * The averaged statistics, shared & private, memory & cpu,
 * occupy the first half of the array. The second half of the
 * array is for current counters, which are averaged into the
 * first set by task_numa_placement.
 */
static inline int task_faults_idx(enum numa_faults_stats s, int nid, int priv)
1188
{
1189
	return NR_NUMA_HINT_FAULT_TYPES * (s * nr_node_ids + nid) + priv;
1190 1191 1192 1193
}

static inline unsigned long task_faults(struct task_struct *p, int nid)
{
1194
	if (!p->numa_faults)
1195 1196
		return 0;

1197 1198
	return p->numa_faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_faults[task_faults_idx(NUMA_MEM, nid, 1)];
1199 1200
}

1201 1202 1203 1204 1205
static inline unsigned long group_faults(struct task_struct *p, int nid)
{
	if (!p->numa_group)
		return 0;

1206 1207
	return p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 0)] +
		p->numa_group->faults[task_faults_idx(NUMA_MEM, nid, 1)];
1208 1209
}

1210 1211
static inline unsigned long group_faults_cpu(struct numa_group *group, int nid)
{
1212 1213
	return group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 0)] +
		group->faults_cpu[task_faults_idx(NUMA_MEM, nid, 1)];
1214 1215
}

1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
static inline unsigned long group_faults_priv(struct numa_group *ng)
{
	unsigned long faults = 0;
	int node;

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

	return faults;
}

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

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

	return faults;
}

1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
/*
 * A node triggering more than 1/3 as many NUMA faults as the maximum is
 * considered part of a numa group's pseudo-interleaving set. Migrations
 * between these nodes are slowed down, to allow things to settle down.
 */
#define ACTIVE_NODE_FRACTION 3

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

1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316
/* Handle placement on systems where not all nodes are directly connected. */
static unsigned long score_nearby_nodes(struct task_struct *p, int nid,
					int maxdist, bool task)
{
	unsigned long score = 0;
	int node;

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

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

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

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

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

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

		score += faults;
	}

	return score;
}

1317 1318 1319 1320 1321 1322
/*
 * 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.
 */
1323 1324
static inline unsigned long task_weight(struct task_struct *p, int nid,
					int dist)
1325
{
1326
	unsigned long faults, total_faults;
1327

1328
	if (!p->numa_faults)
1329 1330 1331 1332 1333 1334 1335
		return 0;

	total_faults = p->total_numa_faults;

	if (!total_faults)
		return 0;

1336
	faults = task_faults(p, nid);
1337 1338
	faults += score_nearby_nodes(p, nid, dist, true);

1339
	return 1000 * faults / total_faults;
1340 1341
}

1342 1343
static inline unsigned long group_weight(struct task_struct *p, int nid,
					 int dist)
1344
{
1345 1346 1347 1348 1349 1350 1351 1352
	unsigned long faults, total_faults;

	if (!p->numa_group)
		return 0;

	total_faults = p->numa_group->total_faults;

	if (!total_faults)
1353 1354
		return 0;

1355
	faults = group_faults(p, nid);
1356 1357
	faults += score_nearby_nodes(p, nid, dist, false);

1358
	return 1000 * faults / total_faults;
1359 1360
}

1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400
bool should_numa_migrate_memory(struct task_struct *p, struct page * page,
				int src_nid, int dst_cpu)
{
	struct numa_group *ng = p->numa_group;
	int dst_nid = cpu_to_node(dst_cpu);
	int last_cpupid, this_cpupid;

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

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

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

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

	/*
1401 1402
	 * Destination node is much more heavily used than the source
	 * node? Allow migration.
1403
	 */
1404 1405
	if (group_faults_cpu(ng, dst_nid) > group_faults_cpu(ng, src_nid) *
					ACTIVE_NODE_FRACTION)
1406 1407 1408
		return true;

	/*
1409 1410 1411 1412 1413 1414
	 * Distribute memory according to CPU & memory use on each node,
	 * with 3/4 hysteresis to avoid unnecessary memory migrations:
	 *
	 * faults_cpu(dst)   3   faults_cpu(src)
	 * --------------- * - > ---------------
	 * faults_mem(dst)   4   faults_mem(src)
1415
	 */
1416 1417
	return group_faults_cpu(ng, dst_nid) * group_faults(p, src_nid) * 3 >
	       group_faults_cpu(ng, src_nid) * group_faults(p, dst_nid) * 4;
1418 1419
}

1420
static unsigned long weighted_cpuload(struct rq *rq);
1421 1422
static unsigned long source_load(int cpu, int type);
static unsigned long target_load(int cpu, int type);
1423
static unsigned long capacity_of(int cpu);
1424

1425
/* Cached statistics for all CPUs within a node */
1426
struct numa_stats {
1427
	unsigned long nr_running;
1428
	unsigned long load;
1429 1430

	/* Total compute capacity of CPUs on a node */
1431
	unsigned long compute_capacity;
1432 1433

	/* Approximate capacity in terms of runnable tasks on a node */
1434
	unsigned long task_capacity;
1435
	int has_free_capacity;
1436
};
1437

1438 1439 1440 1441 1442
/*
 * XXX borrowed from update_sg_lb_stats
 */
static void update_numa_stats(struct numa_stats *ns, int nid)
{
1443 1444
	int smt, cpu, cpus = 0;
	unsigned long capacity;
1445 1446 1447 1448 1449 1450

	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;
1451
		ns->load += weighted_cpuload(rq);
1452
		ns->compute_capacity += capacity_of(cpu);
1453 1454

		cpus++;
1455 1456
	}

1457 1458 1459 1460 1461
	/*
	 * If we raced with hotplug and there are no CPUs left in our mask
	 * the @ns structure is NULL'ed and task_numa_compare() will
	 * not find this node attractive.
	 *
1462 1463
	 * We'll either bail at !has_free_capacity, or we'll detect a huge
	 * imbalance and bail there.
1464 1465 1466 1467
	 */
	if (!cpus)
		return;

1468 1469 1470 1471 1472 1473
	/* smt := ceil(cpus / capacity), assumes: 1 < smt_power < 2 */
	smt = DIV_ROUND_UP(SCHED_CAPACITY_SCALE * cpus, ns->compute_capacity);
	capacity = cpus / smt; /* cores */

	ns->task_capacity = min_t(unsigned, capacity,
		DIV_ROUND_CLOSEST(ns->compute_capacity, SCHED_CAPACITY_SCALE));
1474
	ns->has_free_capacity = (ns->nr_running < ns->task_capacity);
1475 1476
}

1477 1478
struct task_numa_env {
	struct task_struct *p;
1479

1480 1481
	int src_cpu, src_nid;
	int dst_cpu, dst_nid;
1482

1483
	struct numa_stats src_stats, dst_stats;
1484

1485
	int imbalance_pct;
1486
	int dist;
1487 1488 1489

	struct task_struct *best_task;
	long best_imp;
1490 1491 1492
	int best_cpu;
};

1493 1494 1495 1496 1497
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);
1498 1499
	if (p)
		get_task_struct(p);
1500 1501 1502 1503 1504 1505

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

1506
static bool load_too_imbalanced(long src_load, long dst_load,
1507 1508
				struct task_numa_env *env)
{
1509 1510
	long imb, old_imb;
	long orig_src_load, orig_dst_load;
1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521
	long src_capacity, dst_capacity;

	/*
	 * The load is corrected for the CPU capacity available on each node.
	 *
	 * src_load        dst_load
	 * ------------ vs ---------
	 * src_capacity    dst_capacity
	 */
	src_capacity = env->src_stats.compute_capacity;
	dst_capacity = env->dst_stats.compute_capacity;
1522 1523

	/* We care about the slope of the imbalance, not the direction. */
1524 1525
	if (dst_load < src_load)
		swap(dst_load, src_load);
1526 1527

	/* Is the difference below the threshold? */
1528 1529
	imb = dst_load * src_capacity * 100 -
	      src_load * dst_capacity * env->imbalance_pct;
1530 1531 1532 1533 1534
	if (imb <= 0)
		return false;

	/*
	 * The imbalance is above the allowed threshold.
1535
	 * Compare it with the old imbalance.
1536
	 */
1537
	orig_src_load = env->src_stats.load;
1538
	orig_dst_load = env->dst_stats.load;
1539

1540 1541
	if (orig_dst_load < orig_src_load)
		swap(orig_dst_load, orig_src_load);
1542

1543 1544 1545 1546 1547
	old_imb = orig_dst_load * src_capacity * 100 -
		  orig_src_load * dst_capacity * env->imbalance_pct;

	/* Would this change make things worse? */
	return (imb > old_imb);
1548 1549
}

1550 1551 1552 1553 1554 1555
/*
 * 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
 */
1556 1557
static void task_numa_compare(struct task_numa_env *env,
			      long taskimp, long groupimp)
1558 1559 1560 1561
{
	struct rq *src_rq = cpu_rq(env->src_cpu);
	struct rq *dst_rq = cpu_rq(env->dst_cpu);
	struct task_struct *cur;
1562
	long src_load, dst_load;
1563
	long load;
1564
	long imp = env->p->numa_group ? groupimp : taskimp;
1565
	long moveimp = imp;
1566
	int dist = env->dist;
1567 1568

	rcu_read_lock();
1569 1570
	cur = task_rcu_dereference(&dst_rq->curr);
	if (cur && ((cur->flags & PF_EXITING) || is_idle_task(cur)))
1571 1572
		cur = NULL;

1573 1574 1575 1576 1577 1578 1579
	/*
	 * Because we have preemption enabled we can get migrated around and
	 * end try selecting ourselves (current == env->p) as a swap candidate.
	 */
	if (cur == env->p)
		goto unlock;

1580 1581 1582 1583 1584 1585 1586 1587 1588
	/*
	 * "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 */
1589
		if (!cpumask_test_cpu(env->src_cpu, &cur->cpus_allowed))
1590 1591
			goto unlock;

1592 1593
		/*
		 * If dst and source tasks are in the same NUMA group, or not
1594
		 * in any group then look only at task weights.
1595
		 */
1596
		if (cur->numa_group == env->p->numa_group) {
1597 1598
			imp = taskimp + task_weight(cur, env->src_nid, dist) -
			      task_weight(cur, env->dst_nid, dist);
1599 1600 1601 1602 1603 1604
			/*
			 * Add some hysteresis to prevent swapping the
			 * tasks within a group over tiny differences.
			 */
			if (cur->numa_group)
				imp -= imp/16;
1605
		} else {
1606 1607 1608 1609 1610 1611
			/*
			 * Compare the group weights. If a task is all by
			 * itself (not part of a group), use the task weight
			 * instead.
			 */
			if (cur->numa_group)
1612 1613
				imp += group_weight(cur, env->src_nid, dist) -
				       group_weight(cur, env->dst_nid, dist);
1614
			else
1615 1616
				imp += task_weight(cur, env->src_nid, dist) -
				       task_weight(cur, env->dst_nid, dist);
1617
		}
1618 1619
	}

1620
	if (imp <= env->best_imp && moveimp <= env->best_imp)
1621 1622 1623 1624
		goto unlock;

	if (!cur) {
		/* Is there capacity at our destination? */
1625
		if (env->src_stats.nr_running <= env->src_stats.task_capacity &&
1626
		    !env->dst_stats.has_free_capacity)
1627 1628 1629 1630 1631 1632
			goto unlock;

		goto balance;
	}

	/* Balance doesn't matter much if we're running a task per cpu */
1633 1634
	if (imp > env->best_imp && src_rq->nr_running == 1 &&
			dst_rq->nr_running == 1)
1635 1636 1637 1638 1639 1640
		goto assign;

	/*
	 * In the overloaded case, try and keep the load balanced.
	 */
balance:
1641 1642 1643
	load = task_h_load(env->p);
	dst_load = env->dst_stats.load + load;
	src_load = env->src_stats.load - load;
1644

1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
	if (moveimp > imp && moveimp > env->best_imp) {
		/*
		 * If the improvement from just moving env->p direction is
		 * better than swapping tasks around, check if a move is
		 * possible. Store a slightly smaller score than moveimp,
		 * so an actually idle CPU will win.
		 */
		if (!load_too_imbalanced(src_load, dst_load, env)) {
			imp = moveimp - 1;
			cur = NULL;
			goto assign;
		}
	}

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

1662
	if (cur) {
1663 1664 1665
		load = task_h_load(cur);
		dst_load -= load;
		src_load += load;
1666 1667
	}

1668
	if (load_too_imbalanced(src_load, dst_load, env))
1669 1670
		goto unlock;

1671 1672 1673 1674
	/*
	 * One idle CPU per node is evaluated for a task numa move.
	 * Call select_idle_sibling to maybe find a better one.
	 */
1675 1676 1677 1678 1679 1680
	if (!cur) {
		/*
		 * select_idle_siblings() uses an per-cpu cpumask that
		 * can be used from IRQ context.
		 */
		local_irq_disable();
1681 1682
		env->dst_cpu = select_idle_sibling(env->p, env->src_cpu,
						   env->dst_cpu);
1683 1684
		local_irq_enable();
	}
1685

1686 1687 1688 1689 1690 1691
assign:
	task_numa_assign(env, cur, imp);
unlock:
	rcu_read_unlock();
}

1692 1693
static void task_numa_find_cpu(struct task_numa_env *env,
				long taskimp, long groupimp)
1694 1695 1696 1697 1698
{
	int cpu;

	for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
		/* Skip this CPU if the source task cannot migrate */
1699
		if (!cpumask_test_cpu(cpu, &env->p->cpus_allowed))
1700 1701 1702
			continue;

		env->dst_cpu = cpu;
1703
		task_numa_compare(env, taskimp, groupimp);
1704 1705 1706
	}
}

1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723
/* Only move tasks to a NUMA node less busy than the current node. */
static bool numa_has_capacity(struct task_numa_env *env)
{
	struct numa_stats *src = &env->src_stats;
	struct numa_stats *dst = &env->dst_stats;

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

	/*
	 * Only consider a task move if the source has a higher load
	 * than the destination, corrected for CPU capacity on each node.
	 *
	 *      src->load                dst->load
	 * --------------------- vs ---------------------
	 * src->compute_capacity    dst->compute_capacity
	 */
1724 1725 1726
	if (src->load * dst->compute_capacity * env->imbalance_pct >

	    dst->load * src->compute_capacity * 100)
1727 1728 1729 1730 1731
		return true;

	return false;
}

1732 1733 1734 1735
static int task_numa_migrate(struct task_struct *p)
{
	struct task_numa_env env = {
		.p = p,
1736

1737
		.src_cpu = task_cpu(p),
I
Ingo Molnar 已提交
1738
		.src_nid = task_node(p),
1739 1740 1741 1742 1743

		.imbalance_pct = 112,

		.best_task = NULL,
		.best_imp = 0,
1744
		.best_cpu = -1,
1745 1746
	};
	struct sched_domain *sd;
1747
	unsigned long taskweight, groupweight;
1748
	int nid, ret, dist;
1749
	long taskimp, groupimp;
1750

1751
	/*
1752 1753 1754 1755 1756 1757
	 * 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.
1758 1759
	 */
	rcu_read_lock();
1760
	sd = rcu_dereference(per_cpu(sd_numa, env.src_cpu));
1761 1762
	if (sd)
		env.imbalance_pct = 100 + (sd->imbalance_pct - 100) / 2;
1763 1764
	rcu_read_unlock();

1765 1766 1767 1768 1769 1770 1771
	/*
	 * Cpusets can break the scheduler domain tree into smaller
	 * balance domains, some of which do not cross NUMA boundaries.
	 * Tasks that are "trapped" in such domains cannot be migrated
	 * elsewhere, so there is no point in (re)trying.
	 */
	if (unlikely(!sd)) {
1772
		p->numa_preferred_nid = task_node(p);
1773 1774 1775
		return -EINVAL;
	}

1776
	env.dst_nid = p->numa_preferred_nid;
1777 1778 1779 1780 1781 1782
	dist = env.dist = node_distance(env.src_nid, env.dst_nid);
	taskweight = task_weight(p, env.src_nid, dist);
	groupweight = group_weight(p, env.src_nid, dist);
	update_numa_stats(&env.src_stats, env.src_nid);
	taskimp = task_weight(p, env.dst_nid, dist) - taskweight;
	groupimp = group_weight(p, env.dst_nid, dist) - groupweight;
1783
	update_numa_stats(&env.dst_stats, env.dst_nid);
1784

1785
	/* Try to find a spot on the preferred nid. */
1786 1787
	if (numa_has_capacity(&env))
		task_numa_find_cpu(&env, taskimp, groupimp);
1788

1789 1790 1791 1792 1793 1794 1795
	/*
	 * Look at other nodes in these cases:
	 * - there is no space available on the preferred_nid
	 * - the task is part of a numa_group that is interleaved across
	 *   multiple NUMA nodes; in order to better consolidate the group,
	 *   we need to check other locations.
	 */
1796
	if (env.best_cpu == -1 || (p->numa_group && p->numa_group->active_nodes > 1)) {
1797 1798 1799
		for_each_online_node(nid) {
			if (nid == env.src_nid || nid == p->numa_preferred_nid)
				continue;
1800

1801
			dist = node_distance(env.src_nid, env.dst_nid);
1802 1803 1804 1805 1806
			if (sched_numa_topology_type == NUMA_BACKPLANE &&
						dist != env.dist) {
				taskweight = task_weight(p, env.src_nid, dist);
				groupweight = group_weight(p, env.src_nid, dist);
			}
1807

1808
			/* Only consider nodes where both task and groups benefit */
1809 1810
			taskimp = task_weight(p, nid, dist) - taskweight;
			groupimp = group_weight(p, nid, dist) - groupweight;
1811
			if (taskimp < 0 && groupimp < 0)
1812 1813
				continue;

1814
			env.dist = dist;
1815 1816
			env.dst_nid = nid;
			update_numa_stats(&env.dst_stats, env.dst_nid);
1817 1818
			if (numa_has_capacity(&env))
				task_numa_find_cpu(&env, taskimp, groupimp);
1819 1820 1821
		}
	}

1822 1823 1824 1825 1826 1827 1828 1829
	/*
	 * If the task is part of a workload that spans multiple NUMA nodes,
	 * and is migrating into one of the workload's active nodes, remember
	 * this node as the task's preferred numa node, so the workload can
	 * settle down.
	 * A task that migrated to a second choice node will be better off
	 * trying for a better one later. Do not set the preferred node here.
	 */
1830
	if (p->numa_group) {
1831 1832
		struct numa_group *ng = p->numa_group;

1833 1834 1835 1836 1837
		if (env.best_cpu == -1)
			nid = env.src_nid;
		else
			nid = env.dst_nid;

1838
		if (ng->active_nodes > 1 && numa_is_active_node(env.dst_nid, ng))
1839 1840 1841 1842 1843 1844
			sched_setnuma(p, env.dst_nid);
	}

	/* No better CPU than the current one was found. */
	if (env.best_cpu == -1)
		return -EAGAIN;
1845

1846 1847 1848 1849
	/*
	 * Reset the scan period if the task is being rescheduled on an
	 * alternative node to recheck if the tasks is now properly placed.
	 */
1850
	p->numa_scan_period = task_scan_start(p);
1851

1852
	if (env.best_task == NULL) {
1853 1854 1855
		ret = migrate_task_to(p, env.best_cpu);
		if (ret != 0)
			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
1856 1857 1858 1859
		return ret;
	}

	ret = migrate_swap(p, env.best_task);
1860 1861
	if (ret != 0)
		trace_sched_stick_numa(p, env.src_cpu, task_cpu(env.best_task));
1862 1863
	put_task_struct(env.best_task);
	return ret;
1864 1865
}

1866 1867 1868
/* Attempt to migrate a task to a CPU on the preferred node. */
static void numa_migrate_preferred(struct task_struct *p)
{
1869 1870
	unsigned long interval = HZ;

1871
	/* This task has no NUMA fault statistics yet */
1872
	if (unlikely(p->numa_preferred_nid == -1 || !p->numa_faults))
1873 1874
		return;

1875
	/* Periodically retry migrating the task to the preferred node */
1876 1877
	interval = min(interval, msecs_to_jiffies(p->numa_scan_period) / 16);
	p->numa_migrate_retry = jiffies + interval;
1878 1879

	/* Success if task is already running on preferred CPU */
1880
	if (task_node(p) == p->numa_preferred_nid)
1881 1882 1883
		return;

	/* Otherwise, try migrate to a CPU on the preferred node */
1884
	task_numa_migrate(p);
1885 1886
}

1887
/*
1888
 * Find out how many nodes on the workload is actively running on. Do this by
1889 1890 1891 1892
 * tracking the nodes from which NUMA hinting faults are triggered. This can
 * be different from the set of nodes where the workload's memory is currently
 * located.
 */
1893
static void numa_group_count_active_nodes(struct numa_group *numa_group)
1894 1895
{
	unsigned long faults, max_faults = 0;
1896
	int nid, active_nodes = 0;
1897 1898 1899 1900 1901 1902 1903 1904 1905

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
		if (faults > max_faults)
			max_faults = faults;
	}

	for_each_online_node(nid) {
		faults = group_faults_cpu(numa_group, nid);
1906 1907
		if (faults * ACTIVE_NODE_FRACTION > max_faults)
			active_nodes++;
1908
	}
1909 1910 1911

	numa_group->max_faults_cpu = max_faults;
	numa_group->active_nodes = active_nodes;
1912 1913
}

1914 1915 1916
/*
 * When adapting the scan rate, the period is divided into NUMA_PERIOD_SLOTS
 * increments. The more local the fault statistics are, the higher the scan
1917 1918 1919
 * period will be for the next scan window. If local/(local+remote) ratio is
 * below NUMA_PERIOD_THRESHOLD (where range of ratio is 1..NUMA_PERIOD_SLOTS)
 * the scan period will decrease. Aim for 70% local accesses.
1920 1921
 */
#define NUMA_PERIOD_SLOTS 10
1922
#define NUMA_PERIOD_THRESHOLD 7
1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933

/*
 * Increase the scan period (slow down scanning) if the majority of
 * our memory is already on our local node, or if the majority of
 * the page accesses are shared with other processes.
 * Otherwise, decrease the scan period.
 */
static void update_task_scan_period(struct task_struct *p,
			unsigned long shared, unsigned long private)
{
	unsigned int period_slot;
1934
	int lr_ratio, ps_ratio;
1935 1936 1937 1938 1939 1940 1941 1942
	int diff;

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

	/*
	 * If there were no record hinting faults then either the task is
	 * completely idle or all activity is areas that are not of interest
1943 1944 1945
	 * to automatic numa balancing. Related to that, if there were failed
	 * migration then it implies we are migrating too quickly or the local
	 * node is overloaded. In either case, scan slower
1946
	 */
1947
	if (local + shared == 0 || p->numa_faults_locality[2]) {
1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963
		p->numa_scan_period = min(p->numa_scan_period_max,
			p->numa_scan_period << 1);

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

		return;
	}

	/*
	 * Prepare to scale scan period relative to the current period.
	 *	 == NUMA_PERIOD_THRESHOLD scan period stays the same
	 *       <  NUMA_PERIOD_THRESHOLD scan period decreases (scan faster)
	 *	 >= NUMA_PERIOD_THRESHOLD scan period increases (scan slower)
	 */
	period_slot = DIV_ROUND_UP(p->numa_scan_period, NUMA_PERIOD_SLOTS);
1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
	lr_ratio = (local * NUMA_PERIOD_SLOTS) / (local + remote);
	ps_ratio = (private * NUMA_PERIOD_SLOTS) / (private + shared);

	if (ps_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are local. There is no need to
		 * do fast NUMA scanning, since memory is already local.
		 */
		int slot = ps_ratio - NUMA_PERIOD_THRESHOLD;
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else if (lr_ratio >= NUMA_PERIOD_THRESHOLD) {
		/*
		 * Most memory accesses are shared with other tasks.
		 * There is no point in continuing fast NUMA scanning,
		 * since other tasks may just move the memory elsewhere.
		 */
		int slot = lr_ratio - NUMA_PERIOD_THRESHOLD;
1983 1984 1985 1986 1987
		if (!slot)
			slot = 1;
		diff = slot * period_slot;
	} else {
		/*
1988 1989 1990
		 * Private memory faults exceed (SLOTS-THRESHOLD)/SLOTS,
		 * yet they are not on the local NUMA node. Speed up
		 * NUMA scanning to get the memory moved over.
1991
		 */
1992 1993
		int ratio = max(lr_ratio, ps_ratio);
		diff = -(NUMA_PERIOD_THRESHOLD - ratio) * period_slot;
1994 1995 1996 1997 1998 1999 2000
	}

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

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
/*
 * Get the fraction of time the task has been running since the last
 * NUMA placement cycle. The scheduler keeps similar statistics, but
 * decays those on a 32ms period, which is orders of magnitude off
 * from the dozens-of-seconds NUMA balancing period. Use the scheduler
 * stats only if the task is so new there are no NUMA statistics yet.
 */
static u64 numa_get_avg_runtime(struct task_struct *p, u64 *period)
{
	u64 runtime, delta, now;
	/* Use the start of this time slice to avoid calculations. */
	now = p->se.exec_start;
	runtime = p->se.sum_exec_runtime;

	if (p->last_task_numa_placement) {
		delta = runtime - p->last_sum_exec_runtime;
		*period = now - p->last_task_numa_placement;
	} else {
2019
		delta = p->se.avg.load_sum;
2020
		*period = LOAD_AVG_MAX;
2021 2022 2023 2024 2025 2026 2027 2028
	}

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

	return delta;
}

2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075
/*
 * Determine the preferred nid for a task in a numa_group. This needs to
 * be done in a way that produces consistent results with group_weight,
 * otherwise workloads might not converge.
 */
static int preferred_group_nid(struct task_struct *p, int nid)
{
	nodemask_t nodes;
	int dist;

	/* Direct connections between all NUMA nodes. */
	if (sched_numa_topology_type == NUMA_DIRECT)
		return nid;

	/*
	 * On a system with glueless mesh NUMA topology, group_weight
	 * scores nodes according to the number of NUMA hinting faults on
	 * both the node itself, and on nearby nodes.
	 */
	if (sched_numa_topology_type == NUMA_GLUELESS_MESH) {
		unsigned long score, max_score = 0;
		int node, max_node = nid;

		dist = sched_max_numa_distance;

		for_each_online_node(node) {
			score = group_weight(p, node, dist);
			if (score > max_score) {
				max_score = score;
				max_node = node;
			}
		}
		return max_node;
	}

	/*
	 * Finding the preferred nid in a system with NUMA backplane
	 * interconnect topology is more involved. The goal is to locate
	 * tasks from numa_groups near each other in the system, and
	 * untangle workloads from different sides of the system. This requires
	 * searching down the hierarchy of node groups, recursively searching
	 * inside the highest scoring group of nodes. The nodemask tricks
	 * keep the complexity of the search down.
	 */
	nodes = node_online_map;
	for (dist = sched_max_numa_distance; dist > LOCAL_DISTANCE; dist--) {
		unsigned long max_faults = 0;
2076
		nodemask_t max_group = NODE_MASK_NONE;
2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109
		int a, b;

		/* Are there nodes at this distance from each other? */
		if (!find_numa_distance(dist))
			continue;

		for_each_node_mask(a, nodes) {
			unsigned long faults = 0;
			nodemask_t this_group;
			nodes_clear(this_group);

			/* Sum group's NUMA faults; includes a==b case. */
			for_each_node_mask(b, nodes) {
				if (node_distance(a, b) < dist) {
					faults += group_faults(p, b);
					node_set(b, this_group);
					node_clear(b, nodes);
				}
			}

			/* Remember the top group. */
			if (faults > max_faults) {
				max_faults = faults;
				max_group = this_group;
				/*
				 * subtle: at the smallest distance there is
				 * just one node left in each "group", the
				 * winner is the preferred nid.
				 */
				nid = a;
			}
		}
		/* Next round, evaluate the nodes within max_group. */
2110 2111
		if (!max_faults)
			break;
2112 2113 2114 2115 2116
		nodes = max_group;
	}
	return nid;
}

2117 2118
static void task_numa_placement(struct task_struct *p)
{
2119 2120
	int seq, nid, max_nid = -1, max_group_nid = -1;
	unsigned long max_faults = 0, max_group_faults = 0;
2121
	unsigned long fault_types[2] = { 0, 0 };
2122 2123
	unsigned long total_faults;
	u64 runtime, period;
2124
	spinlock_t *group_lock = NULL;
2125

2126 2127 2128 2129 2130
	/*
	 * The p->mm->numa_scan_seq field gets updated without
	 * exclusive access. Use READ_ONCE() here to ensure
	 * that the field is read in a single access:
	 */
2131
	seq = READ_ONCE(p->mm->numa_scan_seq);
2132 2133 2134
	if (p->numa_scan_seq == seq)
		return;
	p->numa_scan_seq = seq;
2135
	p->numa_scan_period_max = task_scan_max(p);
2136

2137 2138 2139 2140
	total_faults = p->numa_faults_locality[0] +
		       p->numa_faults_locality[1];
	runtime = numa_get_avg_runtime(p, &period);

2141 2142 2143
	/* If the task is part of a group prevent parallel updates to group stats */
	if (p->numa_group) {
		group_lock = &p->numa_group->lock;
2144
		spin_lock_irq(group_lock);
2145 2146
	}

2147 2148
	/* Find the node with the highest number of faults */
	for_each_online_node(nid) {
2149 2150
		/* Keep track of the offsets in numa_faults array */
		int mem_idx, membuf_idx, cpu_idx, cpubuf_idx;
2151
		unsigned long faults = 0, group_faults = 0;
2152
		int priv;
2153

2154
		for (priv = 0; priv < NR_NUMA_HINT_FAULT_TYPES; priv++) {
2155
			long diff, f_diff, f_weight;
2156

2157 2158 2159 2160
			mem_idx = task_faults_idx(NUMA_MEM, nid, priv);
			membuf_idx = task_faults_idx(NUMA_MEMBUF, nid, priv);
			cpu_idx = task_faults_idx(NUMA_CPU, nid, priv);
			cpubuf_idx = task_faults_idx(NUMA_CPUBUF, nid, priv);
2161

2162
			/* Decay existing window, copy faults since last scan */
2163 2164 2165
			diff = p->numa_faults[membuf_idx] - p->numa_faults[mem_idx] / 2;
			fault_types[priv] += p->numa_faults[membuf_idx];
			p->numa_faults[membuf_idx] = 0;
2166

2167 2168 2169 2170 2171 2172 2173 2174
			/*
			 * Normalize the faults_from, so all tasks in a group
			 * count according to CPU use, instead of by the raw
			 * number of faults. Tasks with little runtime have
			 * little over-all impact on throughput, and thus their
			 * faults are less important.
			 */
			f_weight = div64_u64(runtime << 16, period + 1);
2175
			f_weight = (f_weight * p->numa_faults[cpubuf_idx]) /
2176
				   (total_faults + 1);
2177 2178
			f_diff = f_weight - p->numa_faults[cpu_idx] / 2;
			p->numa_faults[cpubuf_idx] = 0;
2179

2180 2181 2182
			p->numa_faults[mem_idx] += diff;
			p->numa_faults[cpu_idx] += f_diff;
			faults += p->numa_faults[mem_idx];
2183
			p->total_numa_faults += diff;
2184
			if (p->numa_group) {
2185 2186 2187 2188 2189 2190 2191 2192 2193
				/*
				 * safe because we can only change our own group
				 *
				 * mem_idx represents the offset for a given
				 * nid and priv in a specific region because it
				 * is at the beginning of the numa_faults array.
				 */
				p->numa_group->faults[mem_idx] += diff;
				p->numa_group->faults_cpu[mem_idx] += f_diff;
2194
				p->numa_group->total_faults += diff;
2195
				group_faults += p->numa_group->faults[mem_idx];
2196
			}
2197 2198
		}

2199 2200 2201 2202
		if (faults > max_faults) {
			max_faults = faults;
			max_nid = nid;
		}
2203 2204 2205 2206 2207 2208 2209

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

2210 2211
	update_task_scan_period(p, fault_types[0], fault_types[1]);

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

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

		if (task_node(p) != p->numa_preferred_nid)
			numa_migrate_preferred(p);
2225
	}
2226 2227
}

2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238
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);
}

2239 2240
static void task_numa_group(struct task_struct *p, int cpupid, int flags,
			int *priv)
2241 2242 2243 2244 2245 2246 2247 2248 2249
{
	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) +
2250
				    4*nr_node_ids*sizeof(unsigned long);
2251 2252 2253 2254 2255 2256

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

		atomic_set(&grp->refcount, 1);
2257 2258
		grp->active_nodes = 1;
		grp->max_faults_cpu = 0;
2259
		spin_lock_init(&grp->lock);
2260
		grp->gid = p->pid;
2261
		/* Second half of the array tracks nids where faults happen */
2262 2263
		grp->faults_cpu = grp->faults + NR_NUMA_HINT_FAULT_TYPES *
						nr_node_ids;
2264

2265
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2266
			grp->faults[i] = p->numa_faults[i];
2267

2268
		grp->total_faults = p->total_numa_faults;
2269

2270 2271 2272 2273 2274
		grp->nr_tasks++;
		rcu_assign_pointer(p->numa_group, grp);
	}

	rcu_read_lock();
2275
	tsk = READ_ONCE(cpu_rq(cpu)->curr);
2276 2277

	if (!cpupid_match_pid(tsk, cpupid))
2278
		goto no_join;
2279 2280 2281

	grp = rcu_dereference(tsk->numa_group);
	if (!grp)
2282
		goto no_join;
2283 2284 2285

	my_grp = p->numa_group;
	if (grp == my_grp)
2286
		goto no_join;
2287 2288 2289 2290 2291 2292

	/*
	 * 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)
2293
		goto no_join;
2294 2295 2296 2297 2298

	/*
	 * Tie-break on the grp address.
	 */
	if (my_grp->nr_tasks == grp->nr_tasks && my_grp > grp)
2299
		goto no_join;
2300

2301 2302 2303 2304 2305 2306 2307
	/* 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;
2308

2309 2310 2311
	/* Update priv based on whether false sharing was detected */
	*priv = !join;

2312
	if (join && !get_numa_group(grp))
2313
		goto no_join;
2314 2315 2316 2317 2318 2319

	rcu_read_unlock();

	if (!join)
		return;

2320 2321
	BUG_ON(irqs_disabled());
	double_lock_irq(&my_grp->lock, &grp->lock);
2322

2323
	for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++) {
2324 2325
		my_grp->faults[i] -= p->numa_faults[i];
		grp->faults[i] += p->numa_faults[i];
2326
	}
2327 2328
	my_grp->total_faults -= p->total_numa_faults;
	grp->total_faults += p->total_numa_faults;
2329 2330 2331 2332 2333

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

	spin_unlock(&my_grp->lock);
2334
	spin_unlock_irq(&grp->lock);
2335 2336 2337 2338

	rcu_assign_pointer(p->numa_group, grp);

	put_numa_group(my_grp);
2339 2340 2341 2342 2343
	return;

no_join:
	rcu_read_unlock();
	return;
2344 2345 2346 2347 2348
}

void task_numa_free(struct task_struct *p)
{
	struct numa_group *grp = p->numa_group;
2349
	void *numa_faults = p->numa_faults;
2350 2351
	unsigned long flags;
	int i;
2352 2353

	if (grp) {
2354
		spin_lock_irqsave(&grp->lock, flags);
2355
		for (i = 0; i < NR_NUMA_HINT_FAULT_STATS * nr_node_ids; i++)
2356
			grp->faults[i] -= p->numa_faults[i];
2357
		grp->total_faults -= p->total_numa_faults;
2358

2359
		grp->nr_tasks--;
2360
		spin_unlock_irqrestore(&grp->lock, flags);
2361
		RCU_INIT_POINTER(p->numa_group, NULL);
2362 2363 2364
		put_numa_group(grp);
	}

2365
	p->numa_faults = NULL;
2366
	kfree(numa_faults);
2367 2368
}

2369 2370 2371
/*
 * Got a PROT_NONE fault for a page on @node.
 */
2372
void task_numa_fault(int last_cpupid, int mem_node, int pages, int flags)
2373 2374
{
	struct task_struct *p = current;
2375
	bool migrated = flags & TNF_MIGRATED;
2376
	int cpu_node = task_node(current);
2377
	int local = !!(flags & TNF_FAULT_LOCAL);
2378
	struct numa_group *ng;
2379
	int priv;
2380

2381
	if (!static_branch_likely(&sched_numa_balancing))
2382 2383
		return;

2384 2385 2386 2387
	/* for example, ksmd faulting in a user's mm */
	if (!p->mm)
		return;

2388
	/* Allocate buffer to track faults on a per-node basis */
2389 2390
	if (unlikely(!p->numa_faults)) {
		int size = sizeof(*p->numa_faults) *
2391
			   NR_NUMA_HINT_FAULT_BUCKETS * nr_node_ids;
2392

2393 2394
		p->numa_faults = kzalloc(size, GFP_KERNEL|__GFP_NOWARN);
		if (!p->numa_faults)
2395
			return;
2396

2397
		p->total_numa_faults = 0;
2398
		memset(p->numa_faults_locality, 0, sizeof(p->numa_faults_locality));
2399
	}
2400

2401 2402 2403 2404 2405 2406 2407 2408
	/*
	 * 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);
2409
		if (!priv && !(flags & TNF_NO_GROUP))
2410
			task_numa_group(p, last_cpupid, flags, &priv);
2411 2412
	}

2413 2414 2415 2416 2417 2418
	/*
	 * If a workload spans multiple NUMA nodes, a shared fault that
	 * occurs wholly within the set of nodes that the workload is
	 * actively using should be counted as local. This allows the
	 * scan rate to slow down when a workload has settled down.
	 */
2419 2420 2421 2422
	ng = p->numa_group;
	if (!priv && !local && ng && ng->active_nodes > 1 &&
				numa_is_active_node(cpu_node, ng) &&
				numa_is_active_node(mem_node, ng))
2423 2424
		local = 1;

2425
	task_numa_placement(p);
2426

2427 2428 2429 2430 2431
	/*
	 * Retry task to preferred node migration periodically, in case it
	 * case it previously failed, or the scheduler moved us.
	 */
	if (time_after(jiffies, p->numa_migrate_retry))
2432 2433
		numa_migrate_preferred(p);

I
Ingo Molnar 已提交
2434 2435
	if (migrated)
		p->numa_pages_migrated += pages;
2436 2437
	if (flags & TNF_MIGRATE_FAIL)
		p->numa_faults_locality[2] += pages;
I
Ingo Molnar 已提交
2438

2439 2440
	p->numa_faults[task_faults_idx(NUMA_MEMBUF, mem_node, priv)] += pages;
	p->numa_faults[task_faults_idx(NUMA_CPUBUF, cpu_node, priv)] += pages;
2441
	p->numa_faults_locality[local] += pages;
2442 2443
}

2444 2445
static void reset_ptenuma_scan(struct task_struct *p)
{
2446 2447 2448 2449 2450 2451 2452 2453
	/*
	 * We only did a read acquisition of the mmap sem, so
	 * p->mm->numa_scan_seq is written to without exclusive access
	 * and the update is not guaranteed to be atomic. That's not
	 * much of an issue though, since this is just used for
	 * statistical sampling. Use READ_ONCE/WRITE_ONCE, which are not
	 * expensive, to avoid any form of compiler optimizations:
	 */
2454
	WRITE_ONCE(p->mm->numa_scan_seq, READ_ONCE(p->mm->numa_scan_seq) + 1);
2455 2456 2457
	p->mm->numa_scan_offset = 0;
}

2458 2459 2460 2461 2462 2463 2464 2465 2466
/*
 * 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;
2467
	u64 runtime = p->se.sum_exec_runtime;
2468
	struct vm_area_struct *vma;
2469
	unsigned long start, end;
2470
	unsigned long nr_pte_updates = 0;
2471
	long pages, virtpages;
2472

2473
	SCHED_WARN_ON(p != container_of(work, struct task_struct, numa_work));
2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486

	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;

2487
	if (!mm->numa_next_scan) {
2488 2489
		mm->numa_next_scan = now +
			msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2490 2491
	}

2492 2493 2494 2495 2496 2497 2498
	/*
	 * Enforce maximal scan/migration frequency..
	 */
	migrate = mm->numa_next_scan;
	if (time_before(now, migrate))
		return;

2499 2500
	if (p->numa_scan_period == 0) {
		p->numa_scan_period_max = task_scan_max(p);
2501
		p->numa_scan_period = task_scan_start(p);
2502
	}
2503

2504
	next_scan = now + msecs_to_jiffies(p->numa_scan_period);
2505 2506 2507
	if (cmpxchg(&mm->numa_next_scan, migrate, next_scan) != migrate)
		return;

2508 2509 2510 2511 2512 2513
	/*
	 * Delay this task enough that another task of this mm will likely win
	 * the next time around.
	 */
	p->node_stamp += 2 * TICK_NSEC;

2514 2515 2516
	start = mm->numa_scan_offset;
	pages = sysctl_numa_balancing_scan_size;
	pages <<= 20 - PAGE_SHIFT; /* MB in pages */
2517
	virtpages = pages * 8;	   /* Scan up to this much virtual space */
2518 2519
	if (!pages)
		return;
2520

2521

2522 2523
	if (!down_read_trylock(&mm->mmap_sem))
		return;
2524
	vma = find_vma(mm, start);
2525 2526
	if (!vma) {
		reset_ptenuma_scan(p);
2527
		start = 0;
2528 2529
		vma = mm->mmap;
	}
2530
	for (; vma; vma = vma->vm_next) {
2531
		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
2532
			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
2533
			continue;
2534
		}
2535

2536 2537 2538 2539 2540 2541 2542 2543 2544 2545
		/*
		 * Shared library pages mapped by multiple processes are not
		 * migrated as it is expected they are cache replicated. Avoid
		 * hinting faults in read-only file-backed mappings or the vdso
		 * as migrating the pages will be of marginal benefit.
		 */
		if (!vma->vm_mm ||
		    (vma->vm_file && (vma->vm_flags & (VM_READ|VM_WRITE)) == (VM_READ)))
			continue;

M
Mel Gorman 已提交
2546 2547 2548 2549 2550 2551
		/*
		 * Skip inaccessible VMAs to avoid any confusion between
		 * PROT_NONE and NUMA hinting ptes
		 */
		if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
			continue;
2552

2553 2554 2555 2556
		do {
			start = max(start, vma->vm_start);
			end = ALIGN(start + (pages << PAGE_SHIFT), HPAGE_SIZE);
			end = min(end, vma->vm_end);
2557
			nr_pte_updates = change_prot_numa(vma, start, end);
2558 2559

			/*
2560 2561 2562 2563 2564 2565
			 * Try to scan sysctl_numa_balancing_size worth of
			 * hpages that have at least one present PTE that
			 * is not already pte-numa. If the VMA contains
			 * areas that are unused or already full of prot_numa
			 * PTEs, scan up to virtpages, to skip through those
			 * areas faster.
2566 2567 2568
			 */
			if (nr_pte_updates)
				pages -= (end - start) >> PAGE_SHIFT;
2569
			virtpages -= (end - start) >> PAGE_SHIFT;
2570

2571
			start = end;
2572
			if (pages <= 0 || virtpages <= 0)
2573
				goto out;
2574 2575

			cond_resched();
2576
		} while (end != vma->vm_end);
2577
	}
2578

2579
out:
2580
	/*
P
Peter Zijlstra 已提交
2581 2582 2583 2584
	 * 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.
2585 2586
	 */
	if (vma)
2587
		mm->numa_scan_offset = start;
2588 2589 2590
	else
		reset_ptenuma_scan(p);
	up_read(&mm->mmap_sem);
2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601

	/*
	 * Make sure tasks use at least 32x as much time to run other code
	 * than they used here, to limit NUMA PTE scanning overhead to 3% max.
	 * Usually update_task_scan_period slows down scanning enough; on an
	 * overloaded system we need to limit overhead on a per task basis.
	 */
	if (unlikely(p->se.sum_exec_runtime != runtime)) {
		u64 diff = p->se.sum_exec_runtime - runtime;
		p->node_stamp += 32 * diff;
	}
2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626
}

/*
 * 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;

2627
	if (now > curr->node_stamp + period) {
2628
		if (!curr->node_stamp)
2629
			curr->numa_scan_period = task_scan_start(curr);
2630
		curr->node_stamp += period;
2631 2632 2633 2634 2635 2636 2637

		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);
		}
	}
}
2638

2639 2640 2641 2642
#else
static void task_tick_numa(struct rq *rq, struct task_struct *curr)
{
}
2643 2644 2645 2646 2647 2648 2649 2650

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)
{
}
2651

2652 2653
#endif /* CONFIG_NUMA_BALANCING */

2654 2655 2656 2657
static void
account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
	update_load_add(&cfs_rq->load, se->load.weight);
2658
	if (!parent_entity(se))
2659
		update_load_add(&rq_of(cfs_rq)->load, se->load.weight);
2660
#ifdef CONFIG_SMP
2661 2662 2663 2664 2665 2666
	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);
	}
2667
#endif
2668 2669 2670 2671 2672 2673 2674
	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);
2675
	if (!parent_entity(se))
2676
		update_load_sub(&rq_of(cfs_rq)->load, se->load.weight);
2677
#ifdef CONFIG_SMP
2678 2679
	if (entity_is_task(se)) {
		account_numa_dequeue(rq_of(cfs_rq), task_of(se));
2680
		list_del_init(&se->group_node);
2681
	}
2682
#endif
2683 2684 2685
	cfs_rq->nr_running--;
}

2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724
/*
 * Signed add and clamp on underflow.
 *
 * Explicitly do a load-store to ensure the intermediate value never hits
 * memory. This allows lockless observations without ever seeing the negative
 * values.
 */
#define add_positive(_ptr, _val) do {                           \
	typeof(_ptr) ptr = (_ptr);                              \
	typeof(_val) val = (_val);                              \
	typeof(*ptr) res, var = READ_ONCE(*ptr);                \
								\
	res = var + val;                                        \
								\
	if (val < 0 && res > var)                               \
		res = 0;                                        \
								\
	WRITE_ONCE(*ptr, res);                                  \
} while (0)

/*
 * Unsigned subtract and clamp on underflow.
 *
 * Explicitly do a load-store to ensure the intermediate value never hits
 * memory. This allows lockless observations without ever seeing the negative
 * values.
 */
#define sub_positive(_ptr, _val) do {				\
	typeof(_ptr) ptr = (_ptr);				\
	typeof(*ptr) val = (_val);				\
	typeof(*ptr) res, var = READ_ONCE(*ptr);		\
	res = var - val;					\
	if (res > var)						\
		res = 0;					\
	WRITE_ONCE(*ptr, res);					\
} while (0)

#ifdef CONFIG_SMP
/*
2725
 * XXX we want to get rid of these helpers and use the full load resolution.
2726 2727 2728 2729 2730 2731
 */
static inline long se_weight(struct sched_entity *se)
{
	return scale_load_down(se->load.weight);
}

2732 2733 2734 2735 2736
static inline long se_runnable(struct sched_entity *se)
{
	return scale_load_down(se->runnable_weight);
}

2737 2738 2739
static inline void
enqueue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2740 2741 2742 2743
	cfs_rq->runnable_weight += se->runnable_weight;

	cfs_rq->avg.runnable_load_avg += se->avg.runnable_load_avg;
	cfs_rq->avg.runnable_load_sum += se_runnable(se) * se->avg.runnable_load_sum;
2744 2745 2746 2747 2748
}

static inline void
dequeue_runnable_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
2749 2750 2751 2752 2753
	cfs_rq->runnable_weight -= se->runnable_weight;

	sub_positive(&cfs_rq->avg.runnable_load_avg, se->avg.runnable_load_avg);
	sub_positive(&cfs_rq->avg.runnable_load_sum,
		     se_runnable(se) * se->avg.runnable_load_sum);
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779
}

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

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

2780
static void reweight_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
2781
			    unsigned long weight, unsigned long runnable)
2782 2783 2784 2785 2786 2787 2788 2789 2790 2791
{
	if (se->on_rq) {
		/* commit outstanding execution time */
		if (cfs_rq->curr == se)
			update_curr(cfs_rq);
		account_entity_dequeue(cfs_rq, se);
		dequeue_runnable_load_avg(cfs_rq, se);
	}
	dequeue_load_avg(cfs_rq, se);

2792
	se->runnable_weight = runnable;
2793 2794 2795
	update_load_set(&se->load, weight);

#ifdef CONFIG_SMP
2796 2797 2798 2799 2800 2801 2802
	do {
		u32 divider = LOAD_AVG_MAX - 1024 + se->avg.period_contrib;

		se->avg.load_avg = div_u64(se_weight(se) * se->avg.load_sum, divider);
		se->avg.runnable_load_avg =
			div_u64(se_runnable(se) * se->avg.runnable_load_sum, divider);
	} while (0);
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818
#endif

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

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

2819
	reweight_entity(cfs_rq, se, weight, weight);
2820 2821 2822
	load->inv_weight = sched_prio_to_wmult[prio];
}

2823 2824
#ifdef CONFIG_FAIR_GROUP_SCHED
# ifdef CONFIG_SMP
2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885
/*
 * All this does is approximate the hierarchical proportion which includes that
 * global sum we all love to hate.
 *
 * That is, the weight of a group entity, is the proportional share of the
 * group weight based on the group runqueue weights. That is:
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = -----------------------------               (1)
 *			  \Sum grq->load.weight
 *
 * Now, because computing that sum is prohibitively expensive to compute (been
 * there, done that) we approximate it with this average stuff. The average
 * moves slower and therefore the approximation is cheaper and more stable.
 *
 * So instead of the above, we substitute:
 *
 *   grq->load.weight -> grq->avg.load_avg                         (2)
 *
 * which yields the following:
 *
 *                     tg->weight * grq->avg.load_avg
 *   ge->load.weight = ------------------------------              (3)
 *				tg->load_avg
 *
 * Where: tg->load_avg ~= \Sum grq->avg.load_avg
 *
 * That is shares_avg, and it is right (given the approximation (2)).
 *
 * The problem with it is that because the average is slow -- it was designed
 * to be exactly that of course -- this leads to transients in boundary
 * conditions. In specific, the case where the group was idle and we start the
 * one task. It takes time for our CPU's grq->avg.load_avg to build up,
 * yielding bad latency etc..
 *
 * Now, in that special case (1) reduces to:
 *
 *                     tg->weight * grq->load.weight
 *   ge->load.weight = ----------------------------- = tg>weight   (4)
 *			    grp->load.weight
 *
 * That is, the sum collapses because all other CPUs are idle; the UP scenario.
 *
 * So what we do is modify our approximation (3) to approach (4) in the (near)
 * UP case, like:
 *
 *   ge->load.weight =
 *
 *              tg->weight * grq->load.weight
 *     ---------------------------------------------------         (5)
 *     tg->load_avg - grq->avg.load_avg + grq->load.weight
 *
 *
 * And that is shares_weight and is icky. In the (near) UP case it approaches
 * (4) while in the normal case it approaches (3). It consistently
 * overestimates the ge->load.weight and therefore:
 *
 *   \Sum ge->load.weight >= tg->weight
 *
 * hence icky!
 */
2886
static long calc_cfs_shares(struct cfs_rq *cfs_rq)
2887
{
2888 2889 2890 2891
	long tg_weight, tg_shares, load, shares;
	struct task_group *tg = cfs_rq->tg;

	tg_shares = READ_ONCE(tg->shares);
2892 2893

	/*
2894 2895
	 * Because (5) drops to 0 when the cfs_rq is idle, we need to use (3)
	 * as a lower bound.
2896
	 */
2897
	load = max(scale_load_down(cfs_rq->load.weight), cfs_rq->avg.load_avg);
2898

2899
	tg_weight = atomic_long_read(&tg->load_avg);
2900

2901 2902 2903
	/* Ensure tg_weight >= load */
	tg_weight -= cfs_rq->tg_load_avg_contrib;
	tg_weight += load;
2904

2905
	shares = (tg_shares * load);
2906 2907
	if (tg_weight)
		shares /= tg_weight;
2908

2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920
	/*
	 * MIN_SHARES has to be unscaled here to support per-CPU partitioning
	 * of a group with small tg->shares value. It is a floor value which is
	 * assigned as a minimum load.weight to the sched_entity representing
	 * the group on a CPU.
	 *
	 * E.g. on 64-bit for a group with tg->shares of scale_load(15)=15*1024
	 * on an 8-core system with 8 tasks each runnable on one CPU shares has
	 * to be 15*1024*1/8=1920 instead of scale_load(MIN_SHARES)=2*1024. In
	 * case no task is runnable on a CPU MIN_SHARES=2 should be returned
	 * instead of 0.
	 */
2921
	return clamp_t(long, shares, MIN_SHARES, tg_shares);
2922 2923
}
# endif /* CONFIG_SMP */
2924

2925 2926
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq);

2927 2928 2929 2930 2931
/*
 * Recomputes the group entity based on the current state of its group
 * runqueue.
 */
static void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
2932
{
2933 2934
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
	long shares, runnable;
P
Peter Zijlstra 已提交
2935

2936
	if (!gcfs_rq)
2937 2938
		return;

2939
	if (throttled_hierarchy(gcfs_rq))
P
Peter Zijlstra 已提交
2940
		return;
2941

2942
#ifndef CONFIG_SMP
2943
	runnable = shares = READ_ONCE(gcfs_rq->tg->shares);
2944 2945

	if (likely(se->load.weight == shares))
2946
		return;
2947
#else
2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958
	shares = calc_cfs_shares(gcfs_rq);
	/*
	 * The hierarchical runnable load metric is the proportional part
	 * of this group's runnable_load_avg / load_avg.
	 *
	 * Note: we need to deal with very sporadic 'runnable > load' cases
	 * due to numerical instability.
	 */
	runnable = shares * gcfs_rq->avg.runnable_load_avg;
	if (runnable)
		runnable /= max(gcfs_rq->avg.load_avg, gcfs_rq->avg.runnable_load_avg);
2959
#endif
P
Peter Zijlstra 已提交
2960

2961
	reweight_entity(cfs_rq_of(se), se, shares, runnable);
P
Peter Zijlstra 已提交
2962
}
2963

P
Peter Zijlstra 已提交
2964
#else /* CONFIG_FAIR_GROUP_SCHED */
2965
static inline void update_cfs_group(struct sched_entity *se)
P
Peter Zijlstra 已提交
2966 2967 2968 2969
{
}
#endif /* CONFIG_FAIR_GROUP_SCHED */

2970 2971
static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq)
{
2972 2973 2974
	struct rq *rq = rq_of(cfs_rq);

	if (&rq->cfs == cfs_rq) {
2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990
		/*
		 * There are a few boundary cases this might miss but it should
		 * get called often enough that that should (hopefully) not be
		 * a real problem -- added to that it only calls on the local
		 * CPU, so if we enqueue remotely we'll miss an update, but
		 * the next tick/schedule should update.
		 *
		 * It will not get called when we go idle, because the idle
		 * thread is a different class (!fair), nor will the utilization
		 * number include things like RT tasks.
		 *
		 * As is, the util number is not freq-invariant (we'd have to
		 * implement arch_scale_freq_capacity() for that).
		 *
		 * See cpu_util().
		 */
2991
		cpufreq_update_util(rq, 0);
2992 2993 2994
	}
}

2995
#ifdef CONFIG_SMP
2996 2997 2998 2999
/*
 * Approximate:
 *   val * y^n,    where y^32 ~= 0.5 (~1 scheduling period)
 */
3000
static u64 decay_load(u64 val, u64 n)
3001
{
3002 3003
	unsigned int local_n;

3004
	if (unlikely(n > LOAD_AVG_PERIOD * 63))
3005 3006 3007 3008 3009 3010 3011
		return 0;

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

	/*
	 * As y^PERIOD = 1/2, we can combine
3012 3013
	 *    y^n = 1/2^(n/PERIOD) * y^(n%PERIOD)
	 * With a look-up table which covers y^n (n<PERIOD)
3014 3015 3016 3017 3018 3019
	 *
	 * To achieve constant time decay_load.
	 */
	if (unlikely(local_n >= LOAD_AVG_PERIOD)) {
		val >>= local_n / LOAD_AVG_PERIOD;
		local_n %= LOAD_AVG_PERIOD;
3020 3021
	}

3022 3023
	val = mul_u64_u32_shr(val, runnable_avg_yN_inv[local_n], 32);
	return val;
3024 3025
}

3026
static u32 __accumulate_pelt_segments(u64 periods, u32 d1, u32 d3)
3027
{
3028
	u32 c1, c2, c3 = d3; /* y^0 == 1 */
3029

3030
	/*
P
Peter Zijlstra 已提交
3031
	 * c1 = d1 y^p
3032
	 */
3033
	c1 = decay_load((u64)d1, periods);
3034 3035

	/*
P
Peter Zijlstra 已提交
3036
	 *            p-1
3037 3038
	 * c2 = 1024 \Sum y^n
	 *            n=1
3039
	 *
3040 3041
	 *              inf        inf
	 *    = 1024 ( \Sum y^n - \Sum y^n - y^0 )
P
Peter Zijlstra 已提交
3042
	 *              n=0        n=p
3043
	 */
3044
	c2 = LOAD_AVG_MAX - decay_load(LOAD_AVG_MAX, periods) - 1024;
3045 3046

	return c1 + c2 + c3;
3047 3048
}

3049
#define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT)
3050

3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061
/*
 * Accumulate the three separate parts of the sum; d1 the remainder
 * of the last (incomplete) period, d2 the span of full periods and d3
 * the remainder of the (incomplete) current period.
 *
 *           d1          d2           d3
 *           ^           ^            ^
 *           |           |            |
 *         |<->|<----------------->|<--->|
 * ... |---x---|------| ... |------|-----x (now)
 *
P
Peter Zijlstra 已提交
3062 3063 3064
 *                           p-1
 * u' = (u + d1) y^p + 1024 \Sum y^n + d3 y^0
 *                           n=1
3065
 *
P
Peter Zijlstra 已提交
3066
 *    = u y^p +					(Step 1)
3067
 *
P
Peter Zijlstra 已提交
3068 3069 3070
 *                     p-1
 *      d1 y^p + 1024 \Sum y^n + d3 y^0		(Step 2)
 *                     n=1
3071 3072 3073
 */
static __always_inline u32
accumulate_sum(u64 delta, int cpu, struct sched_avg *sa,
3074
	       unsigned long load, unsigned long runnable, int running)
3075 3076
{
	unsigned long scale_freq, scale_cpu;
3077
	u32 contrib = (u32)delta; /* p == 0 -> delta < 1024 */
3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090
	u64 periods;

	scale_freq = arch_scale_freq_capacity(NULL, cpu);
	scale_cpu = arch_scale_cpu_capacity(NULL, cpu);

	delta += sa->period_contrib;
	periods = delta / 1024; /* A period is 1024us (~1ms) */

	/*
	 * Step 1: decay old *_sum if we crossed period boundaries.
	 */
	if (periods) {
		sa->load_sum = decay_load(sa->load_sum, periods);
3091 3092
		sa->runnable_load_sum =
			decay_load(sa->runnable_load_sum, periods);
3093 3094
		sa->util_sum = decay_load((u64)(sa->util_sum), periods);

3095 3096 3097 3098 3099 3100 3101
		/*
		 * Step 2
		 */
		delta %= 1024;
		contrib = __accumulate_pelt_segments(periods,
				1024 - sa->period_contrib, delta);
	}
3102 3103 3104
	sa->period_contrib = delta;

	contrib = cap_scale(contrib, scale_freq);
3105 3106 3107 3108
	if (load)
		sa->load_sum += load * contrib;
	if (runnable)
		sa->runnable_load_sum += runnable * contrib;
3109 3110 3111 3112 3113 3114
	if (running)
		sa->util_sum += contrib * scale_cpu;

	return periods;
}

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
/*
 * 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}]
 */
3143
static __always_inline int
3144
___update_load_sum(u64 now, int cpu, struct sched_avg *sa,
3145
		  unsigned long load, unsigned long runnable, int running)
3146
{
3147
	u64 delta;
3148

3149
	delta = now - sa->last_update_time;
3150 3151 3152 3153 3154
	/*
	 * 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) {
3155
		sa->last_update_time = now;
3156 3157 3158 3159 3160 3161 3162 3163 3164 3165
		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;
3166 3167

	sa->last_update_time += delta << 10;
3168

3169 3170 3171 3172 3173 3174 3175 3176 3177
	/*
	 * running is a subset of runnable (weight) so running can't be set if
	 * runnable is clear. But there are some corner cases where the current
	 * se has been already dequeued but cfs_rq->curr still points to it.
	 * This means that weight will be 0 but not running for a sched_entity
	 * but also for a cfs_rq if the latter becomes idle. As an example,
	 * this happens during idle_balance() which calls
	 * update_blocked_averages()
	 */
3178 3179
	if (!load)
		runnable = running = 0;
3180

3181 3182 3183 3184 3185 3186 3187
	/*
	 * Now we know we crossed measurement unit boundaries. The *_avg
	 * accrues by two steps:
	 *
	 * Step 1: accumulate *_sum since last_update_time. If we haven't
	 * crossed period boundaries, finish.
	 */
3188
	if (!accumulate_sum(delta, cpu, sa, load, runnable, running))
3189
		return 0;
3190

3191 3192 3193 3194
	return 1;
}

static __always_inline void
3195
___update_load_avg(struct sched_avg *sa, unsigned long load, unsigned long runnable)
3196 3197 3198
{
	u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;

3199 3200 3201
	/*
	 * Step 2: update *_avg.
	 */
3202 3203
	sa->load_avg = div_u64(load * sa->load_sum, divider);
	sa->runnable_load_avg =	div_u64(runnable * sa->runnable_load_sum, divider);
3204 3205
	sa->util_avg = sa->util_sum / divider;
}
3206

3207 3208 3209
/*
 * sched_entity:
 *
3210 3211 3212 3213 3214 3215 3216
 *   task:
 *     se_runnable() == se_weight()
 *
 *   group: [ see update_cfs_group() ]
 *     se_weight()   = tg->weight * grq->load_avg / tg->load_avg
 *     se_runnable() = se_weight(se) * grq->runnable_load_avg / grq->load_avg
 *
3217 3218 3219
 *   load_sum := runnable_sum
 *   load_avg = se_weight(se) * runnable_avg
 *
3220 3221 3222 3223 3224
 *   runnable_load_sum := runnable_sum
 *   runnable_load_avg = se_runnable(se) * runnable_avg
 *
 * XXX collapse load_sum and runnable_load_sum
 *
3225 3226 3227 3228
 * cfq_rs:
 *
 *   load_sum = \Sum se_weight(se) * se->avg.load_sum
 *   load_avg = \Sum se->avg.load_avg
3229 3230 3231
 *
 *   runnable_load_sum = \Sum se_runnable(se) * se->avg.runnable_load_sum
 *   runnable_load_avg = \Sum se->avg.runable_load_avg
3232 3233
 */

3234 3235 3236
static int
__update_load_avg_blocked_se(u64 now, int cpu, struct sched_entity *se)
{
3237 3238 3239 3240 3241
	if (entity_is_task(se))
		se->runnable_weight = se->load.weight;

	if (___update_load_sum(now, cpu, &se->avg, 0, 0, 0)) {
		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
3242 3243 3244 3245
		return 1;
	}

	return 0;
3246 3247 3248 3249 3250
}

static int
__update_load_avg_se(u64 now, int cpu, struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3251 3252 3253 3254 3255
	if (entity_is_task(se))
		se->runnable_weight = se->load.weight;

	if (___update_load_sum(now, cpu, &se->avg, !!se->on_rq, !!se->on_rq,
				cfs_rq->curr == se)) {
3256

3257
		___update_load_avg(&se->avg, se_weight(se), se_runnable(se));
3258 3259 3260 3261
		return 1;
	}

	return 0;
3262 3263 3264 3265 3266
}

static int
__update_load_avg_cfs_rq(u64 now, int cpu, struct cfs_rq *cfs_rq)
{
3267 3268
	if (___update_load_sum(now, cpu, &cfs_rq->avg,
				scale_load_down(cfs_rq->load.weight),
3269 3270 3271 3272
				scale_load_down(cfs_rq->runnable_weight),
				cfs_rq->curr != NULL)) {

		___update_load_avg(&cfs_rq->avg, 1, 1);
3273 3274 3275 3276
		return 1;
	}

	return 0;
3277 3278
}

3279
#ifdef CONFIG_FAIR_GROUP_SCHED
3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292
/**
 * update_tg_load_avg - update the tg's load avg
 * @cfs_rq: the cfs_rq whose avg changed
 * @force: update regardless of how small the difference
 *
 * This function 'ensures': tg->load_avg := \Sum tg->cfs_rq[]->avg.load.
 * However, because tg->load_avg is a global value there are performance
 * considerations.
 *
 * In order to avoid having to look at the other cfs_rq's, we use a
 * differential update where we store the last value we propagated. This in
 * turn allows skipping updates if the differential is 'small'.
 *
3293
 * Updating tg's load_avg is necessary before update_cfs_share().
3294
 */
3295
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force)
3296
{
3297
	long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
3298

3299 3300 3301 3302 3303 3304
	/*
	 * No need to update load_avg for root_task_group as it is not used.
	 */
	if (cfs_rq->tg == &root_task_group)
		return;

3305 3306 3307
	if (force || abs(delta) > cfs_rq->tg_load_avg_contrib / 64) {
		atomic_long_add(delta, &cfs_rq->tg->load_avg);
		cfs_rq->tg_load_avg_contrib = cfs_rq->avg.load_avg;
3308
	}
3309
}
3310

3311 3312 3313 3314 3315 3316 3317 3318
/*
 * Called within set_task_rq() right before setting a task's cpu. The
 * caller only guarantees p->pi_lock is held; no other assumptions,
 * including the state of rq->lock, should be made.
 */
void set_task_rq_fair(struct sched_entity *se,
		      struct cfs_rq *prev, struct cfs_rq *next)
{
3319 3320 3321
	u64 p_last_update_time;
	u64 n_last_update_time;

3322 3323 3324 3325 3326 3327 3328 3329 3330 3331
	if (!sched_feat(ATTACH_AGE_LOAD))
		return;

	/*
	 * We are supposed to update the task to "current" time, then its up to
	 * date and ready to go to new CPU/cfs_rq. But we have difficulty in
	 * getting what current time is, so simply throw away the out-of-date
	 * time. This will result in the wakee task is less decayed, but giving
	 * the wakee more load sounds not bad.
	 */
3332 3333
	if (!(se->avg.last_update_time && prev))
		return;
3334 3335

#ifndef CONFIG_64BIT
3336
	{
3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350
		u64 p_last_update_time_copy;
		u64 n_last_update_time_copy;

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

			smp_rmb();

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

		} while (p_last_update_time != p_last_update_time_copy ||
			 n_last_update_time != n_last_update_time_copy);
3351
	}
3352
#else
3353 3354
	p_last_update_time = prev->avg.last_update_time;
	n_last_update_time = next->avg.last_update_time;
3355
#endif
3356 3357
	__update_load_avg_blocked_se(p_last_update_time, cpu_of(rq_of(prev)), se);
	se->avg.last_update_time = n_last_update_time;
3358
}
3359

3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 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

/*
 * When on migration a sched_entity joins/leaves the PELT hierarchy, we need to
 * propagate its contribution. The key to this propagation is the invariant
 * that for each group:
 *
 *   ge->avg == grq->avg						(1)
 *
 * _IFF_ we look at the pure running and runnable sums. Because they
 * represent the very same entity, just at different points in the hierarchy.
 *
 *
 * Per the above update_tg_cfs_util() is trivial (and still 'wrong') and
 * simply copies the running sum over.
 *
 * However, update_tg_cfs_runnable() is more complex. So we have:
 *
 *   ge->avg.load_avg = ge->load.weight * ge->avg.runnable_avg		(2)
 *
 * And since, like util, the runnable part should be directly transferable,
 * the following would _appear_ to be the straight forward approach:
 *
 *   grq->avg.load_avg = grq->load.weight * grq->avg.running_avg	(3)
 *
 * And per (1) we have:
 *
 *   ge->avg.running_avg == grq->avg.running_avg
 *
 * Which gives:
 *
 *                      ge->load.weight * grq->avg.load_avg
 *   ge->avg.load_avg = -----------------------------------		(4)
 *                               grq->load.weight
 *
 * Except that is wrong!
 *
 * Because while for entities historical weight is not important and we
 * really only care about our future and therefore can consider a pure
 * runnable sum, runqueues can NOT do this.
 *
 * We specifically want runqueues to have a load_avg that includes
 * historical weights. Those represent the blocked load, the load we expect
 * to (shortly) return to us. This only works by keeping the weights as
 * integral part of the sum. We therefore cannot decompose as per (3).
 *
 * OK, so what then?
 *
 *
 * Another way to look at things is:
 *
 *   grq->avg.load_avg = \Sum se->avg.load_avg
 *
 * Therefore, per (2):
 *
 *   grq->avg.load_avg = \Sum se->load.weight * se->avg.runnable_avg
 *
 * And the very thing we're propagating is a change in that sum (someone
 * joined/left). So we can easily know the runnable change, which would be, per
 * (2) the already tracked se->load_avg divided by the corresponding
 * se->weight.
 *
 * Basically (4) but in differential form:
 *
 *   d(runnable_avg) += se->avg.load_avg / se->load.weight
 *								   (5)
 *   ge->avg.load_avg += ge->load.weight * d(runnable_avg)
 */

3428
static inline void
3429
update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446
{
	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;

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

	/* Set new sched_entity's utilization */
	se->avg.util_avg = gcfs_rq->avg.util_avg;
	se->avg.util_sum = se->avg.util_avg * LOAD_AVG_MAX;

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

static inline void
3447
update_tg_cfs_runnable(struct cfs_rq *cfs_rq, struct sched_entity *se, struct cfs_rq *gcfs_rq)
3448
{
3449
	long runnable_sum = gcfs_rq->prop_runnable_sum;
3450 3451
	long runnable_load_avg, load_avg;
	s64 runnable_load_sum, load_sum;
3452

3453 3454
	if (!runnable_sum)
		return;
3455

3456
	gcfs_rq->prop_runnable_sum = 0;
3457

3458 3459
	load_sum = (s64)se_weight(se) * runnable_sum;
	load_avg = div_s64(load_sum, LOAD_AVG_MAX);
3460

3461 3462
	add_positive(&se->avg.load_sum, runnable_sum);
	add_positive(&se->avg.load_avg, load_avg);
3463

3464 3465
	add_positive(&cfs_rq->avg.load_avg, load_avg);
	add_positive(&cfs_rq->avg.load_sum, load_sum);
3466

3467 3468 3469 3470 3471 3472
	runnable_load_sum = (s64)se_runnable(se) * runnable_sum;
	runnable_load_avg = div_s64(runnable_load_sum, LOAD_AVG_MAX);

	add_positive(&se->avg.runnable_load_sum, runnable_sum);
	add_positive(&se->avg.runnable_load_avg, runnable_load_avg);

3473
	if (se->on_rq) {
3474 3475
		add_positive(&cfs_rq->avg.runnable_load_avg, runnable_load_avg);
		add_positive(&cfs_rq->avg.runnable_load_sum, runnable_load_sum);
3476 3477 3478
	}
}

3479
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum)
3480
{
3481 3482
	cfs_rq->propagate = 1;
	cfs_rq->prop_runnable_sum += runnable_sum;
3483 3484 3485 3486 3487
}

/* Update task and its cfs_rq load average */
static inline int propagate_entity_load_avg(struct sched_entity *se)
{
3488
	struct cfs_rq *cfs_rq, *gcfs_rq;
3489 3490 3491 3492

	if (entity_is_task(se))
		return 0;

3493 3494
	gcfs_rq = group_cfs_rq(se);
	if (!gcfs_rq->propagate)
3495 3496
		return 0;

3497 3498
	gcfs_rq->propagate = 0;

3499 3500
	cfs_rq = cfs_rq_of(se);

3501
	add_tg_cfs_propagate(cfs_rq, gcfs_rq->prop_runnable_sum);
3502

3503 3504
	update_tg_cfs_util(cfs_rq, se, gcfs_rq);
	update_tg_cfs_runnable(cfs_rq, se, gcfs_rq);
3505 3506 3507 3508

	return 1;
}

3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527
/*
 * Check if we need to update the load and the utilization of a blocked
 * group_entity:
 */
static inline bool skip_blocked_update(struct sched_entity *se)
{
	struct cfs_rq *gcfs_rq = group_cfs_rq(se);

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

	/*
	 * If there is a pending propagation, we have to update the load and
	 * the utilization of the sched_entity:
	 */
3528
	if (gcfs_rq->propagate)
3529 3530 3531 3532 3533 3534 3535 3536 3537 3538
		return false;

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

3539
#else /* CONFIG_FAIR_GROUP_SCHED */
3540

3541
static inline void update_tg_load_avg(struct cfs_rq *cfs_rq, int force) {}
3542 3543 3544 3545 3546 3547

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

3548
static inline void add_tg_cfs_propagate(struct cfs_rq *cfs_rq, long runnable_sum) {}
3549

3550
#endif /* CONFIG_FAIR_GROUP_SCHED */
3551

3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562
/**
 * update_cfs_rq_load_avg - update the cfs_rq's load/util averages
 * @now: current time, as per cfs_rq_clock_task()
 * @cfs_rq: cfs_rq to update
 *
 * The cfs_rq avg is the direct sum of all its entities (blocked and runnable)
 * avg. The immediate corollary is that all (fair) tasks must be attached, see
 * post_init_entity_util_avg().
 *
 * cfs_rq->avg is used for task_h_load() and update_cfs_share() for example.
 *
3563 3564 3565 3566
 * Returns true if the load decayed or we removed load.
 *
 * Since both these conditions indicate a changed cfs_rq->avg.load we should
 * call update_tg_load_avg() when this function returns true.
3567
 */
3568
static inline int
3569
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3570
{
3571
	unsigned long removed_load = 0, removed_util = 0, removed_runnable_sum = 0;
3572
	struct sched_avg *sa = &cfs_rq->avg;
3573
	int decayed = 0;
3574

3575 3576
	if (cfs_rq->removed.nr) {
		unsigned long r;
3577
		u32 divider = LOAD_AVG_MAX - 1024 + sa->period_contrib;
3578 3579 3580 3581

		raw_spin_lock(&cfs_rq->removed.lock);
		swap(cfs_rq->removed.util_avg, removed_util);
		swap(cfs_rq->removed.load_avg, removed_load);
3582
		swap(cfs_rq->removed.runnable_sum, removed_runnable_sum);
3583 3584 3585 3586
		cfs_rq->removed.nr = 0;
		raw_spin_unlock(&cfs_rq->removed.lock);

		r = removed_load;
3587
		sub_positive(&sa->load_avg, r);
3588
		sub_positive(&sa->load_sum, r * divider);
3589

3590
		r = removed_util;
3591
		sub_positive(&sa->util_avg, r);
3592
		sub_positive(&sa->util_sum, r * divider);
3593

3594
		add_tg_cfs_propagate(cfs_rq, -(long)removed_runnable_sum);
3595 3596

		decayed = 1;
3597
	}
3598

3599
	decayed |= __update_load_avg_cfs_rq(now, cpu_of(rq_of(cfs_rq)), cfs_rq);
3600

3601 3602 3603 3604
#ifndef CONFIG_64BIT
	smp_wmb();
	cfs_rq->load_last_update_time_copy = sa->last_update_time;
#endif
3605

3606
	if (decayed)
3607
		cfs_rq_util_change(cfs_rq);
3608

3609
	return decayed;
3610 3611
}

3612 3613 3614 3615 3616 3617 3618 3619
/**
 * attach_entity_load_avg - attach this entity to its cfs_rq load avg
 * @cfs_rq: cfs_rq to attach to
 * @se: sched_entity to attach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3620 3621
static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3622 3623 3624 3625 3626 3627 3628 3629 3630
	u32 divider = LOAD_AVG_MAX - 1024 + cfs_rq->avg.period_contrib;

	/*
	 * When we attach the @se to the @cfs_rq, we must align the decay
	 * window because without that, really weird and wonderful things can
	 * happen.
	 *
	 * XXX illustrate
	 */
3631
	se->avg.last_update_time = cfs_rq->avg.last_update_time;
3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649
	se->avg.period_contrib = cfs_rq->avg.period_contrib;

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

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

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

3650
	enqueue_load_avg(cfs_rq, se);
3651 3652
	cfs_rq->avg.util_avg += se->avg.util_avg;
	cfs_rq->avg.util_sum += se->avg.util_sum;
3653 3654

	add_tg_cfs_propagate(cfs_rq, se->avg.load_sum);
3655 3656

	cfs_rq_util_change(cfs_rq);
3657 3658
}

3659 3660 3661 3662 3663 3664 3665 3666
/**
 * detach_entity_load_avg - detach this entity from its cfs_rq load avg
 * @cfs_rq: cfs_rq to detach from
 * @se: sched_entity to detach
 *
 * Must call update_cfs_rq_load_avg() before this, since we rely on
 * cfs_rq->avg.last_update_time being current.
 */
3667 3668
static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
3669
	dequeue_load_avg(cfs_rq, se);
3670 3671
	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
3672 3673

	add_tg_cfs_propagate(cfs_rq, -se->avg.load_sum);
3674 3675

	cfs_rq_util_change(cfs_rq);
3676 3677
}

3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711
/*
 * Optional action to be done while updating the load average
 */
#define UPDATE_TG	0x1
#define SKIP_AGE_LOAD	0x2
#define DO_ATTACH	0x4

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

	/*
	 * Track task load average for carrying it to new CPU after migrated, and
	 * track group sched_entity load average for task_h_load calc in migration
	 */
	if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
		__update_load_avg_se(now, cpu, cfs_rq, se);

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

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

		attach_entity_load_avg(cfs_rq, se);
		update_tg_load_avg(cfs_rq, 0);

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

3712
#ifndef CONFIG_64BIT
3713 3714
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
3715
	u64 last_update_time_copy;
3716
	u64 last_update_time;
3717

3718 3719 3720 3721 3722
	do {
		last_update_time_copy = cfs_rq->load_last_update_time_copy;
		smp_rmb();
		last_update_time = cfs_rq->avg.last_update_time;
	} while (last_update_time != last_update_time_copy);
3723 3724 3725

	return last_update_time;
}
3726
#else
3727 3728 3729 3730
static inline u64 cfs_rq_last_update_time(struct cfs_rq *cfs_rq)
{
	return cfs_rq->avg.last_update_time;
}
3731 3732
#endif

3733 3734 3735 3736 3737 3738 3739 3740 3741 3742
/*
 * Synchronize entity load avg of dequeued entity without locking
 * the previous rq.
 */
void sync_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
	u64 last_update_time;

	last_update_time = cfs_rq_last_update_time(cfs_rq);
3743
	__update_load_avg_blocked_se(last_update_time, cpu_of(rq_of(cfs_rq)), se);
3744 3745
}

3746 3747 3748 3749 3750 3751 3752
/*
 * Task first catches up with cfs_rq, and then subtract
 * itself from the cfs_rq (task must be off the queue now).
 */
void remove_entity_load_avg(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
3753
	unsigned long flags;
3754 3755

	/*
3756 3757 3758 3759 3760 3761 3762
	 * tasks cannot exit without having gone through wake_up_new_task() ->
	 * post_init_entity_util_avg() which will have added things to the
	 * cfs_rq, so we can remove unconditionally.
	 *
	 * Similarly for groups, they will have passed through
	 * post_init_entity_util_avg() before unregister_sched_fair_group()
	 * calls this.
3763 3764
	 */

3765
	sync_entity_load_avg(se);
3766 3767 3768 3769 3770

	raw_spin_lock_irqsave(&cfs_rq->removed.lock, flags);
	++cfs_rq->removed.nr;
	cfs_rq->removed.util_avg	+= se->avg.util_avg;
	cfs_rq->removed.load_avg	+= se->avg.load_avg;
3771
	cfs_rq->removed.runnable_sum	+= se->avg.load_sum; /* == runnable_sum */
3772
	raw_spin_unlock_irqrestore(&cfs_rq->removed.lock, flags);
3773
}
3774

3775 3776
static inline unsigned long cfs_rq_runnable_load_avg(struct cfs_rq *cfs_rq)
{
3777
	return cfs_rq->avg.runnable_load_avg;
3778 3779 3780 3781 3782 3783 3784
}

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

3785
static int idle_balance(struct rq *this_rq, struct rq_flags *rf);
3786

3787 3788
#else /* CONFIG_SMP */

3789
static inline int
3790
update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
3791 3792 3793 3794
{
	return 0;
}

3795 3796
#define UPDATE_TG	0x0
#define SKIP_AGE_LOAD	0x0
3797
#define DO_ATTACH	0x0
3798

3799
static inline void update_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se, int not_used1)
3800
{
3801
	cfs_rq_util_change(cfs_rq);
3802 3803
}

3804
static inline void remove_entity_load_avg(struct sched_entity *se) {}
3805

3806 3807 3808 3809 3810
static inline void
attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}
static inline void
detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se) {}

3811
static inline int idle_balance(struct rq *rq, struct rq_flags *rf)
3812 3813 3814 3815
{
	return 0;
}

3816
#endif /* CONFIG_SMP */
3817

P
Peter Zijlstra 已提交
3818 3819 3820 3821 3822 3823 3824 3825 3826
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)
3827
		schedstat_inc(cfs_rq->nr_spread_over);
P
Peter Zijlstra 已提交
3828 3829 3830
#endif
}

3831 3832 3833
static void
place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
{
3834
	u64 vruntime = cfs_rq->min_vruntime;
P
Peter Zijlstra 已提交
3835

3836 3837 3838 3839 3840 3841
	/*
	 * 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
3842
	if (initial && sched_feat(START_DEBIT))
3843
		vruntime += sched_vslice(cfs_rq, se);
3844

3845
	/* sleeps up to a single latency don't count. */
3846
	if (!initial) {
3847
		unsigned long thresh = sysctl_sched_latency;
3848

3849 3850 3851 3852 3853 3854
		/*
		 * Halve their sleep time's effect, to allow
		 * for a gentler effect of sleepers:
		 */
		if (sched_feat(GENTLE_FAIR_SLEEPERS))
			thresh >>= 1;
3855

3856
		vruntime -= thresh;
3857 3858
	}

3859
	/* ensure we never gain time by being placed backwards. */
3860
	se->vruntime = max_vruntime(se->vruntime, vruntime);
3861 3862
}

3863 3864
static void check_enqueue_throttle(struct cfs_rq *cfs_rq);

3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876
static inline void check_schedstat_required(void)
{
#ifdef CONFIG_SCHEDSTATS
	if (schedstat_enabled())
		return;

	/* Force schedstat enabled if a dependent tracepoint is active */
	if (trace_sched_stat_wait_enabled()    ||
			trace_sched_stat_sleep_enabled()   ||
			trace_sched_stat_iowait_enabled()  ||
			trace_sched_stat_blocked_enabled() ||
			trace_sched_stat_runtime_enabled())  {
3877
		printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
3878
			     "stat_blocked and stat_runtime require the "
3879
			     "kernel parameter schedstats=enable or "
3880 3881 3882 3883 3884
			     "kernel.sched_schedstats=1\n");
	}
#endif
}

3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903

/*
 * MIGRATION
 *
 *	dequeue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way the vruntime transition between RQs is done when both
 * min_vruntime are up-to-date.
 *
 * WAKEUP (remote)
 *
3904
 *	->migrate_task_rq_fair() (p->state == TASK_WAKING)
3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915
 *	  vruntime -= min_vruntime
 *
 *	enqueue
 *	  update_curr()
 *	    update_min_vruntime()
 *	  vruntime += min_vruntime
 *
 * this way we don't have the most up-to-date min_vruntime on the originating
 * CPU and an up-to-date min_vruntime on the destination CPU.
 */

3916
static void
3917
enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
3918
{
3919 3920 3921
	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
	bool curr = cfs_rq->curr == se;

3922
	/*
3923 3924
	 * If we're the current task, we must renormalise before calling
	 * update_curr().
3925
	 */
3926
	if (renorm && curr)
3927 3928
		se->vruntime += cfs_rq->min_vruntime;

3929 3930
	update_curr(cfs_rq);

3931
	/*
3932 3933 3934 3935
	 * Otherwise, renormalise after, such that we're placed at the current
	 * moment in time, instead of some random moment in the past. Being
	 * placed in the past could significantly boost this task to the
	 * fairness detriment of existing tasks.
3936
	 */
3937 3938 3939
	if (renorm && !curr)
		se->vruntime += cfs_rq->min_vruntime;

3940 3941 3942 3943 3944 3945 3946 3947
	/*
	 * When enqueuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Add its load to cfs_rq->runnable_avg
	 *   - For group_entity, update its weight to reflect the new share of
	 *     its group cfs_rq
	 *   - Add its new weight to cfs_rq->load.weight
	 */
3948
	update_load_avg(cfs_rq, se, UPDATE_TG | DO_ATTACH);
3949
	update_cfs_group(se);
3950
	enqueue_runnable_load_avg(cfs_rq, se);
3951
	account_entity_enqueue(cfs_rq, se);
3952

3953
	if (flags & ENQUEUE_WAKEUP)
3954
		place_entity(cfs_rq, se, 0);
3955

3956
	check_schedstat_required();
3957 3958
	update_stats_enqueue(cfs_rq, se, flags);
	check_spread(cfs_rq, se);
3959
	if (!curr)
3960
		__enqueue_entity(cfs_rq, se);
P
Peter Zijlstra 已提交
3961
	se->on_rq = 1;
3962

3963
	if (cfs_rq->nr_running == 1) {
3964
		list_add_leaf_cfs_rq(cfs_rq);
3965 3966
		check_enqueue_throttle(cfs_rq);
	}
3967 3968
}

3969
static void __clear_buddies_last(struct sched_entity *se)
P
Peter Zijlstra 已提交
3970
{
3971 3972
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3973
		if (cfs_rq->last != se)
3974
			break;
3975 3976

		cfs_rq->last = NULL;
3977 3978
	}
}
P
Peter Zijlstra 已提交
3979

3980 3981 3982 3983
static void __clear_buddies_next(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3984
		if (cfs_rq->next != se)
3985
			break;
3986 3987

		cfs_rq->next = NULL;
3988
	}
P
Peter Zijlstra 已提交
3989 3990
}

3991 3992 3993 3994
static void __clear_buddies_skip(struct sched_entity *se)
{
	for_each_sched_entity(se) {
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
3995
		if (cfs_rq->skip != se)
3996
			break;
3997 3998

		cfs_rq->skip = NULL;
3999 4000 4001
	}
}

P
Peter Zijlstra 已提交
4002 4003
static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
4004 4005 4006 4007 4008
	if (cfs_rq->last == se)
		__clear_buddies_last(se);

	if (cfs_rq->next == se)
		__clear_buddies_next(se);
4009 4010 4011

	if (cfs_rq->skip == se)
		__clear_buddies_skip(se);
P
Peter Zijlstra 已提交
4012 4013
}

4014
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4015

4016
static void
4017
dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
4018
{
4019 4020 4021 4022
	/*
	 * Update run-time statistics of the 'current'.
	 */
	update_curr(cfs_rq);
4023 4024 4025 4026 4027 4028 4029 4030 4031

	/*
	 * When dequeuing a sched_entity, we must:
	 *   - Update loads to have both entity and cfs_rq synced with now.
	 *   - Substract its load from the cfs_rq->runnable_avg.
	 *   - Substract its previous weight from cfs_rq->load.weight.
	 *   - For group entity, update its weight to reflect the new share
	 *     of its group cfs_rq.
	 */
4032
	update_load_avg(cfs_rq, se, UPDATE_TG);
4033
	dequeue_runnable_load_avg(cfs_rq, se);
4034

4035
	update_stats_dequeue(cfs_rq, se, flags);
P
Peter Zijlstra 已提交
4036

P
Peter Zijlstra 已提交
4037
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4038

4039
	if (se != cfs_rq->curr)
4040
		__dequeue_entity(cfs_rq, se);
4041
	se->on_rq = 0;
4042
	account_entity_dequeue(cfs_rq, se);
4043 4044

	/*
4045 4046 4047 4048
	 * Normalize after update_curr(); which will also have moved
	 * min_vruntime if @se is the one holding it back. But before doing
	 * update_min_vruntime() again, which will discount @se's position and
	 * can move min_vruntime forward still more.
4049
	 */
4050
	if (!(flags & DEQUEUE_SLEEP))
4051
		se->vruntime -= cfs_rq->min_vruntime;
4052

4053 4054 4055
	/* return excess runtime on last dequeue */
	return_cfs_rq_runtime(cfs_rq);

4056
	update_cfs_group(se);
4057 4058 4059 4060 4061 4062 4063 4064 4065

	/*
	 * Now advance min_vruntime if @se was the entity holding it back,
	 * except when: DEQUEUE_SAVE && !DEQUEUE_MOVE, in this case we'll be
	 * put back on, and if we advance min_vruntime, we'll be placed back
	 * further than we started -- ie. we'll be penalized.
	 */
	if ((flags & (DEQUEUE_SAVE | DEQUEUE_MOVE)) == DEQUEUE_SAVE)
		update_min_vruntime(cfs_rq);
4066 4067 4068 4069 4070
}

/*
 * Preempt the current task with a newly woken task if needed:
 */
4071
static void
I
Ingo Molnar 已提交
4072
check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4073
{
4074
	unsigned long ideal_runtime, delta_exec;
4075 4076
	struct sched_entity *se;
	s64 delta;
4077

P
Peter Zijlstra 已提交
4078
	ideal_runtime = sched_slice(cfs_rq, curr);
4079
	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
4080
	if (delta_exec > ideal_runtime) {
4081
		resched_curr(rq_of(cfs_rq));
4082 4083 4084 4085 4086
		/*
		 * The current task ran long enough, ensure it doesn't get
		 * re-elected due to buddy favours.
		 */
		clear_buddies(cfs_rq, curr);
4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097
		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;

4098 4099
	se = __pick_first_entity(cfs_rq);
	delta = curr->vruntime - se->vruntime;
4100

4101 4102
	if (delta < 0)
		return;
4103

4104
	if (delta > ideal_runtime)
4105
		resched_curr(rq_of(cfs_rq));
4106 4107
}

4108
static void
4109
set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
4110
{
4111 4112 4113 4114 4115 4116 4117
	/* '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.
		 */
4118
		update_stats_wait_end(cfs_rq, se);
4119
		__dequeue_entity(cfs_rq, se);
4120
		update_load_avg(cfs_rq, se, UPDATE_TG);
4121 4122
	}

4123
	update_stats_curr_start(cfs_rq, se);
4124
	cfs_rq->curr = se;
4125

I
Ingo Molnar 已提交
4126 4127 4128 4129 4130
	/*
	 * 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):
	 */
4131
	if (schedstat_enabled() && rq_of(cfs_rq)->load.weight >= 2*se->load.weight) {
4132 4133 4134
		schedstat_set(se->statistics.slice_max,
			max((u64)schedstat_val(se->statistics.slice_max),
			    se->sum_exec_runtime - se->prev_sum_exec_runtime));
I
Ingo Molnar 已提交
4135
	}
4136

4137
	se->prev_sum_exec_runtime = se->sum_exec_runtime;
4138 4139
}

4140 4141 4142
static int
wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se);

4143 4144 4145 4146 4147 4148 4149
/*
 * 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
 */
4150 4151
static struct sched_entity *
pick_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *curr)
4152
{
4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163
	struct sched_entity *left = __pick_first_entity(cfs_rq);
	struct sched_entity *se;

	/*
	 * If curr is set we have to see if its left of the leftmost entity
	 * still in the tree, provided there was anything in the tree at all.
	 */
	if (!left || (curr && entity_before(curr, left)))
		left = curr;

	se = left; /* ideally we run the leftmost entity */
4164

4165 4166 4167 4168 4169
	/*
	 * Avoid running the skip buddy, if running something else can
	 * be done without getting too unfair.
	 */
	if (cfs_rq->skip == se) {
4170 4171 4172 4173 4174 4175 4176 4177 4178 4179
		struct sched_entity *second;

		if (se == curr) {
			second = __pick_first_entity(cfs_rq);
		} else {
			second = __pick_next_entity(se);
			if (!second || (curr && entity_before(curr, second)))
				second = curr;
		}

4180 4181 4182
		if (second && wakeup_preempt_entity(second, left) < 1)
			se = second;
	}
4183

4184 4185 4186 4187 4188 4189
	/*
	 * 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;

4190 4191 4192 4193 4194 4195
	/*
	 * 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;

4196
	clear_buddies(cfs_rq, se);
P
Peter Zijlstra 已提交
4197 4198

	return se;
4199 4200
}

4201
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq);
4202

4203
static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
4204 4205 4206 4207 4208 4209
{
	/*
	 * If still on the runqueue then deactivate_task()
	 * was not called and update_curr() has to be done:
	 */
	if (prev->on_rq)
4210
		update_curr(cfs_rq);
4211

4212 4213 4214
	/* throttle cfs_rqs exceeding runtime */
	check_cfs_rq_runtime(cfs_rq);

4215
	check_spread(cfs_rq, prev);
4216

4217
	if (prev->on_rq) {
4218
		update_stats_wait_start(cfs_rq, prev);
4219 4220
		/* Put 'current' back into the tree. */
		__enqueue_entity(cfs_rq, prev);
4221
		/* in !on_rq case, update occurred at dequeue */
4222
		update_load_avg(cfs_rq, prev, 0);
4223
	}
4224
	cfs_rq->curr = NULL;
4225 4226
}

P
Peter Zijlstra 已提交
4227 4228
static void
entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
4229 4230
{
	/*
4231
	 * Update run-time statistics of the 'current'.
4232
	 */
4233
	update_curr(cfs_rq);
4234

4235 4236 4237
	/*
	 * Ensure that runnable average is periodically updated.
	 */
4238
	update_load_avg(cfs_rq, curr, UPDATE_TG);
4239
	update_cfs_group(curr);
4240

P
Peter Zijlstra 已提交
4241 4242 4243 4244 4245
#ifdef CONFIG_SCHED_HRTICK
	/*
	 * queued ticks are scheduled to match the slice, so don't bother
	 * validating it and just reschedule.
	 */
4246
	if (queued) {
4247
		resched_curr(rq_of(cfs_rq));
4248 4249
		return;
	}
P
Peter Zijlstra 已提交
4250 4251 4252 4253 4254 4255 4256 4257
	/*
	 * 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
4258
	if (cfs_rq->nr_running > 1)
I
Ingo Molnar 已提交
4259
		check_preempt_tick(cfs_rq, curr);
4260 4261
}

4262 4263 4264 4265 4266 4267

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

#ifdef CONFIG_CFS_BANDWIDTH
4268 4269

#ifdef HAVE_JUMP_LABEL
4270
static struct static_key __cfs_bandwidth_used;
4271 4272 4273

static inline bool cfs_bandwidth_used(void)
{
4274
	return static_key_false(&__cfs_bandwidth_used);
4275 4276
}

4277
void cfs_bandwidth_usage_inc(void)
4278
{
4279 4280 4281 4282 4283 4284
	static_key_slow_inc(&__cfs_bandwidth_used);
}

void cfs_bandwidth_usage_dec(void)
{
	static_key_slow_dec(&__cfs_bandwidth_used);
4285 4286 4287 4288 4289 4290 4291
}
#else /* HAVE_JUMP_LABEL */
static bool cfs_bandwidth_used(void)
{
	return true;
}

4292 4293
void cfs_bandwidth_usage_inc(void) {}
void cfs_bandwidth_usage_dec(void) {}
4294 4295
#endif /* HAVE_JUMP_LABEL */

4296 4297 4298 4299 4300 4301 4302 4303
/*
 * default period for cfs group bandwidth.
 * default: 0.1s, units: nanoseconds
 */
static inline u64 default_cfs_period(void)
{
	return 100000000ULL;
}
4304 4305 4306 4307 4308 4309

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

P
Paul Turner 已提交
4310 4311 4312 4313 4314 4315 4316
/*
 * 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
 */
4317
void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
P
Paul Turner 已提交
4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328
{
	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);
}

4329 4330 4331 4332 4333
static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
{
	return &tg->cfs_bandwidth;
}

4334 4335 4336 4337
/* 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))
4338
		return cfs_rq->throttled_clock_task - cfs_rq->throttled_clock_task_time;
4339

4340
	return rq_clock_task(rq_of(cfs_rq)) - cfs_rq->throttled_clock_task_time;
4341 4342
}

4343 4344
/* returns 0 on failure to allocate runtime */
static int assign_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4345 4346 4347
{
	struct task_group *tg = cfs_rq->tg;
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
P
Paul Turner 已提交
4348
	u64 amount = 0, min_amount, expires;
4349 4350 4351 4352 4353 4354 4355

	/* 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;
4356
	else {
P
Peter Zijlstra 已提交
4357
		start_cfs_bandwidth(cfs_b);
4358 4359 4360 4361 4362 4363

		if (cfs_b->runtime > 0) {
			amount = min(cfs_b->runtime, min_amount);
			cfs_b->runtime -= amount;
			cfs_b->idle = 0;
		}
4364
	}
P
Paul Turner 已提交
4365
	expires = cfs_b->runtime_expires;
4366 4367 4368
	raw_spin_unlock(&cfs_b->lock);

	cfs_rq->runtime_remaining += amount;
P
Paul Turner 已提交
4369 4370 4371 4372 4373 4374 4375
	/*
	 * 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;
4376 4377

	return cfs_rq->runtime_remaining > 0;
4378 4379
}

P
Paul Turner 已提交
4380 4381 4382 4383 4384
/*
 * 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)
4385
{
P
Paul Turner 已提交
4386 4387 4388
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);

	/* if the deadline is ahead of our clock, nothing to do */
4389
	if (likely((s64)(rq_clock(rq_of(cfs_rq)) - cfs_rq->runtime_expires) < 0))
4390 4391
		return;

P
Paul Turner 已提交
4392 4393 4394 4395 4396 4397 4398 4399 4400
	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
4401 4402 4403
	 * whether the global deadline has advanced. It is valid to compare
	 * cfs_b->runtime_expires without any locks since we only care about
	 * exact equality, so a partial write will still work.
P
Paul Turner 已提交
4404 4405
	 */

4406
	if (cfs_rq->runtime_expires != cfs_b->runtime_expires) {
P
Paul Turner 已提交
4407 4408 4409 4410 4411 4412 4413 4414
		/* 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;
	}
}

4415
static void __account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
P
Paul Turner 已提交
4416 4417
{
	/* dock delta_exec before expiring quota (as it could span periods) */
4418
	cfs_rq->runtime_remaining -= delta_exec;
P
Paul Turner 已提交
4419 4420 4421
	expire_cfs_rq_runtime(cfs_rq);

	if (likely(cfs_rq->runtime_remaining > 0))
4422 4423
		return;

4424 4425 4426 4427 4428
	/*
	 * 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))
4429
		resched_curr(rq_of(cfs_rq));
4430 4431
}

4432
static __always_inline
4433
void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec)
4434
{
4435
	if (!cfs_bandwidth_used() || !cfs_rq->runtime_enabled)
4436 4437 4438 4439 4440
		return;

	__account_cfs_rq_runtime(cfs_rq, delta_exec);
}

4441 4442
static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
4443
	return cfs_bandwidth_used() && cfs_rq->throttled;
4444 4445
}

4446 4447 4448
/* check whether cfs_rq, or any parent, is throttled */
static inline int throttled_hierarchy(struct cfs_rq *cfs_rq)
{
4449
	return cfs_bandwidth_used() && cfs_rq->throttle_count;
4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476
}

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

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

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

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

	cfs_rq->throttle_count--;
	if (!cfs_rq->throttle_count) {
4477
		/* adjust cfs_rq_clock_task() */
4478
		cfs_rq->throttled_clock_task_time += rq_clock_task(rq) -
4479
					     cfs_rq->throttled_clock_task;
4480 4481 4482 4483 4484 4485 4486 4487 4488 4489
	}

	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)];

4490 4491
	/* group is entering throttled state, stop time */
	if (!cfs_rq->throttle_count)
4492
		cfs_rq->throttled_clock_task = rq_clock_task(rq);
4493 4494 4495 4496 4497
	cfs_rq->throttle_count++;

	return 0;
}

4498
static void throttle_cfs_rq(struct cfs_rq *cfs_rq)
4499 4500 4501 4502 4503
{
	struct rq *rq = rq_of(cfs_rq);
	struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg);
	struct sched_entity *se;
	long task_delta, dequeue = 1;
P
Peter Zijlstra 已提交
4504
	bool empty;
4505 4506 4507

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

4508
	/* freeze hierarchy runnable averages while throttled */
4509 4510 4511
	rcu_read_lock();
	walk_tg_tree_from(cfs_rq->tg, tg_throttle_down, tg_nop, (void *)rq);
	rcu_read_unlock();
4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528

	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)
4529
		sub_nr_running(rq, task_delta);
4530 4531

	cfs_rq->throttled = 1;
4532
	cfs_rq->throttled_clock = rq_clock(rq);
4533
	raw_spin_lock(&cfs_b->lock);
4534
	empty = list_empty(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4535

4536 4537 4538 4539 4540
	/*
	 * Add to the _head_ of the list, so that an already-started
	 * distribute_cfs_runtime will not see us
	 */
	list_add_rcu(&cfs_rq->throttled_list, &cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4541 4542 4543 4544 4545 4546 4547 4548

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

4549 4550 4551
	raw_spin_unlock(&cfs_b->lock);
}

4552
void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
4553 4554 4555 4556 4557 4558 4559
{
	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;

4560
	se = cfs_rq->tg->se[cpu_of(rq)];
4561 4562

	cfs_rq->throttled = 0;
4563 4564 4565

	update_rq_clock(rq);

4566
	raw_spin_lock(&cfs_b->lock);
4567
	cfs_b->throttled_time += rq_clock(rq) - cfs_rq->throttled_clock;
4568 4569 4570
	list_del_rcu(&cfs_rq->throttled_list);
	raw_spin_unlock(&cfs_b->lock);

4571 4572 4573
	/* update hierarchical throttle state */
	walk_tg_tree_from(cfs_rq->tg, tg_nop, tg_unthrottle_up, (void *)rq);

4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591
	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)
4592
		add_nr_running(rq, task_delta);
4593 4594 4595

	/* determine whether we need to wake up potentially idle cpu */
	if (rq->curr == rq->idle && rq->cfs.nr_running)
4596
		resched_curr(rq);
4597 4598 4599 4600 4601 4602
}

static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
		u64 remaining, u64 expires)
{
	struct cfs_rq *cfs_rq;
4603 4604
	u64 runtime;
	u64 starting_runtime = remaining;
4605 4606 4607 4608 4609

	rcu_read_lock();
	list_for_each_entry_rcu(cfs_rq, &cfs_b->throttled_cfs_rq,
				throttled_list) {
		struct rq *rq = rq_of(cfs_rq);
4610
		struct rq_flags rf;
4611

4612
		rq_lock(rq, &rf);
4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628
		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:
4629
		rq_unlock(rq, &rf);
4630 4631 4632 4633 4634 4635

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

4636
	return starting_runtime - remaining;
4637 4638
}

4639 4640 4641 4642 4643 4644 4645 4646
/*
 * 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)
{
4647
	u64 runtime, runtime_expires;
4648
	int throttled;
4649 4650 4651

	/* no need to continue the timer with no bandwidth constraint */
	if (cfs_b->quota == RUNTIME_INF)
4652
		goto out_deactivate;
4653

4654
	throttled = !list_empty(&cfs_b->throttled_cfs_rq);
4655
	cfs_b->nr_periods += overrun;
4656

4657 4658 4659 4660 4661 4662
	/*
	 * idle depends on !throttled (for the case of a large deficit), and if
	 * we're going inactive then everything else can be deferred
	 */
	if (cfs_b->idle && !throttled)
		goto out_deactivate;
P
Paul Turner 已提交
4663 4664 4665

	__refill_cfs_bandwidth_runtime(cfs_b);

4666 4667 4668
	if (!throttled) {
		/* mark as potentially idle for the upcoming period */
		cfs_b->idle = 1;
4669
		return 0;
4670 4671
	}

4672 4673 4674
	/* account preceding periods in which throttling occurred */
	cfs_b->nr_throttled += overrun;

4675 4676 4677
	runtime_expires = cfs_b->runtime_expires;

	/*
4678 4679 4680 4681 4682
	 * This check is repeated as we are holding onto the new bandwidth while
	 * we unthrottle. This can potentially race with an unthrottled group
	 * trying to acquire new bandwidth from the global pool. This can result
	 * in us over-using our runtime if it is all used during this loop, but
	 * only by limited amounts in that extreme case.
4683
	 */
4684 4685
	while (throttled && cfs_b->runtime > 0) {
		runtime = cfs_b->runtime;
4686 4687 4688 4689 4690 4691 4692
		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);
4693 4694

		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4695
	}
4696

4697 4698 4699 4700 4701 4702 4703
	/*
	 * 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;
4704

4705 4706 4707 4708
	return 0;

out_deactivate:
	return 1;
4709
}
4710

4711 4712 4713 4714 4715 4716 4717
/* 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;

4718 4719 4720 4721
/*
 * Are we near the end of the current quota period?
 *
 * Requires cfs_b->lock for hrtimer_expires_remaining to be safe against the
4722
 * hrtimer base being cleared by hrtimer_start. In the case of
4723 4724
 * migrate_hrtimers, base is never cleared, so we are fine.
 */
4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749
static int runtime_refresh_within(struct cfs_bandwidth *cfs_b, u64 min_expire)
{
	struct hrtimer *refresh_timer = &cfs_b->period_timer;
	u64 remaining;

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

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

	return 0;
}

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

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

P
Peter Zijlstra 已提交
4750 4751 4752
	hrtimer_start(&cfs_b->slack_timer,
			ns_to_ktime(cfs_bandwidth_slack_period),
			HRTIMER_MODE_REL);
4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781
}

/* 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)
{
4782 4783 4784
	if (!cfs_bandwidth_used())
		return;

4785
	if (!cfs_rq->runtime_enabled || cfs_rq->nr_running)
4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800
		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 */
4801 4802 4803
	raw_spin_lock(&cfs_b->lock);
	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
		raw_spin_unlock(&cfs_b->lock);
4804
		return;
4805
	}
4806

4807
	if (cfs_b->quota != RUNTIME_INF && cfs_b->runtime > slice)
4808
		runtime = cfs_b->runtime;
4809

4810 4811 4812 4813 4814 4815 4816 4817 4818 4819
	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)
4820
		cfs_b->runtime -= min(runtime, cfs_b->runtime);
4821 4822 4823
	raw_spin_unlock(&cfs_b->lock);
}

4824 4825 4826 4827 4828 4829 4830
/*
 * 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)
{
4831 4832 4833
	if (!cfs_bandwidth_used())
		return;

4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847
	/* 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);
}

4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861
static void sync_throttle(struct task_group *tg, int cpu)
{
	struct cfs_rq *pcfs_rq, *cfs_rq;

	if (!cfs_bandwidth_used())
		return;

	if (!tg->parent)
		return;

	cfs_rq = tg->cfs_rq[cpu];
	pcfs_rq = tg->parent->cfs_rq[cpu];

	cfs_rq->throttle_count = pcfs_rq->throttle_count;
4862
	cfs_rq->throttled_clock_task = rq_clock_task(cpu_rq(cpu));
4863 4864
}

4865
/* conditionally throttle active cfs_rq's from put_prev_entity() */
4866
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq)
4867
{
4868
	if (!cfs_bandwidth_used())
4869
		return false;
4870

4871
	if (likely(!cfs_rq->runtime_enabled || cfs_rq->runtime_remaining > 0))
4872
		return false;
4873 4874 4875 4876 4877 4878

	/*
	 * 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))
4879
		return true;
4880 4881

	throttle_cfs_rq(cfs_rq);
4882
	return true;
4883
}
4884 4885 4886 4887 4888

static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
{
	struct cfs_bandwidth *cfs_b =
		container_of(timer, struct cfs_bandwidth, slack_timer);
P
Peter Zijlstra 已提交
4889

4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901
	do_sched_cfs_slack_timer(cfs_b);

	return HRTIMER_NORESTART;
}

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

4902
	raw_spin_lock(&cfs_b->lock);
4903
	for (;;) {
P
Peter Zijlstra 已提交
4904
		overrun = hrtimer_forward_now(timer, cfs_b->period);
4905 4906 4907 4908 4909
		if (!overrun)
			break;

		idle = do_sched_cfs_period_timer(cfs_b, overrun);
	}
P
Peter Zijlstra 已提交
4910 4911
	if (idle)
		cfs_b->period_active = 0;
4912
	raw_spin_unlock(&cfs_b->lock);
4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924

	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
}

void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
	raw_spin_lock_init(&cfs_b->lock);
	cfs_b->runtime = 0;
	cfs_b->quota = RUNTIME_INF;
	cfs_b->period = ns_to_ktime(default_cfs_period());

	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
P
Peter Zijlstra 已提交
4925
	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936
	cfs_b->period_timer.function = sched_cfs_period_timer;
	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	cfs_b->slack_timer.function = sched_cfs_slack_timer;
}

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

P
Peter Zijlstra 已提交
4937
void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
4938
{
P
Peter Zijlstra 已提交
4939
	lockdep_assert_held(&cfs_b->lock);
4940

P
Peter Zijlstra 已提交
4941 4942 4943 4944 4945
	if (!cfs_b->period_active) {
		cfs_b->period_active = 1;
		hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
		hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
	}
4946 4947 4948 4949
}

static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
{
4950 4951 4952 4953
	/* init_cfs_bandwidth() was not called */
	if (!cfs_b->throttled_cfs_rq.next)
		return;

4954 4955 4956 4957
	hrtimer_cancel(&cfs_b->period_timer);
	hrtimer_cancel(&cfs_b->slack_timer);
}

4958 4959 4960 4961 4962 4963 4964 4965
/*
 * Both these cpu hotplug callbacks race against unregister_fair_sched_group()
 *
 * The race is harmless, since modifying bandwidth settings of unhooked group
 * bits doesn't do much.
 */

/* cpu online calback */
4966 4967
static void __maybe_unused update_runtime_enabled(struct rq *rq)
{
4968
	struct task_group *tg;
4969

4970 4971 4972 4973 4974 4975
	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
4976 4977 4978 4979 4980

		raw_spin_lock(&cfs_b->lock);
		cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
		raw_spin_unlock(&cfs_b->lock);
	}
4981
	rcu_read_unlock();
4982 4983
}

4984
/* cpu offline callback */
4985
static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
4986
{
4987 4988 4989 4990 4991 4992 4993
	struct task_group *tg;

	lockdep_assert_held(&rq->lock);

	rcu_read_lock();
	list_for_each_entry_rcu(tg, &task_groups, list) {
		struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
4994 4995 4996 4997 4998 4999 5000 5001

		if (!cfs_rq->runtime_enabled)
			continue;

		/*
		 * clock_task is not advancing so we just need to make sure
		 * there's some valid quota amount
		 */
5002
		cfs_rq->runtime_remaining = 1;
5003 5004 5005 5006 5007 5008
		/*
		 * Offline rq is schedulable till cpu is completely disabled
		 * in take_cpu_down(), so we prevent new cfs throttling here.
		 */
		cfs_rq->runtime_enabled = 0;

5009 5010 5011
		if (cfs_rq_throttled(cfs_rq))
			unthrottle_cfs_rq(cfs_rq);
	}
5012
	rcu_read_unlock();
5013 5014 5015
}

#else /* CONFIG_CFS_BANDWIDTH */
5016 5017
static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq)
{
5018
	return rq_clock_task(rq_of(cfs_rq));
5019 5020
}

5021
static void account_cfs_rq_runtime(struct cfs_rq *cfs_rq, u64 delta_exec) {}
5022
static bool check_cfs_rq_runtime(struct cfs_rq *cfs_rq) { return false; }
5023
static void check_enqueue_throttle(struct cfs_rq *cfs_rq) {}
5024
static inline void sync_throttle(struct task_group *tg, int cpu) {}
5025
static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq) {}
5026 5027 5028 5029 5030

static inline int cfs_rq_throttled(struct cfs_rq *cfs_rq)
{
	return 0;
}
5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041

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;
}
5042 5043 5044 5045 5046

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) {}
5047 5048
#endif

5049 5050 5051 5052 5053
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) {}
5054
static inline void update_runtime_enabled(struct rq *rq) {}
5055
static inline void unthrottle_offline_cfs_rqs(struct rq *rq) {}
5056 5057 5058

#endif /* CONFIG_CFS_BANDWIDTH */

5059 5060 5061 5062
/**************************************************
 * CFS operations on tasks:
 */

P
Peter Zijlstra 已提交
5063 5064 5065 5066 5067 5068
#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);

5069
	SCHED_WARN_ON(task_rq(p) != rq);
P
Peter Zijlstra 已提交
5070

5071
	if (rq->cfs.h_nr_running > 1) {
P
Peter Zijlstra 已提交
5072 5073 5074 5075 5076 5077
		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)
5078
				resched_curr(rq);
P
Peter Zijlstra 已提交
5079 5080
			return;
		}
5081
		hrtick_start(rq, delta);
P
Peter Zijlstra 已提交
5082 5083
	}
}
5084 5085 5086 5087 5088 5089 5090 5091 5092 5093

/*
 * 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;

5094
	if (!hrtick_enabled(rq) || curr->sched_class != &fair_sched_class)
5095 5096 5097 5098 5099
		return;

	if (cfs_rq_of(&curr->se)->nr_running < sched_nr_latency)
		hrtick_start_fair(rq, curr);
}
5100
#else /* !CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
5101 5102 5103 5104
static inline void
hrtick_start_fair(struct rq *rq, struct task_struct *p)
{
}
5105 5106 5107 5108

static inline void hrtick_update(struct rq *rq)
{
}
P
Peter Zijlstra 已提交
5109 5110
#endif

5111 5112 5113 5114 5115
/*
 * 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:
 */
5116
static void
5117
enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5118 5119
{
	struct cfs_rq *cfs_rq;
5120
	struct sched_entity *se = &p->se;
5121

5122 5123 5124 5125 5126 5127
	/*
	 * If in_iowait is set, the code below may not trigger any cpufreq
	 * utilization updates, so do it here explicitly with the IOWAIT flag
	 * passed.
	 */
	if (p->in_iowait)
5128
		cpufreq_update_util(rq, SCHED_CPUFREQ_IOWAIT);
5129

5130
	for_each_sched_entity(se) {
5131
		if (se->on_rq)
5132 5133
			break;
		cfs_rq = cfs_rq_of(se);
5134
		enqueue_entity(cfs_rq, se, flags);
5135 5136 5137 5138 5139 5140

		/*
		 * 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.
5141
		 */
5142 5143
		if (cfs_rq_throttled(cfs_rq))
			break;
5144
		cfs_rq->h_nr_running++;
5145

5146
		flags = ENQUEUE_WAKEUP;
5147
	}
P
Peter Zijlstra 已提交
5148

P
Peter Zijlstra 已提交
5149
	for_each_sched_entity(se) {
5150
		cfs_rq = cfs_rq_of(se);
5151
		cfs_rq->h_nr_running++;
P
Peter Zijlstra 已提交
5152

5153 5154 5155
		if (cfs_rq_throttled(cfs_rq))
			break;

5156
		update_load_avg(cfs_rq, se, UPDATE_TG);
5157
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5158 5159
	}

Y
Yuyang Du 已提交
5160
	if (!se)
5161
		add_nr_running(rq, 1);
Y
Yuyang Du 已提交
5162

5163
	hrtick_update(rq);
5164 5165
}

5166 5167
static void set_next_buddy(struct sched_entity *se);

5168 5169 5170 5171 5172
/*
 * The dequeue_task method is called before nr_running is
 * decreased. We remove the task from the rbtree and
 * update the fair scheduling stats:
 */
5173
static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
5174 5175
{
	struct cfs_rq *cfs_rq;
5176
	struct sched_entity *se = &p->se;
5177
	int task_sleep = flags & DEQUEUE_SLEEP;
5178 5179 5180

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
5181
		dequeue_entity(cfs_rq, se, flags);
5182 5183 5184 5185 5186 5187 5188 5189 5190

		/*
		 * 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;
5191
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5192

5193
		/* Don't dequeue parent if it has other entities besides us */
5194
		if (cfs_rq->load.weight) {
5195 5196
			/* Avoid re-evaluating load for this entity: */
			se = parent_entity(se);
5197 5198 5199 5200
			/*
			 * Bias pick_next to pick a task from this cfs_rq, as
			 * p is sleeping when it is within its sched_slice.
			 */
5201 5202
			if (task_sleep && se && !throttled_hierarchy(cfs_rq))
				set_next_buddy(se);
5203
			break;
5204
		}
5205
		flags |= DEQUEUE_SLEEP;
5206
	}
P
Peter Zijlstra 已提交
5207

P
Peter Zijlstra 已提交
5208
	for_each_sched_entity(se) {
5209
		cfs_rq = cfs_rq_of(se);
5210
		cfs_rq->h_nr_running--;
P
Peter Zijlstra 已提交
5211

5212 5213 5214
		if (cfs_rq_throttled(cfs_rq))
			break;

5215
		update_load_avg(cfs_rq, se, UPDATE_TG);
5216
		update_cfs_group(se);
P
Peter Zijlstra 已提交
5217 5218
	}

Y
Yuyang Du 已提交
5219
	if (!se)
5220
		sub_nr_running(rq, 1);
Y
Yuyang Du 已提交
5221

5222
	hrtick_update(rq);
5223 5224
}

5225
#ifdef CONFIG_SMP
5226 5227 5228 5229 5230

/* Working cpumask for: load_balance, load_balance_newidle. */
DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
DEFINE_PER_CPU(cpumask_var_t, select_idle_mask);

5231
#ifdef CONFIG_NO_HZ_COMMON
5232 5233 5234 5235 5236
/*
 * per rq 'load' arrray crap; XXX kill this.
 */

/*
5237
 * The exact cpuload calculated at every tick would be:
5238
 *
5239 5240 5241 5242 5243 5244 5245
 *   load' = (1 - 1/2^i) * load + (1/2^i) * cur_load
 *
 * If a cpu misses updates for n ticks (as it was idle) and update gets
 * called on the n+1-th tick when cpu may be busy, then we have:
 *
 *   load_n   = (1 - 1/2^i)^n * load_0
 *   load_n+1 = (1 - 1/2^i)   * load_n + (1/2^i) * cur_load
5246 5247 5248
 *
 * decay_load_missed() below does efficient calculation of
 *
5249 5250 5251 5252 5253 5254
 *   load' = (1 - 1/2^i)^n * load
 *
 * Because x^(n+m) := x^n * x^m we can decompose any x^n in power-of-2 factors.
 * This allows us to precompute the above in said factors, thereby allowing the
 * reduction of an arbitrary n in O(log_2 n) steps. (See also
 * fixed_power_int())
5255
 *
5256
 * The calculation is approximated on a 128 point scale.
5257 5258
 */
#define DEGRADE_SHIFT		7
5259 5260 5261 5262 5263 5264 5265 5266 5267

static const u8 degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128};
static const u8 degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = {
	{   0,   0,  0,  0,  0,  0, 0, 0 },
	{  64,  32,  8,  0,  0,  0, 0, 0 },
	{  96,  72, 40, 12,  1,  0, 0, 0 },
	{ 112,  98, 75, 43, 15,  1, 0, 0 },
	{ 120, 112, 98, 76, 45, 16, 2, 0 }
};
5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296

/*
 * Update cpu_load for any missed ticks, due to tickless idle. The backlog
 * would be when CPU is idle and so we just decay the old load without
 * adding any new load.
 */
static unsigned long
decay_load_missed(unsigned long load, unsigned long missed_updates, int idx)
{
	int j = 0;

	if (!missed_updates)
		return load;

	if (missed_updates >= degrade_zero_ticks[idx])
		return 0;

	if (idx == 1)
		return load >> missed_updates;

	while (missed_updates) {
		if (missed_updates % 2)
			load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT;

		missed_updates >>= 1;
		j++;
	}
	return load;
}
5297
#endif /* CONFIG_NO_HZ_COMMON */
5298

5299
/**
5300
 * __cpu_load_update - update the rq->cpu_load[] statistics
5301 5302 5303 5304
 * @this_rq: The rq to update statistics for
 * @this_load: The current load
 * @pending_updates: The number of missed updates
 *
5305
 * Update rq->cpu_load[] statistics. This function is usually called every
5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331
 * scheduler tick (TICK_NSEC).
 *
 * This function computes a decaying average:
 *
 *   load[i]' = (1 - 1/2^i) * load[i] + (1/2^i) * load
 *
 * Because of NOHZ it might not get called on every tick which gives need for
 * the @pending_updates argument.
 *
 *   load[i]_n = (1 - 1/2^i) * load[i]_n-1 + (1/2^i) * load_n-1
 *             = A * load[i]_n-1 + B ; A := (1 - 1/2^i), B := (1/2^i) * load
 *             = A * (A * load[i]_n-2 + B) + B
 *             = A * (A * (A * load[i]_n-3 + B) + B) + B
 *             = A^3 * load[i]_n-3 + (A^2 + A + 1) * B
 *             = A^n * load[i]_0 + (A^(n-1) + A^(n-2) + ... + 1) * B
 *             = A^n * load[i]_0 + ((1 - A^n) / (1 - A)) * B
 *             = (1 - 1/2^i)^n * (load[i]_0 - load) + load
 *
 * In the above we've assumed load_n := load, which is true for NOHZ_FULL as
 * any change in load would have resulted in the tick being turned back on.
 *
 * For regular NOHZ, this reduces to:
 *
 *   load[i]_n = (1 - 1/2^i)^n * load[i]_0
 *
 * see decay_load_misses(). For NOHZ_FULL we get to subtract and add the extra
5332
 * term.
5333
 */
5334 5335
static void cpu_load_update(struct rq *this_rq, unsigned long this_load,
			    unsigned long pending_updates)
5336
{
5337
	unsigned long __maybe_unused tickless_load = this_rq->cpu_load[0];
5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348
	int i, scale;

	this_rq->nr_load_updates++;

	/* Update our load: */
	this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */
	for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) {
		unsigned long old_load, new_load;

		/* scale is effectively 1 << i now, and >> i divides by scale */

5349
		old_load = this_rq->cpu_load[i];
5350
#ifdef CONFIG_NO_HZ_COMMON
5351
		old_load = decay_load_missed(old_load, pending_updates - 1, i);
5352 5353 5354 5355 5356 5357 5358 5359 5360
		if (tickless_load) {
			old_load -= decay_load_missed(tickless_load, pending_updates - 1, i);
			/*
			 * old_load can never be a negative value because a
			 * decayed tickless_load cannot be greater than the
			 * original tickless_load.
			 */
			old_load += tickless_load;
		}
5361
#endif
5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376
		new_load = this_load;
		/*
		 * Round up the averaging division if load is increasing. This
		 * prevents us from getting stuck on 9 if the load is 10, for
		 * example.
		 */
		if (new_load > old_load)
			new_load += scale - 1;

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

	sched_avg_update(this_rq);
}

5377
/* Used instead of source_load when we know the type == 0 */
5378
static unsigned long weighted_cpuload(struct rq *rq)
5379
{
5380
	return cfs_rq_runnable_load_avg(&rq->cfs);
5381 5382
}

5383
#ifdef CONFIG_NO_HZ_COMMON
5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400
/*
 * There is no sane way to deal with nohz on smp when using jiffies because the
 * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
 * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
 *
 * Therefore we need to avoid the delta approach from the regular tick when
 * possible since that would seriously skew the load calculation. This is why we
 * use cpu_load_update_periodic() for CPUs out of nohz. However we'll rely on
 * jiffies deltas for updates happening while in nohz mode (idle ticks, idle
 * loop exit, nohz_idle_balance, nohz full exit...)
 *
 * This means we might still be one tick off for nohz periods.
 */

static void cpu_load_update_nohz(struct rq *this_rq,
				 unsigned long curr_jiffies,
				 unsigned long load)
5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411
{
	unsigned long pending_updates;

	pending_updates = curr_jiffies - this_rq->last_load_update_tick;
	if (pending_updates) {
		this_rq->last_load_update_tick = curr_jiffies;
		/*
		 * In the regular NOHZ case, we were idle, this means load 0.
		 * In the NOHZ_FULL case, we were non-idle, we should consider
		 * its weighted load.
		 */
5412
		cpu_load_update(this_rq, load, pending_updates);
5413 5414 5415
	}
}

5416 5417 5418 5419
/*
 * Called from nohz_idle_balance() to update the load ratings before doing the
 * idle balance.
 */
5420
static void cpu_load_update_idle(struct rq *this_rq)
5421 5422 5423 5424
{
	/*
	 * bail if there's load or we're actually up-to-date.
	 */
5425
	if (weighted_cpuload(this_rq))
5426 5427
		return;

5428
	cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), 0);
5429 5430 5431
}

/*
5432 5433 5434 5435
 * Record CPU load on nohz entry so we know the tickless load to account
 * on nohz exit. cpu_load[0] happens then to be updated more frequently
 * than other cpu_load[idx] but it should be fine as cpu_load readers
 * shouldn't rely into synchronized cpu_load[*] updates.
5436
 */
5437
void cpu_load_update_nohz_start(void)
5438 5439
{
	struct rq *this_rq = this_rq();
5440 5441 5442 5443 5444 5445

	/*
	 * This is all lockless but should be fine. If weighted_cpuload changes
	 * concurrently we'll exit nohz. And cpu_load write can race with
	 * cpu_load_update_idle() but both updater would be writing the same.
	 */
5446
	this_rq->cpu_load[0] = weighted_cpuload(this_rq);
5447 5448 5449 5450 5451 5452 5453
}

/*
 * Account the tickless load in the end of a nohz frame.
 */
void cpu_load_update_nohz_stop(void)
{
5454
	unsigned long curr_jiffies = READ_ONCE(jiffies);
5455 5456
	struct rq *this_rq = this_rq();
	unsigned long load;
5457
	struct rq_flags rf;
5458 5459 5460 5461

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

5462
	load = weighted_cpuload(this_rq);
5463
	rq_lock(this_rq, &rf);
5464
	update_rq_clock(this_rq);
5465
	cpu_load_update_nohz(this_rq, curr_jiffies, load);
5466
	rq_unlock(this_rq, &rf);
5467
}
5468 5469 5470 5471 5472 5473 5474 5475
#else /* !CONFIG_NO_HZ_COMMON */
static inline void cpu_load_update_nohz(struct rq *this_rq,
					unsigned long curr_jiffies,
					unsigned long load) { }
#endif /* CONFIG_NO_HZ_COMMON */

static void cpu_load_update_periodic(struct rq *this_rq, unsigned long load)
{
5476
#ifdef CONFIG_NO_HZ_COMMON
5477 5478
	/* See the mess around cpu_load_update_nohz(). */
	this_rq->last_load_update_tick = READ_ONCE(jiffies);
5479
#endif
5480 5481
	cpu_load_update(this_rq, load, 1);
}
5482 5483 5484 5485

/*
 * Called from scheduler_tick()
 */
5486
void cpu_load_update_active(struct rq *this_rq)
5487
{
5488
	unsigned long load = weighted_cpuload(this_rq);
5489 5490 5491 5492 5493

	if (tick_nohz_tick_stopped())
		cpu_load_update_nohz(this_rq, READ_ONCE(jiffies), load);
	else
		cpu_load_update_periodic(this_rq, load);
5494 5495
}

5496 5497 5498 5499 5500 5501 5502 5503 5504 5505
/*
 * 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);
5506
	unsigned long total = weighted_cpuload(rq);
5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520

	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);
5521
	unsigned long total = weighted_cpuload(rq);
5522 5523 5524 5525 5526 5527 5528

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

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

5529
static unsigned long capacity_of(int cpu)
5530
{
5531
	return cpu_rq(cpu)->cpu_capacity;
5532 5533
}

5534 5535 5536 5537 5538
static unsigned long capacity_orig_of(int cpu)
{
	return cpu_rq(cpu)->cpu_capacity_orig;
}

5539 5540 5541
static unsigned long cpu_avg_load_per_task(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5542
	unsigned long nr_running = READ_ONCE(rq->cfs.h_nr_running);
5543
	unsigned long load_avg = weighted_cpuload(rq);
5544 5545

	if (nr_running)
5546
		return load_avg / nr_running;
5547 5548 5549 5550

	return 0;
}

P
Peter Zijlstra 已提交
5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567
static void record_wakee(struct task_struct *p)
{
	/*
	 * Only decay a single time; tasks that have less then 1 wakeup per
	 * jiffy will not have built up many flips.
	 */
	if (time_after(jiffies, current->wakee_flip_decay_ts + HZ)) {
		current->wakee_flips >>= 1;
		current->wakee_flip_decay_ts = jiffies;
	}

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

M
Mike Galbraith 已提交
5568 5569
/*
 * Detect M:N waker/wakee relationships via a switching-frequency heuristic.
P
Peter Zijlstra 已提交
5570
 *
M
Mike Galbraith 已提交
5571
 * A waker of many should wake a different task than the one last awakened
P
Peter Zijlstra 已提交
5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583
 * at a frequency roughly N times higher than one of its wakees.
 *
 * In order to determine whether we should let the load spread vs consolidating
 * to shared cache, we look for a minimum 'flip' frequency of llc_size in one
 * partner, and a factor of lls_size higher frequency in the other.
 *
 * With both conditions met, we can be relatively sure that the relationship is
 * non-monogamous, with partner count exceeding socket size.
 *
 * Waker/wakee being client/server, worker/dispatcher, interrupt source or
 * whatever is irrelevant, spread criteria is apparent partner count exceeds
 * socket size.
M
Mike Galbraith 已提交
5584
 */
5585 5586
static int wake_wide(struct task_struct *p)
{
M
Mike Galbraith 已提交
5587 5588
	unsigned int master = current->wakee_flips;
	unsigned int slave = p->wakee_flips;
5589
	int factor = this_cpu_read(sd_llc_size);
5590

M
Mike Galbraith 已提交
5591 5592 5593 5594 5595
	if (master < slave)
		swap(master, slave);
	if (slave < factor || master < slave * factor)
		return 0;
	return 1;
5596 5597
}

5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665
struct llc_stats {
	unsigned long	nr_running;
	unsigned long	load;
	unsigned long	capacity;
	int		has_capacity;
};

static bool get_llc_stats(struct llc_stats *stats, int cpu)
{
	struct sched_domain_shared *sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));

	if (!sds)
		return false;

	stats->nr_running	= READ_ONCE(sds->nr_running);
	stats->load		= READ_ONCE(sds->load);
	stats->capacity		= READ_ONCE(sds->capacity);
	stats->has_capacity	= stats->nr_running < per_cpu(sd_llc_size, cpu);

	return true;
}

/*
 * Can a task be moved from prev_cpu to this_cpu without causing a load
 * imbalance that would trigger the load balancer?
 *
 * Since we're running on 'stale' values, we might in fact create an imbalance
 * but recomputing these values is expensive, as that'd mean iteration 2 cache
 * domains worth of CPUs.
 */
static bool
wake_affine_llc(struct sched_domain *sd, struct task_struct *p,
		int this_cpu, int prev_cpu, int sync)
{
	struct llc_stats prev_stats, this_stats;
	s64 this_eff_load, prev_eff_load;
	unsigned long task_load;

	if (!get_llc_stats(&prev_stats, prev_cpu) ||
	    !get_llc_stats(&this_stats, this_cpu))
		return false;

	/*
	 * If sync wakeup then subtract the (maximum possible)
	 * effect of the currently running task from the load
	 * of the current LLC.
	 */
	if (sync) {
		unsigned long current_load = task_h_load(current);

		/* in this case load hits 0 and this LLC is considered 'idle' */
		if (current_load > this_stats.load)
			return true;

		this_stats.load -= current_load;
	}

	/*
	 * The has_capacity stuff is not SMT aware, but by trying to balance
	 * the nr_running on both ends we try and fill the domain at equal
	 * rates, thereby first consuming cores before siblings.
	 */

	/* if the old cache has capacity, stay there */
	if (prev_stats.has_capacity && prev_stats.nr_running < this_stats.nr_running+1)
		return false;

	/* if this cache has capacity, come here */
5666
	if (this_stats.has_capacity && this_stats.nr_running+1 < prev_stats.nr_running)
5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686
		return true;

	/*
	 * Check to see if we can move the load without causing too much
	 * imbalance.
	 */
	task_load = task_h_load(p);

	this_eff_load = 100;
	this_eff_load *= prev_stats.capacity;

	prev_eff_load = 100 + (sd->imbalance_pct - 100) / 2;
	prev_eff_load *= this_stats.capacity;

	this_eff_load *= this_stats.load + task_load;
	prev_eff_load *= prev_stats.load - task_load;

	return this_eff_load <= prev_eff_load;
}

5687 5688
static int wake_affine(struct sched_domain *sd, struct task_struct *p,
		       int prev_cpu, int sync)
5689
{
5690
	int this_cpu = smp_processor_id();
5691
	bool affine;
5692

5693
	/*
5694 5695 5696
	 * Default to no affine wakeups; wake_affine() should not effect a task
	 * placement the load-balancer feels inclined to undo. The conservative
	 * option is therefore to not move tasks when they wake up.
5697
	 */
5698 5699 5700 5701 5702 5703 5704 5705 5706
	affine = false;

	/*
	 * If the wakeup is across cache domains, try to evaluate if movement
	 * makes sense, otherwise rely on select_idle_siblings() to do
	 * placement inside the cache domain.
	 */
	if (!cpus_share_cache(prev_cpu, this_cpu))
		affine = wake_affine_llc(sd, p, this_cpu, prev_cpu, sync);
5707

5708
	schedstat_inc(p->se.statistics.nr_wakeups_affine_attempts);
5709 5710 5711 5712
	if (affine) {
		schedstat_inc(sd->ttwu_move_affine);
		schedstat_inc(p->se.statistics.nr_wakeups_affine);
	}
5713

5714
	return affine;
5715 5716
}

5717 5718 5719 5720 5721 5722 5723 5724
static inline int task_util(struct task_struct *p);
static int cpu_util_wake(int cpu, struct task_struct *p);

static unsigned long capacity_spare_wake(int cpu, struct task_struct *p)
{
	return capacity_orig_of(cpu) - cpu_util_wake(cpu, p);
}

5725 5726 5727 5728 5729
/*
 * find_idlest_group finds and returns the least busy CPU group within the
 * domain.
 */
static struct sched_group *
P
Peter Zijlstra 已提交
5730
find_idlest_group(struct sched_domain *sd, struct task_struct *p,
5731
		  int this_cpu, int sd_flag)
5732
{
5733
	struct sched_group *idlest = NULL, *group = sd->groups;
5734
	struct sched_group *most_spare_sg = NULL;
5735 5736
	unsigned long min_runnable_load = ULONG_MAX, this_runnable_load = 0;
	unsigned long min_avg_load = ULONG_MAX, this_avg_load = 0;
5737
	unsigned long most_spare = 0, this_spare = 0;
5738
	int load_idx = sd->forkexec_idx;
5739 5740 5741
	int imbalance_scale = 100 + (sd->imbalance_pct-100)/2;
	unsigned long imbalance = scale_load_down(NICE_0_LOAD) *
				(sd->imbalance_pct-100) / 100;
5742

5743 5744 5745
	if (sd_flag & SD_BALANCE_WAKE)
		load_idx = sd->wake_idx;

5746
	do {
5747 5748
		unsigned long load, avg_load, runnable_load;
		unsigned long spare_cap, max_spare_cap;
5749 5750
		int local_group;
		int i;
5751

5752
		/* Skip over this group if it has no CPUs allowed */
5753
		if (!cpumask_intersects(sched_group_span(group),
5754
					&p->cpus_allowed))
5755 5756 5757
			continue;

		local_group = cpumask_test_cpu(this_cpu,
5758
					       sched_group_span(group));
5759

5760 5761 5762 5763
		/*
		 * Tally up the load of all CPUs in the group and find
		 * the group containing the CPU with most spare capacity.
		 */
5764
		avg_load = 0;
5765
		runnable_load = 0;
5766
		max_spare_cap = 0;
5767

5768
		for_each_cpu(i, sched_group_span(group)) {
5769 5770 5771 5772 5773 5774
			/* Bias balancing toward cpus of our domain */
			if (local_group)
				load = source_load(i, load_idx);
			else
				load = target_load(i, load_idx);

5775 5776 5777
			runnable_load += load;

			avg_load += cfs_rq_load_avg(&cpu_rq(i)->cfs);
5778 5779 5780 5781 5782

			spare_cap = capacity_spare_wake(i, p);

			if (spare_cap > max_spare_cap)
				max_spare_cap = spare_cap;
5783 5784
		}

5785
		/* Adjust by relative CPU capacity of the group */
5786 5787 5788 5789
		avg_load = (avg_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
		runnable_load = (runnable_load * SCHED_CAPACITY_SCALE) /
					group->sgc->capacity;
5790 5791

		if (local_group) {
5792 5793
			this_runnable_load = runnable_load;
			this_avg_load = avg_load;
5794 5795
			this_spare = max_spare_cap;
		} else {
5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810
			if (min_runnable_load > (runnable_load + imbalance)) {
				/*
				 * The runnable load is significantly smaller
				 * so we can pick this new cpu
				 */
				min_runnable_load = runnable_load;
				min_avg_load = avg_load;
				idlest = group;
			} else if ((runnable_load < (min_runnable_load + imbalance)) &&
				   (100*min_avg_load > imbalance_scale*avg_load)) {
				/*
				 * The runnable loads are close so take the
				 * blocked load into account through avg_load.
				 */
				min_avg_load = avg_load;
5811 5812 5813 5814 5815 5816 5817
				idlest = group;
			}

			if (most_spare < max_spare_cap) {
				most_spare = max_spare_cap;
				most_spare_sg = group;
			}
5818 5819 5820
		}
	} while (group = group->next, group != sd->groups);

5821 5822 5823 5824 5825 5826
	/*
	 * The cross-over point between using spare capacity or least load
	 * is too conservative for high utilization tasks on partially
	 * utilized systems if we require spare_capacity > task_util(p),
	 * so we allow for some task stuffing by using
	 * spare_capacity > task_util(p)/2.
5827 5828 5829 5830
	 *
	 * Spare capacity can't be used for fork because the utilization has
	 * not been set yet, we must first select a rq to compute the initial
	 * utilization.
5831
	 */
5832 5833 5834
	if (sd_flag & SD_BALANCE_FORK)
		goto skip_spare;

5835
	if (this_spare > task_util(p) / 2 &&
5836
	    imbalance_scale*this_spare > 100*most_spare)
5837
		return NULL;
5838 5839

	if (most_spare > task_util(p) / 2)
5840 5841
		return most_spare_sg;

5842
skip_spare:
5843 5844 5845 5846
	if (!idlest)
		return NULL;

	if (min_runnable_load > (this_runnable_load + imbalance))
5847
		return NULL;
5848 5849 5850 5851 5852

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

5853 5854 5855 5856 5857 5858 5859 5860 5861 5862
	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;
5863 5864 5865 5866
	unsigned int min_exit_latency = UINT_MAX;
	u64 latest_idle_timestamp = 0;
	int least_loaded_cpu = this_cpu;
	int shallowest_idle_cpu = -1;
5867 5868
	int i;

5869 5870
	/* Check if we have any choice: */
	if (group->group_weight == 1)
5871
		return cpumask_first(sched_group_span(group));
5872

5873
	/* Traverse only the allowed CPUs */
5874
	for_each_cpu_and(i, sched_group_span(group), &p->cpus_allowed) {
5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896
		if (idle_cpu(i)) {
			struct rq *rq = cpu_rq(i);
			struct cpuidle_state *idle = idle_get_state(rq);
			if (idle && idle->exit_latency < min_exit_latency) {
				/*
				 * We give priority to a CPU whose idle state
				 * has the smallest exit latency irrespective
				 * of any idle timestamp.
				 */
				min_exit_latency = idle->exit_latency;
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			} else if ((!idle || idle->exit_latency == min_exit_latency) &&
				   rq->idle_stamp > latest_idle_timestamp) {
				/*
				 * If equal or no active idle state, then
				 * the most recently idled CPU might have
				 * a warmer cache.
				 */
				latest_idle_timestamp = rq->idle_stamp;
				shallowest_idle_cpu = i;
			}
5897
		} else if (shallowest_idle_cpu == -1) {
5898
			load = weighted_cpuload(cpu_rq(i));
5899 5900 5901 5902
			if (load < min_load || (load == min_load && i == this_cpu)) {
				min_load = load;
				least_loaded_cpu = i;
			}
5903 5904 5905
		}
	}

5906
	return shallowest_idle_cpu != -1 ? shallowest_idle_cpu : least_loaded_cpu;
5907
}
5908

5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937
#ifdef CONFIG_SCHED_SMT

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

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

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

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

	return def;
}

/*
 * Scans the local SMT mask to see if the entire core is idle, and records this
 * information in sd_llc_shared->has_idle_cores.
 *
 * Since SMT siblings share all cache levels, inspecting this limited remote
 * state should be fairly cheap.
 */
P
Peter Zijlstra 已提交
5938
void __update_idle_core(struct rq *rq)
5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967
{
	int core = cpu_of(rq);
	int cpu;

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

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

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

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

/*
 * Scan the entire LLC domain for idle cores; this dynamically switches off if
 * there are no idle cores left in the system; tracked through
 * sd_llc->shared->has_idle_cores and enabled through update_idle_core() above.
 */
static int select_idle_core(struct task_struct *p, struct sched_domain *sd, int target)
{
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(select_idle_mask);
5968
	int core, cpu;
5969

P
Peter Zijlstra 已提交
5970 5971 5972
	if (!static_branch_likely(&sched_smt_present))
		return -1;

5973 5974 5975
	if (!test_idle_cores(target, false))
		return -1;

5976
	cpumask_and(cpus, sched_domain_span(sd), &p->cpus_allowed);
5977

5978
	for_each_cpu_wrap(core, cpus, target) {
5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005
		bool idle = true;

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

		if (idle)
			return core;
	}

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

	return -1;
}

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

P
Peter Zijlstra 已提交
6006 6007 6008
	if (!static_branch_likely(&sched_smt_present))
		return -1;

6009
	for_each_cpu(cpu, cpu_smt_mask(target)) {
6010
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036
			continue;
		if (idle_cpu(cpu))
			return cpu;
	}

	return -1;
}

#else /* CONFIG_SCHED_SMT */

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

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

#endif /* CONFIG_SCHED_SMT */

/*
 * Scan the LLC domain for idle CPUs; this is dynamically regulated by
 * comparing the average scan cost (tracked in sd->avg_scan_cost) against the
 * average idle time for this rq (as found in rq->avg_idle).
6037
 */
6038 6039
static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int target)
{
6040
	struct sched_domain *this_sd;
6041
	u64 avg_cost, avg_idle;
6042 6043
	u64 time, cost;
	s64 delta;
6044
	int cpu, nr = INT_MAX;
6045

6046 6047 6048 6049
	this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
	if (!this_sd)
		return -1;

6050 6051 6052 6053
	/*
	 * Due to large variance we need a large fuzz factor; hackbench in
	 * particularly is sensitive here.
	 */
6054 6055 6056 6057
	avg_idle = this_rq()->avg_idle / 512;
	avg_cost = this_sd->avg_scan_cost + 1;

	if (sched_feat(SIS_AVG_CPU) && avg_idle < avg_cost)
6058 6059
		return -1;

6060 6061 6062 6063 6064 6065 6066 6067
	if (sched_feat(SIS_PROP)) {
		u64 span_avg = sd->span_weight * avg_idle;
		if (span_avg > 4*avg_cost)
			nr = div_u64(span_avg, avg_cost);
		else
			nr = 4;
	}

6068 6069
	time = local_clock();

6070
	for_each_cpu_wrap(cpu, sched_domain_span(sd), target) {
6071 6072
		if (!--nr)
			return -1;
6073
		if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088
			continue;
		if (idle_cpu(cpu))
			break;
	}

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

	return cpu;
}

/*
 * Try and locate an idle core/thread in the LLC cache domain.
6089
 */
6090
static int select_idle_sibling(struct task_struct *p, int prev, int target)
6091
{
6092
	struct sched_domain *sd;
6093
	int i;
6094

6095 6096
	if (idle_cpu(target))
		return target;
6097 6098

	/*
6099
	 * If the previous cpu is cache affine and idle, don't be stupid.
6100
	 */
6101 6102
	if (prev != target && cpus_share_cache(prev, target) && idle_cpu(prev))
		return prev;
6103

6104
	sd = rcu_dereference(per_cpu(sd_llc, target));
6105 6106
	if (!sd)
		return target;
6107

6108 6109 6110
	i = select_idle_core(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6111

6112 6113 6114 6115 6116 6117 6118
	i = select_idle_cpu(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;

	i = select_idle_smt(p, sd, target);
	if ((unsigned)i < nr_cpumask_bits)
		return i;
6119

6120 6121
	return target;
}
6122

6123
/*
6124
 * cpu_util returns the amount of capacity of a CPU that is used by CFS
6125
 * tasks. The unit of the return value must be the one of capacity so we can
6126 6127
 * compare the utilization with the capacity of the CPU that is available for
 * CFS task (ie cpu_capacity).
6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147
 *
 * cfs_rq.avg.util_avg is the sum of running time of runnable tasks plus the
 * recent utilization of currently non-runnable tasks on a CPU. It represents
 * the amount of utilization of a CPU in the range [0..capacity_orig] where
 * capacity_orig is the cpu_capacity available at the highest frequency
 * (arch_scale_freq_capacity()).
 * The utilization of a CPU converges towards a sum equal to or less than the
 * current capacity (capacity_curr <= capacity_orig) of the CPU because it is
 * the running time on this CPU scaled by capacity_curr.
 *
 * Nevertheless, cfs_rq.avg.util_avg can be higher than capacity_curr or even
 * higher than capacity_orig because of unfortunate rounding in
 * cfs.avg.util_avg or just after migrating tasks and new task wakeups until
 * the average stabilizes with the new running time. We need to check that the
 * utilization stays within the range of [0..capacity_orig] and cap it if
 * necessary. Without utilization capping, a group could be seen as overloaded
 * (CPU0 utilization at 121% + CPU1 utilization at 80%) whereas CPU1 has 20% of
 * available capacity. We allow utilization to overshoot capacity_curr (but not
 * capacity_orig) as it useful for predicting the capacity required after task
 * migrations (scheduler-driven DVFS).
6148
 */
6149
static int cpu_util(int cpu)
6150
{
6151
	unsigned long util = cpu_rq(cpu)->cfs.avg.util_avg;
6152 6153
	unsigned long capacity = capacity_orig_of(cpu);

6154
	return (util >= capacity) ? capacity : util;
6155
}
6156

6157 6158 6159 6160 6161
static inline int task_util(struct task_struct *p)
{
	return p->se.avg.util_avg;
}

6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179
/*
 * cpu_util_wake: Compute cpu utilization with any contributions from
 * the waking task p removed.
 */
static int cpu_util_wake(int cpu, struct task_struct *p)
{
	unsigned long util, capacity;

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

	capacity = capacity_orig_of(cpu);
	util = max_t(long, cpu_rq(cpu)->cfs.avg.util_avg - task_util(p), 0);

	return (util >= capacity) ? capacity : util;
}

6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197
/*
 * Disable WAKE_AFFINE in the case where task @p doesn't fit in the
 * capacity of either the waking CPU @cpu or the previous CPU @prev_cpu.
 *
 * In that case WAKE_AFFINE doesn't make sense and we'll let
 * BALANCE_WAKE sort things out.
 */
static int wake_cap(struct task_struct *p, int cpu, int prev_cpu)
{
	long min_cap, max_cap;

	min_cap = min(capacity_orig_of(prev_cpu), capacity_orig_of(cpu));
	max_cap = cpu_rq(cpu)->rd->max_cpu_capacity;

	/* Minimum capacity is close to max, no need to abort wake_affine */
	if (max_cap - min_cap < max_cap >> 3)
		return 0;

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

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

6204
/*
6205 6206 6207
 * select_task_rq_fair: Select target runqueue for the waking task in domains
 * that have the 'sd_flag' flag set. In practice, this is SD_BALANCE_WAKE,
 * SD_BALANCE_FORK, or SD_BALANCE_EXEC.
6208
 *
6209 6210
 * Balances load by selecting the idlest cpu in the idlest group, or under
 * certain conditions an idle sibling cpu if the domain has SD_WAKE_AFFINE set.
6211
 *
6212
 * Returns the target cpu number.
6213 6214 6215
 *
 * preempt must be disabled.
 */
6216
static int
6217
select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_flags)
6218
{
6219
	struct sched_domain *tmp, *affine_sd = NULL, *sd = NULL;
6220
	int cpu = smp_processor_id();
M
Mike Galbraith 已提交
6221
	int new_cpu = prev_cpu;
6222
	int want_affine = 0;
6223
	int sync = wake_flags & WF_SYNC;
6224

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

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

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

6246
		if (tmp->flags & sd_flag)
6247
			sd = tmp;
M
Mike Galbraith 已提交
6248 6249
		else if (!want_affine)
			break;
6250 6251
	}

M
Mike Galbraith 已提交
6252 6253
	if (affine_sd) {
		sd = NULL; /* Prefer wake_affine over balance flags */
6254 6255 6256 6257
		if (cpu == prev_cpu)
			goto pick_cpu;

		if (wake_affine(affine_sd, p, prev_cpu, sync))
M
Mike Galbraith 已提交
6258
			new_cpu = cpu;
6259
	}
6260

M
Mike Galbraith 已提交
6261
	if (!sd) {
6262
 pick_cpu:
M
Mike Galbraith 已提交
6263
		if (sd_flag & SD_BALANCE_WAKE) /* XXX always ? */
6264
			new_cpu = select_idle_sibling(p, prev_cpu, new_cpu);
M
Mike Galbraith 已提交
6265 6266

	} else while (sd) {
6267
		struct sched_group *group;
6268
		int weight;
6269

6270
		if (!(sd->flags & sd_flag)) {
6271 6272 6273
			sd = sd->child;
			continue;
		}
6274

6275
		group = find_idlest_group(sd, p, cpu, sd_flag);
6276 6277 6278 6279
		if (!group) {
			sd = sd->child;
			continue;
		}
I
Ingo Molnar 已提交
6280

6281
		new_cpu = find_idlest_cpu(group, p, cpu);
6282 6283 6284 6285
		if (new_cpu == -1 || new_cpu == cpu) {
			/* Now try balancing at a lower domain level of cpu */
			sd = sd->child;
			continue;
6286
		}
6287 6288 6289

		/* Now try balancing at a lower domain level of new_cpu */
		cpu = new_cpu;
6290
		weight = sd->span_weight;
6291 6292
		sd = NULL;
		for_each_domain(cpu, tmp) {
6293
			if (weight <= tmp->span_weight)
6294
				break;
6295
			if (tmp->flags & sd_flag)
6296 6297 6298
				sd = tmp;
		}
		/* while loop will break here if sd == NULL */
6299
	}
6300
	rcu_read_unlock();
6301

6302
	return new_cpu;
6303
}
6304

6305 6306
static void detach_entity_cfs_rq(struct sched_entity *se);

6307 6308 6309
/*
 * 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
6310
 * previous cpu. The caller guarantees p->pi_lock or task_rq(p)->lock is held.
6311
 */
6312
static void migrate_task_rq_fair(struct task_struct *p)
6313
{
6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339
	/*
	 * As blocked tasks retain absolute vruntime the migration needs to
	 * deal with this by subtracting the old and adding the new
	 * min_vruntime -- the latter is done by enqueue_entity() when placing
	 * the task on the new runqueue.
	 */
	if (p->state == TASK_WAKING) {
		struct sched_entity *se = &p->se;
		struct cfs_rq *cfs_rq = cfs_rq_of(se);
		u64 min_vruntime;

#ifndef CONFIG_64BIT
		u64 min_vruntime_copy;

		do {
			min_vruntime_copy = cfs_rq->min_vruntime_copy;
			smp_rmb();
			min_vruntime = cfs_rq->min_vruntime;
		} while (min_vruntime != min_vruntime_copy);
#else
		min_vruntime = cfs_rq->min_vruntime;
#endif

		se->vruntime -= min_vruntime;
	}

6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358
	if (p->on_rq == TASK_ON_RQ_MIGRATING) {
		/*
		 * In case of TASK_ON_RQ_MIGRATING we in fact hold the 'old'
		 * rq->lock and can modify state directly.
		 */
		lockdep_assert_held(&task_rq(p)->lock);
		detach_entity_cfs_rq(&p->se);

	} else {
		/*
		 * We are supposed to update the task to "current" time, then
		 * its up to date and ready to go to new CPU/cfs_rq. But we
		 * have difficulty in getting what current time is, so simply
		 * throw away the out-of-date time. This will result in the
		 * wakee task is less decayed, but giving the wakee more load
		 * sounds not bad.
		 */
		remove_entity_load_avg(&p->se);
	}
6359 6360 6361

	/* Tell new CPU we are migrated */
	p->se.avg.last_update_time = 0;
6362 6363

	/* We have migrated, no longer consider this task hot */
6364
	p->se.exec_start = 0;
6365
}
6366 6367 6368 6369 6370

static void task_dead_fair(struct task_struct *p)
{
	remove_entity_load_avg(&p->se);
}
6371 6372
#endif /* CONFIG_SMP */

P
Peter Zijlstra 已提交
6373 6374
static unsigned long
wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
6375 6376 6377 6378
{
	unsigned long gran = sysctl_sched_wakeup_granularity;

	/*
P
Peter Zijlstra 已提交
6379 6380
	 * Since its curr running now, convert the gran from real-time
	 * to virtual-time in his units.
M
Mike Galbraith 已提交
6381 6382 6383 6384 6385 6386 6387 6388 6389
	 *
	 * 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.
6390
	 */
6391
	return calc_delta_fair(gran, se);
6392 6393
}

6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415
/*
 * 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
6416
	gran = wakeup_gran(curr, se);
6417 6418 6419 6420 6421 6422
	if (vdiff > gran)
		return 1;

	return 0;
}

6423 6424
static void set_last_buddy(struct sched_entity *se)
{
6425 6426 6427
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6428 6429 6430
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6431
		cfs_rq_of(se)->last = se;
6432
	}
6433 6434 6435 6436
}

static void set_next_buddy(struct sched_entity *se)
{
6437 6438 6439
	if (entity_is_task(se) && unlikely(task_of(se)->policy == SCHED_IDLE))
		return;

6440 6441 6442
	for_each_sched_entity(se) {
		if (SCHED_WARN_ON(!se->on_rq))
			return;
6443
		cfs_rq_of(se)->next = se;
6444
	}
6445 6446
}

6447 6448
static void set_skip_buddy(struct sched_entity *se)
{
6449 6450
	for_each_sched_entity(se)
		cfs_rq_of(se)->skip = se;
6451 6452
}

6453 6454 6455
/*
 * Preempt the current task with a newly woken task if needed:
 */
6456
static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
6457 6458
{
	struct task_struct *curr = rq->curr;
6459
	struct sched_entity *se = &curr->se, *pse = &p->se;
6460
	struct cfs_rq *cfs_rq = task_cfs_rq(curr);
6461
	int scale = cfs_rq->nr_running >= sched_nr_latency;
6462
	int next_buddy_marked = 0;
6463

I
Ingo Molnar 已提交
6464 6465 6466
	if (unlikely(se == pse))
		return;

6467
	/*
6468
	 * This is possible from callers such as attach_tasks(), in which we
6469 6470 6471 6472 6473 6474 6475
	 * 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;

6476
	if (sched_feat(NEXT_BUDDY) && scale && !(wake_flags & WF_FORK)) {
M
Mike Galbraith 已提交
6477
		set_next_buddy(pse);
6478 6479
		next_buddy_marked = 1;
	}
P
Peter Zijlstra 已提交
6480

6481 6482 6483
	/*
	 * We can come here with TIF_NEED_RESCHED already set from new task
	 * wake up path.
6484 6485 6486 6487 6488 6489
	 *
	 * 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.
6490 6491 6492 6493
	 */
	if (test_tsk_need_resched(curr))
		return;

6494 6495 6496 6497 6498
	/* Idle tasks are by definition preempted by non-idle tasks. */
	if (unlikely(curr->policy == SCHED_IDLE) &&
	    likely(p->policy != SCHED_IDLE))
		goto preempt;

6499
	/*
6500 6501
	 * Batch and idle tasks do not preempt non-idle tasks (their preemption
	 * is driven by the tick):
6502
	 */
6503
	if (unlikely(p->policy != SCHED_NORMAL) || !sched_feat(WAKEUP_PREEMPTION))
6504
		return;
6505

6506
	find_matching_se(&se, &pse);
6507
	update_curr(cfs_rq_of(se));
6508
	BUG_ON(!pse);
6509 6510 6511 6512 6513 6514 6515
	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);
6516
		goto preempt;
6517
	}
6518

6519
	return;
6520

6521
preempt:
6522
	resched_curr(rq);
6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536
	/*
	 * 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);
6537 6538
}

6539
static struct task_struct *
6540
pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
6541 6542 6543
{
	struct cfs_rq *cfs_rq = &rq->cfs;
	struct sched_entity *se;
6544
	struct task_struct *p;
6545
	int new_tasks;
6546

6547
again:
6548
	if (!cfs_rq->nr_running)
6549
		goto idle;
6550

6551
#ifdef CONFIG_FAIR_GROUP_SCHED
6552
	if (prev->sched_class != &fair_sched_class)
6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571
		goto simple;

	/*
	 * Because of the set_next_buddy() in dequeue_task_fair() it is rather
	 * likely that a next task is from the same cgroup as the current.
	 *
	 * Therefore attempt to avoid putting and setting the entire cgroup
	 * hierarchy, only change the part that actually changes.
	 */

	do {
		struct sched_entity *curr = cfs_rq->curr;

		/*
		 * Since we got here without doing put_prev_entity() we also
		 * have to consider cfs_rq->curr. If it is still a runnable
		 * entity, update_curr() will update its vruntime, otherwise
		 * forget we've ever seen it.
		 */
6572 6573 6574 6575 6576
		if (curr) {
			if (curr->on_rq)
				update_curr(cfs_rq);
			else
				curr = NULL;
6577

6578 6579 6580
			/*
			 * This call to check_cfs_rq_runtime() will do the
			 * throttle and dequeue its entity in the parent(s).
6581
			 * Therefore the nr_running test will indeed
6582 6583
			 * be correct.
			 */
6584 6585 6586 6587 6588 6589
			if (unlikely(check_cfs_rq_runtime(cfs_rq))) {
				cfs_rq = &rq->cfs;

				if (!cfs_rq->nr_running)
					goto idle;

6590
				goto simple;
6591
			}
6592
		}
6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631

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

	p = task_of(se);

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

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

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

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

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

	return p;
simple:
#endif
6632

6633
	put_prev_task(rq, prev);
6634

6635
	do {
6636
		se = pick_next_entity(cfs_rq, NULL);
6637
		set_next_entity(cfs_rq, se);
6638 6639 6640
		cfs_rq = group_cfs_rq(se);
	} while (cfs_rq);

P
Peter Zijlstra 已提交
6641
	p = task_of(se);
6642

6643 6644
	if (hrtick_enabled(rq))
		hrtick_start_fair(rq, p);
P
Peter Zijlstra 已提交
6645 6646

	return p;
6647 6648

idle:
6649 6650
	new_tasks = idle_balance(rq, rf);

6651 6652 6653 6654 6655
	/*
	 * Because idle_balance() releases (and re-acquires) rq->lock, it is
	 * possible for any higher priority task to appear. In that case we
	 * must re-start the pick_next_entity() loop.
	 */
6656
	if (new_tasks < 0)
6657 6658
		return RETRY_TASK;

6659
	if (new_tasks > 0)
6660 6661 6662
		goto again;

	return NULL;
6663 6664 6665 6666 6667
}

/*
 * Account for a descheduled task:
 */
6668
static void put_prev_task_fair(struct rq *rq, struct task_struct *prev)
6669 6670 6671 6672 6673 6674
{
	struct sched_entity *se = &prev->se;
	struct cfs_rq *cfs_rq;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);
6675
		put_prev_entity(cfs_rq, se);
6676 6677 6678
	}
}

6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703
/*
 * 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);
6704 6705 6706 6707 6708
		/*
		 * Tell update_rq_clock() that we've just updated,
		 * so we don't do microscopic update in schedule()
		 * and double the fastpath cost.
		 */
6709
		rq_clock_skip_update(rq, true);
6710 6711 6712 6713 6714
	}

	set_skip_buddy(se);
}

6715 6716 6717 6718
static bool yield_to_task_fair(struct rq *rq, struct task_struct *p, bool preempt)
{
	struct sched_entity *se = &p->se;

6719 6720
	/* throttled hierarchies are not runnable */
	if (!se->on_rq || throttled_hierarchy(cfs_rq_of(se)))
6721 6722 6723 6724 6725 6726 6727 6728 6729 6730
		return false;

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

	yield_task_fair(rq);

	return true;
}

6731
#ifdef CONFIG_SMP
6732
/**************************************************
P
Peter Zijlstra 已提交
6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748
 * 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
6749
 * is derived from the nice value as per sched_prio_to_weight[].
P
Peter Zijlstra 已提交
6750 6751 6752 6753 6754 6755
 *
 * 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)
 *
6756
 * C_i is the compute capacity of cpu i, typically it is the
P
Peter Zijlstra 已提交
6757 6758 6759 6760 6761 6762
 * 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):
 *
6763
 *   imb_i,j = max{ avg(W/C), W_i/C_i } - min{ avg(W/C), W_j/C_j }    (4)
P
Peter Zijlstra 已提交
6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801
 *
 * 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:
 *
6802
 *             log_2 n
P
Peter Zijlstra 已提交
6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847
 *   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.]
6848
 */
6849

6850 6851
static unsigned long __read_mostly max_load_balance_interval = HZ/10;

6852 6853
enum fbq_type { regular, remote, all };

6854
#define LBF_ALL_PINNED	0x01
6855
#define LBF_NEED_BREAK	0x02
6856 6857
#define LBF_DST_PINNED  0x04
#define LBF_SOME_PINNED	0x08
6858 6859 6860 6861 6862

struct lb_env {
	struct sched_domain	*sd;

	struct rq		*src_rq;
6863
	int			src_cpu;
6864 6865 6866 6867

	int			dst_cpu;
	struct rq		*dst_rq;

6868 6869
	struct cpumask		*dst_grpmask;
	int			new_dst_cpu;
6870
	enum cpu_idle_type	idle;
6871
	long			imbalance;
6872 6873 6874
	/* The set of CPUs under consideration for load-balancing */
	struct cpumask		*cpus;

6875
	unsigned int		flags;
6876 6877 6878 6879

	unsigned int		loop;
	unsigned int		loop_break;
	unsigned int		loop_max;
6880 6881

	enum fbq_type		fbq_type;
6882
	struct list_head	tasks;
6883 6884
};

6885 6886 6887
/*
 * Is this task likely cache-hot:
 */
6888
static int task_hot(struct task_struct *p, struct lb_env *env)
6889 6890 6891
{
	s64 delta;

6892 6893
	lockdep_assert_held(&env->src_rq->lock);

6894 6895 6896 6897 6898 6899 6900 6901 6902
	if (p->sched_class != &fair_sched_class)
		return 0;

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

	/*
	 * Buddy candidates are cache hot:
	 */
6903
	if (sched_feat(CACHE_HOT_BUDDY) && env->dst_rq->nr_running &&
6904 6905 6906 6907 6908 6909 6910 6911 6912
			(&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;

6913
	delta = rq_clock_task(env->src_rq) - p->se.exec_start;
6914 6915 6916 6917

	return delta < (s64)sysctl_sched_migration_cost;
}

6918
#ifdef CONFIG_NUMA_BALANCING
6919
/*
6920 6921 6922
 * Returns 1, if task migration degrades locality
 * Returns 0, if task migration improves locality i.e migration preferred.
 * Returns -1, if task migration is not affected by locality.
6923
 */
6924
static int migrate_degrades_locality(struct task_struct *p, struct lb_env *env)
6925
{
6926
	struct numa_group *numa_group = rcu_dereference(p->numa_group);
6927
	unsigned long src_faults, dst_faults;
6928 6929
	int src_nid, dst_nid;

6930
	if (!static_branch_likely(&sched_numa_balancing))
6931 6932
		return -1;

6933
	if (!p->numa_faults || !(env->sd->flags & SD_NUMA))
6934
		return -1;
6935 6936 6937 6938

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

6939
	if (src_nid == dst_nid)
6940
		return -1;
6941

6942 6943 6944 6945 6946 6947 6948
	/* Migrating away from the preferred node is always bad. */
	if (src_nid == p->numa_preferred_nid) {
		if (env->src_rq->nr_running > env->src_rq->nr_preferred_running)
			return 1;
		else
			return -1;
	}
6949

6950 6951
	/* Encourage migration to the preferred node. */
	if (dst_nid == p->numa_preferred_nid)
6952
		return 0;
6953

6954 6955 6956 6957
	/* Leaving a core idle is often worse than degrading locality. */
	if (env->idle != CPU_NOT_IDLE)
		return -1;

6958 6959 6960 6961 6962 6963
	if (numa_group) {
		src_faults = group_faults(p, src_nid);
		dst_faults = group_faults(p, dst_nid);
	} else {
		src_faults = task_faults(p, src_nid);
		dst_faults = task_faults(p, dst_nid);
6964 6965
	}

6966
	return dst_faults < src_faults;
6967 6968
}

6969
#else
6970
static inline int migrate_degrades_locality(struct task_struct *p,
6971 6972
					     struct lb_env *env)
{
6973
	return -1;
6974
}
6975 6976
#endif

6977 6978 6979 6980
/*
 * can_migrate_task - may task p from runqueue rq be migrated to this_cpu?
 */
static
6981
int can_migrate_task(struct task_struct *p, struct lb_env *env)
6982
{
6983
	int tsk_cache_hot;
6984 6985 6986

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

6987 6988
	/*
	 * We do not migrate tasks that are:
6989
	 * 1) throttled_lb_pair, or
6990
	 * 2) cannot be migrated to this CPU due to cpus_allowed, or
6991 6992
	 * 3) running (obviously), or
	 * 4) are cache-hot on their current CPU.
6993
	 */
6994 6995 6996
	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
		return 0;

6997
	if (!cpumask_test_cpu(env->dst_cpu, &p->cpus_allowed)) {
6998
		int cpu;
6999

7000
		schedstat_inc(p->se.statistics.nr_failed_migrations_affine);
7001

7002 7003
		env->flags |= LBF_SOME_PINNED;

7004 7005 7006 7007 7008
		/*
		 * 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.
		 *
7009 7010
		 * Avoid computing new_dst_cpu for NEWLY_IDLE or if we have
		 * already computed one in current iteration.
7011
		 */
7012
		if (env->idle == CPU_NEWLY_IDLE || (env->flags & LBF_DST_PINNED))
7013 7014
			return 0;

7015 7016
		/* Prevent to re-select dst_cpu via env's cpus */
		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
7017
			if (cpumask_test_cpu(cpu, &p->cpus_allowed)) {
7018
				env->flags |= LBF_DST_PINNED;
7019 7020 7021
				env->new_dst_cpu = cpu;
				break;
			}
7022
		}
7023

7024 7025
		return 0;
	}
7026 7027

	/* Record that we found atleast one task that could run on dst_cpu */
7028
	env->flags &= ~LBF_ALL_PINNED;
7029

7030
	if (task_running(env->src_rq, p)) {
7031
		schedstat_inc(p->se.statistics.nr_failed_migrations_running);
7032 7033 7034 7035 7036
		return 0;
	}

	/*
	 * Aggressive migration if:
7037 7038 7039
	 * 1) destination numa is preferred
	 * 2) task is cache cold, or
	 * 3) too many balance attempts have failed.
7040
	 */
7041 7042 7043
	tsk_cache_hot = migrate_degrades_locality(p, env);
	if (tsk_cache_hot == -1)
		tsk_cache_hot = task_hot(p, env);
7044

7045
	if (tsk_cache_hot <= 0 ||
7046
	    env->sd->nr_balance_failed > env->sd->cache_nice_tries) {
7047
		if (tsk_cache_hot == 1) {
7048 7049
			schedstat_inc(env->sd->lb_hot_gained[env->idle]);
			schedstat_inc(p->se.statistics.nr_forced_migrations);
7050
		}
7051 7052 7053
		return 1;
	}

7054
	schedstat_inc(p->se.statistics.nr_failed_migrations_hot);
Z
Zhang Hang 已提交
7055
	return 0;
7056 7057
}

7058
/*
7059 7060 7061 7062 7063 7064 7065
 * detach_task() -- detach the task for the migration specified in env
 */
static void detach_task(struct task_struct *p, struct lb_env *env)
{
	lockdep_assert_held(&env->src_rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
7066
	deactivate_task(env->src_rq, p, DEQUEUE_NOCLOCK);
7067 7068 7069
	set_task_cpu(p, env->dst_cpu);
}

7070
/*
7071
 * detach_one_task() -- tries to dequeue exactly one task from env->src_rq, as
7072 7073
 * part of active balancing operations within "domain".
 *
7074
 * Returns a task if successful and NULL otherwise.
7075
 */
7076
static struct task_struct *detach_one_task(struct lb_env *env)
7077 7078 7079
{
	struct task_struct *p, *n;

7080 7081
	lockdep_assert_held(&env->src_rq->lock);

7082 7083 7084
	list_for_each_entry_safe(p, n, &env->src_rq->cfs_tasks, se.group_node) {
		if (!can_migrate_task(p, env))
			continue;
7085

7086
		detach_task(p, env);
7087

7088
		/*
7089
		 * Right now, this is only the second place where
7090
		 * lb_gained[env->idle] is updated (other is detach_tasks)
7091
		 * so we can safely collect stats here rather than
7092
		 * inside detach_tasks().
7093
		 */
7094
		schedstat_inc(env->sd->lb_gained[env->idle]);
7095
		return p;
7096
	}
7097
	return NULL;
7098 7099
}

7100 7101
static const unsigned int sched_nr_migrate_break = 32;

7102
/*
7103 7104
 * detach_tasks() -- tries to detach up to imbalance weighted load from
 * busiest_rq, as part of a balancing operation within domain "sd".
7105
 *
7106
 * Returns number of detached tasks if successful and 0 otherwise.
7107
 */
7108
static int detach_tasks(struct lb_env *env)
7109
{
7110 7111
	struct list_head *tasks = &env->src_rq->cfs_tasks;
	struct task_struct *p;
7112
	unsigned long load;
7113 7114 7115
	int detached = 0;

	lockdep_assert_held(&env->src_rq->lock);
7116

7117
	if (env->imbalance <= 0)
7118
		return 0;
7119

7120
	while (!list_empty(tasks)) {
7121 7122 7123 7124 7125 7126 7127
		/*
		 * We don't want to steal all, otherwise we may be treated likewise,
		 * which could at worst lead to a livelock crash.
		 */
		if (env->idle != CPU_NOT_IDLE && env->src_rq->nr_running <= 1)
			break;

7128
		p = list_first_entry(tasks, struct task_struct, se.group_node);
7129

7130 7131
		env->loop++;
		/* We've more or less seen every task there is, call it quits */
7132
		if (env->loop > env->loop_max)
7133
			break;
7134 7135

		/* take a breather every nr_migrate tasks */
7136
		if (env->loop > env->loop_break) {
7137
			env->loop_break += sched_nr_migrate_break;
7138
			env->flags |= LBF_NEED_BREAK;
7139
			break;
7140
		}
7141

7142
		if (!can_migrate_task(p, env))
7143 7144 7145
			goto next;

		load = task_h_load(p);
7146

7147
		if (sched_feat(LB_MIN) && load < 16 && !env->sd->nr_balance_failed)
7148 7149
			goto next;

7150
		if ((load / 2) > env->imbalance)
7151
			goto next;
7152

7153 7154 7155 7156
		detach_task(p, env);
		list_add(&p->se.group_node, &env->tasks);

		detached++;
7157
		env->imbalance -= load;
7158 7159

#ifdef CONFIG_PREEMPT
7160 7161
		/*
		 * NEWIDLE balancing is a source of latency, so preemptible
7162
		 * kernels will stop after the first task is detached to minimize
7163 7164
		 * the critical section.
		 */
7165
		if (env->idle == CPU_NEWLY_IDLE)
7166
			break;
7167 7168
#endif

7169 7170 7171 7172
		/*
		 * We only want to steal up to the prescribed amount of
		 * weighted load.
		 */
7173
		if (env->imbalance <= 0)
7174
			break;
7175 7176 7177

		continue;
next:
7178
		list_move_tail(&p->se.group_node, tasks);
7179
	}
7180

7181
	/*
7182 7183 7184
	 * Right now, this is one of only two places we collect this stat
	 * so we can safely collect detach_one_task() stats here rather
	 * than inside detach_one_task().
7185
	 */
7186
	schedstat_add(env->sd->lb_gained[env->idle], detached);
7187

7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198
	return detached;
}

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

	BUG_ON(task_rq(p) != rq);
7199
	activate_task(rq, p, ENQUEUE_NOCLOCK);
7200
	p->on_rq = TASK_ON_RQ_QUEUED;
7201 7202 7203 7204 7205 7206 7207 7208 7209
	check_preempt_curr(rq, p, 0);
}

/*
 * attach_one_task() -- attaches the task returned from detach_one_task() to
 * its new rq.
 */
static void attach_one_task(struct rq *rq, struct task_struct *p)
{
7210 7211 7212
	struct rq_flags rf;

	rq_lock(rq, &rf);
7213
	update_rq_clock(rq);
7214
	attach_task(rq, p);
7215
	rq_unlock(rq, &rf);
7216 7217 7218 7219 7220 7221 7222 7223 7224 7225
}

/*
 * attach_tasks() -- attaches all tasks detached by detach_tasks() to their
 * new rq.
 */
static void attach_tasks(struct lb_env *env)
{
	struct list_head *tasks = &env->tasks;
	struct task_struct *p;
7226
	struct rq_flags rf;
7227

7228
	rq_lock(env->dst_rq, &rf);
7229
	update_rq_clock(env->dst_rq);
7230 7231 7232 7233

	while (!list_empty(tasks)) {
		p = list_first_entry(tasks, struct task_struct, se.group_node);
		list_del_init(&p->se.group_node);
7234

7235 7236 7237
		attach_task(env->dst_rq, p);
	}

7238
	rq_unlock(env->dst_rq, &rf);
7239 7240
}

P
Peter Zijlstra 已提交
7241
#ifdef CONFIG_FAIR_GROUP_SCHED
7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253

static inline bool cfs_rq_is_decayed(struct cfs_rq *cfs_rq)
{
	if (cfs_rq->load.weight)
		return false;

	if (cfs_rq->avg.load_sum)
		return false;

	if (cfs_rq->avg.util_sum)
		return false;

7254
	if (cfs_rq->avg.runnable_load_sum)
7255 7256 7257 7258 7259
		return false;

	return true;
}

7260
static void update_blocked_averages(int cpu)
7261 7262
{
	struct rq *rq = cpu_rq(cpu);
7263
	struct cfs_rq *cfs_rq, *pos;
7264
	struct rq_flags rf;
7265

7266
	rq_lock_irqsave(rq, &rf);
7267
	update_rq_clock(rq);
7268

7269 7270 7271 7272
	/*
	 * Iterates the task_group tree in a bottom up fashion, see
	 * list_add_leaf_cfs_rq() for details.
	 */
7273
	for_each_leaf_cfs_rq_safe(rq, cfs_rq, pos) {
7274 7275
		struct sched_entity *se;

7276 7277 7278
		/* throttled entities do not contribute to load */
		if (throttled_hierarchy(cfs_rq))
			continue;
7279

7280
		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq))
7281
			update_tg_load_avg(cfs_rq, 0);
7282

7283 7284 7285
		/* Propagate pending load changes to the parent, if any: */
		se = cfs_rq->tg->se[cpu];
		if (se && !skip_blocked_update(se))
7286
			update_load_avg(cfs_rq_of(se), se, 0);
7287 7288 7289 7290 7291 7292 7293

		/*
		 * There can be a lot of idle CPU cgroups.  Don't let fully
		 * decayed cfs_rqs linger on the list.
		 */
		if (cfs_rq_is_decayed(cfs_rq))
			list_del_leaf_cfs_rq(cfs_rq);
7294
	}
7295
	rq_unlock_irqrestore(rq, &rf);
7296 7297
}

7298
/*
7299
 * Compute the hierarchical load factor for cfs_rq and all its ascendants.
7300 7301 7302
 * This needs to be done in a top-down fashion because the load of a child
 * group is a fraction of its parents load.
 */
7303
static void update_cfs_rq_h_load(struct cfs_rq *cfs_rq)
7304
{
7305 7306
	struct rq *rq = rq_of(cfs_rq);
	struct sched_entity *se = cfs_rq->tg->se[cpu_of(rq)];
7307
	unsigned long now = jiffies;
7308
	unsigned long load;
7309

7310
	if (cfs_rq->last_h_load_update == now)
7311 7312
		return;

7313 7314 7315 7316 7317 7318 7319
	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;
	}
7320

7321
	if (!se) {
7322
		cfs_rq->h_load = cfs_rq_load_avg(cfs_rq);
7323 7324 7325 7326 7327
		cfs_rq->last_h_load_update = now;
	}

	while ((se = cfs_rq->h_load_next) != NULL) {
		load = cfs_rq->h_load;
7328 7329
		load = div64_ul(load * se->avg.load_avg,
			cfs_rq_load_avg(cfs_rq) + 1);
7330 7331 7332 7333
		cfs_rq = group_cfs_rq(se);
		cfs_rq->h_load = load;
		cfs_rq->last_h_load_update = now;
	}
7334 7335
}

7336
static unsigned long task_h_load(struct task_struct *p)
P
Peter Zijlstra 已提交
7337
{
7338
	struct cfs_rq *cfs_rq = task_cfs_rq(p);
P
Peter Zijlstra 已提交
7339

7340
	update_cfs_rq_h_load(cfs_rq);
7341
	return div64_ul(p->se.avg.load_avg * cfs_rq->h_load,
7342
			cfs_rq_load_avg(cfs_rq) + 1);
P
Peter Zijlstra 已提交
7343 7344
}
#else
7345
static inline void update_blocked_averages(int cpu)
7346
{
7347 7348
	struct rq *rq = cpu_rq(cpu);
	struct cfs_rq *cfs_rq = &rq->cfs;
7349
	struct rq_flags rf;
7350

7351
	rq_lock_irqsave(rq, &rf);
7352
	update_rq_clock(rq);
7353
	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
7354
	rq_unlock_irqrestore(rq, &rf);
7355 7356
}

7357
static unsigned long task_h_load(struct task_struct *p)
7358
{
7359
	return p->se.avg.load_avg;
7360
}
P
Peter Zijlstra 已提交
7361
#endif
7362 7363

/********** Helpers for find_busiest_group ************************/
7364 7365 7366 7367 7368 7369 7370

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

7371 7372 7373 7374 7375 7376 7377
/*
 * 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
7378
	unsigned long load_per_task;
7379
	unsigned long group_capacity;
7380
	unsigned long group_util; /* Total utilization of the group */
7381 7382 7383
	unsigned int sum_nr_running; /* Nr tasks running in the group */
	unsigned int idle_cpus;
	unsigned int group_weight;
7384
	enum group_type group_type;
7385
	int group_no_capacity;
7386 7387 7388 7389
#ifdef CONFIG_NUMA_BALANCING
	unsigned int nr_numa_running;
	unsigned int nr_preferred_running;
#endif
7390 7391
};

J
Joonsoo Kim 已提交
7392 7393 7394 7395 7396 7397 7398
/*
 * 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 */
7399
	unsigned long total_running;
J
Joonsoo Kim 已提交
7400
	unsigned long total_load;	/* Total load of all groups in sd */
7401
	unsigned long total_capacity;	/* Total capacity of all groups in sd */
J
Joonsoo Kim 已提交
7402 7403 7404
	unsigned long avg_load;	/* Average load across all groups in sd */

	struct sg_lb_stats busiest_stat;/* Statistics of the busiest group */
7405
	struct sg_lb_stats local_stat;	/* Statistics of the local group */
J
Joonsoo Kim 已提交
7406 7407
};

7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418
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,
7419
		.total_running = 0UL,
7420
		.total_load = 0UL,
7421
		.total_capacity = 0UL,
7422 7423
		.busiest_stat = {
			.avg_load = 0UL,
7424 7425
			.sum_nr_running = 0,
			.group_type = group_other,
7426 7427 7428 7429
		},
	};
}

7430 7431 7432
/**
 * get_sd_load_idx - Obtain the load index for a given sched domain.
 * @sd: The sched_domain whose load_idx is to be obtained.
7433
 * @idle: The idle status of the CPU for whose sd load_idx is obtained.
7434 7435
 *
 * Return: The load index.
7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457
 */
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;
}

7458
static unsigned long scale_rt_capacity(int cpu)
7459 7460
{
	struct rq *rq = cpu_rq(cpu);
7461
	u64 total, used, age_stamp, avg;
7462
	s64 delta;
7463

7464 7465 7466 7467
	/*
	 * Since we're reading these variables without serialization make sure
	 * we read them once before doing sanity checks on them.
	 */
7468 7469
	age_stamp = READ_ONCE(rq->age_stamp);
	avg = READ_ONCE(rq->rt_avg);
7470
	delta = __rq_clock_broken(rq) - age_stamp;
7471

7472 7473 7474 7475
	if (unlikely(delta < 0))
		delta = 0;

	total = sched_avg_period() + delta;
7476

7477
	used = div_u64(avg, total);
7478

7479 7480
	if (likely(used < SCHED_CAPACITY_SCALE))
		return SCHED_CAPACITY_SCALE - used;
7481

7482
	return 1;
7483 7484
}

7485
static void update_cpu_capacity(struct sched_domain *sd, int cpu)
7486
{
7487
	unsigned long capacity = arch_scale_cpu_capacity(sd, cpu);
7488 7489
	struct sched_group *sdg = sd->groups;

7490
	cpu_rq(cpu)->cpu_capacity_orig = capacity;
7491

7492
	capacity *= scale_rt_capacity(cpu);
7493
	capacity >>= SCHED_CAPACITY_SHIFT;
7494

7495 7496
	if (!capacity)
		capacity = 1;
7497

7498 7499
	cpu_rq(cpu)->cpu_capacity = capacity;
	sdg->sgc->capacity = capacity;
7500
	sdg->sgc->min_capacity = capacity;
7501 7502
}

7503
void update_group_capacity(struct sched_domain *sd, int cpu)
7504 7505 7506
{
	struct sched_domain *child = sd->child;
	struct sched_group *group, *sdg = sd->groups;
7507
	unsigned long capacity, min_capacity;
7508 7509 7510 7511
	unsigned long interval;

	interval = msecs_to_jiffies(sd->balance_interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);
7512
	sdg->sgc->next_update = jiffies + interval;
7513 7514

	if (!child) {
7515
		update_cpu_capacity(sd, cpu);
7516 7517 7518
		return;
	}

7519
	capacity = 0;
7520
	min_capacity = ULONG_MAX;
7521

P
Peter Zijlstra 已提交
7522 7523 7524 7525 7526 7527
	if (child->flags & SD_OVERLAP) {
		/*
		 * SD_OVERLAP domains cannot assume that child groups
		 * span the current group.
		 */

7528
		for_each_cpu(cpu, sched_group_span(sdg)) {
7529
			struct sched_group_capacity *sgc;
7530
			struct rq *rq = cpu_rq(cpu);
7531

7532
			/*
7533
			 * build_sched_domains() -> init_sched_groups_capacity()
7534 7535 7536
			 * gets here before we've attached the domains to the
			 * runqueues.
			 *
7537 7538
			 * Use capacity_of(), which is set irrespective of domains
			 * in update_cpu_capacity().
7539
			 *
7540
			 * This avoids capacity from being 0 and
7541 7542 7543
			 * causing divide-by-zero issues on boot.
			 */
			if (unlikely(!rq->sd)) {
7544
				capacity += capacity_of(cpu);
7545 7546 7547
			} else {
				sgc = rq->sd->groups->sgc;
				capacity += sgc->capacity;
7548
			}
7549

7550
			min_capacity = min(capacity, min_capacity);
7551
		}
P
Peter Zijlstra 已提交
7552 7553 7554 7555
	} else  {
		/*
		 * !SD_OVERLAP domains can assume that child groups
		 * span the current group.
7556
		 */
P
Peter Zijlstra 已提交
7557 7558 7559

		group = child->groups;
		do {
7560 7561 7562 7563
			struct sched_group_capacity *sgc = group->sgc;

			capacity += sgc->capacity;
			min_capacity = min(sgc->min_capacity, min_capacity);
P
Peter Zijlstra 已提交
7564 7565 7566
			group = group->next;
		} while (group != child->groups);
	}
7567

7568
	sdg->sgc->capacity = capacity;
7569
	sdg->sgc->min_capacity = min_capacity;
7570 7571
}

7572
/*
7573 7574 7575
 * Check whether the capacity of the rq has been noticeably reduced by side
 * activity. The imbalance_pct is used for the threshold.
 * Return true is the capacity is reduced
7576 7577
 */
static inline int
7578
check_cpu_capacity(struct rq *rq, struct sched_domain *sd)
7579
{
7580 7581
	return ((rq->cpu_capacity * sd->imbalance_pct) <
				(rq->cpu_capacity_orig * 100));
7582 7583
}

7584 7585
/*
 * Group imbalance indicates (and tries to solve) the problem where balancing
7586
 * groups is inadequate due to ->cpus_allowed constraints.
7587 7588 7589 7590 7591
 *
 * 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:
 *
7592 7593
 *	{ 0 1 2 3 } { 4 5 6 7 }
 *	        *     * * *
7594 7595 7596 7597 7598 7599
 *
 * 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
7600 7601
 * by noticing the lower domain failed to reach balance and had difficulty
 * moving tasks due to affinity constraints.
7602 7603
 *
 * When this is so detected; this group becomes a candidate for busiest; see
7604
 * update_sd_pick_busiest(). And calculate_imbalance() and
7605
 * find_busiest_group() avoid some of the usual balance conditions to allow it
7606 7607 7608 7609 7610 7611 7612
 * 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.
 */

7613
static inline int sg_imbalanced(struct sched_group *group)
7614
{
7615
	return group->sgc->imbalance;
7616 7617
}

7618
/*
7619 7620 7621
 * group_has_capacity returns true if the group has spare capacity that could
 * be used by some tasks.
 * We consider that a group has spare capacity if the  * number of task is
7622 7623
 * smaller than the number of CPUs or if the utilization is lower than the
 * available capacity for CFS tasks.
7624 7625 7626 7627 7628
 * For the latter, we use a threshold to stabilize the state, to take into
 * account the variance of the tasks' load and to return true if the available
 * capacity in meaningful for the load balancer.
 * As an example, an available capacity of 1% can appear but it doesn't make
 * any benefit for the load balance.
7629
 */
7630 7631
static inline bool
group_has_capacity(struct lb_env *env, struct sg_lb_stats *sgs)
7632
{
7633 7634
	if (sgs->sum_nr_running < sgs->group_weight)
		return true;
7635

7636
	if ((sgs->group_capacity * 100) >
7637
			(sgs->group_util * env->sd->imbalance_pct))
7638
		return true;
7639

7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655
	return false;
}

/*
 *  group_is_overloaded returns true if the group has more tasks than it can
 *  handle.
 *  group_is_overloaded is not equals to !group_has_capacity because a group
 *  with the exact right number of tasks, has no more spare capacity but is not
 *  overloaded so both group_has_capacity and group_is_overloaded return
 *  false.
 */
static inline bool
group_is_overloaded(struct lb_env *env, struct sg_lb_stats *sgs)
{
	if (sgs->sum_nr_running <= sgs->group_weight)
		return false;
7656

7657
	if ((sgs->group_capacity * 100) <
7658
			(sgs->group_util * env->sd->imbalance_pct))
7659
		return true;
7660

7661
	return false;
7662 7663
}

7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674
/*
 * group_smaller_cpu_capacity: Returns true if sched_group sg has smaller
 * per-CPU capacity than sched_group ref.
 */
static inline bool
group_smaller_cpu_capacity(struct sched_group *sg, struct sched_group *ref)
{
	return sg->sgc->min_capacity * capacity_margin <
						ref->sgc->min_capacity * 1024;
}

7675 7676 7677
static inline enum
group_type group_classify(struct sched_group *group,
			  struct sg_lb_stats *sgs)
7678
{
7679
	if (sgs->group_no_capacity)
7680 7681 7682 7683 7684 7685 7686 7687
		return group_overloaded;

	if (sg_imbalanced(group))
		return group_imbalanced;

	return group_other;
}

7688 7689
/**
 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
7690
 * @env: The load balancing environment.
7691 7692 7693 7694
 * @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.
7695
 * @overload: Indicate more than one runnable task for any CPU.
7696
 */
7697 7698
static inline void update_sg_lb_stats(struct lb_env *env,
			struct sched_group *group, int load_idx,
7699 7700
			int local_group, struct sg_lb_stats *sgs,
			bool *overload)
7701
{
7702
	unsigned long load;
7703
	int i, nr_running;
7704

7705 7706
	memset(sgs, 0, sizeof(*sgs));

7707
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
7708 7709 7710
		struct rq *rq = cpu_rq(i);

		/* Bias balancing toward cpus of our domain */
7711
		if (local_group)
7712
			load = target_load(i, load_idx);
7713
		else
7714 7715 7716
			load = source_load(i, load_idx);

		sgs->group_load += load;
7717
		sgs->group_util += cpu_util(i);
7718
		sgs->sum_nr_running += rq->cfs.h_nr_running;
7719

7720 7721
		nr_running = rq->nr_running;
		if (nr_running > 1)
7722 7723
			*overload = true;

7724 7725 7726 7727
#ifdef CONFIG_NUMA_BALANCING
		sgs->nr_numa_running += rq->nr_numa_running;
		sgs->nr_preferred_running += rq->nr_preferred_running;
#endif
7728
		sgs->sum_weighted_load += weighted_cpuload(rq);
7729 7730 7731 7732
		/*
		 * No need to call idle_cpu() if nr_running is not 0
		 */
		if (!nr_running && idle_cpu(i))
7733
			sgs->idle_cpus++;
7734 7735
	}

7736 7737
	/* Adjust by relative CPU capacity of the group */
	sgs->group_capacity = group->sgc->capacity;
7738
	sgs->avg_load = (sgs->group_load*SCHED_CAPACITY_SCALE) / sgs->group_capacity;
7739

7740
	if (sgs->sum_nr_running)
7741
		sgs->load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
7742

7743
	sgs->group_weight = group->group_weight;
7744

7745
	sgs->group_no_capacity = group_is_overloaded(env, sgs);
7746
	sgs->group_type = group_classify(group, sgs);
7747 7748
}

7749 7750
/**
 * update_sd_pick_busiest - return 1 on busiest group
7751
 * @env: The load balancing environment.
7752 7753
 * @sds: sched_domain statistics
 * @sg: sched_group candidate to be checked for being the busiest
7754
 * @sgs: sched_group statistics
7755 7756 7757
 *
 * Determine if @sg is a busier group than the previously selected
 * busiest group.
7758 7759 7760
 *
 * Return: %true if @sg is a busier group than the previously selected
 * busiest group. %false otherwise.
7761
 */
7762
static bool update_sd_pick_busiest(struct lb_env *env,
7763 7764
				   struct sd_lb_stats *sds,
				   struct sched_group *sg,
7765
				   struct sg_lb_stats *sgs)
7766
{
7767
	struct sg_lb_stats *busiest = &sds->busiest_stat;
7768

7769
	if (sgs->group_type > busiest->group_type)
7770 7771
		return true;

7772 7773 7774 7775 7776 7777
	if (sgs->group_type < busiest->group_type)
		return false;

	if (sgs->avg_load <= busiest->avg_load)
		return false;

7778 7779 7780 7781 7782 7783 7784 7785 7786 7787 7788 7789 7790 7791
	if (!(env->sd->flags & SD_ASYM_CPUCAPACITY))
		goto asym_packing;

	/*
	 * Candidate sg has no more than one task per CPU and
	 * has higher per-CPU capacity. Migrating tasks to less
	 * capable CPUs may harm throughput. Maximize throughput,
	 * power/energy consequences are not considered.
	 */
	if (sgs->sum_nr_running <= sgs->group_weight &&
	    group_smaller_cpu_capacity(sds->local, sg))
		return false;

asym_packing:
7792 7793
	/* This is the busiest node in its class. */
	if (!(env->sd->flags & SD_ASYM_PACKING))
7794 7795
		return true;

7796 7797 7798
	/* No ASYM_PACKING if target cpu is already busy */
	if (env->idle == CPU_NOT_IDLE)
		return true;
7799
	/*
T
Tim Chen 已提交
7800 7801 7802
	 * ASYM_PACKING needs to move all the work to the highest
	 * prority CPUs in the group, therefore mark all groups
	 * of lower priority than ourself as busy.
7803
	 */
T
Tim Chen 已提交
7804 7805
	if (sgs->sum_nr_running &&
	    sched_asym_prefer(env->dst_cpu, sg->asym_prefer_cpu)) {
7806 7807 7808
		if (!sds->busiest)
			return true;

T
Tim Chen 已提交
7809 7810 7811
		/* Prefer to move from lowest priority cpu's work */
		if (sched_asym_prefer(sds->busiest->asym_prefer_cpu,
				      sg->asym_prefer_cpu))
7812 7813 7814 7815 7816 7817
			return true;
	}

	return false;
}

7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839 7840 7841 7842 7843 7844 7845 7846 7847
#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 */

7848
/**
7849
 * update_sd_lb_stats - Update sched_domain's statistics for load balancing.
7850
 * @env: The load balancing environment.
7851 7852
 * @sds: variable to hold the statistics for this sched_domain.
 */
7853
static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sds)
7854
{
7855
	struct sched_domain_shared *shared = env->sd->shared;
7856 7857
	struct sched_domain *child = env->sd->child;
	struct sched_group *sg = env->sd->groups;
7858
	struct sg_lb_stats *local = &sds->local_stat;
J
Joonsoo Kim 已提交
7859
	struct sg_lb_stats tmp_sgs;
7860
	int load_idx, prefer_sibling = 0;
7861
	bool overload = false;
7862 7863 7864 7865

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

7866
	load_idx = get_sd_load_idx(env->sd, env->idle);
7867 7868

	do {
J
Joonsoo Kim 已提交
7869
		struct sg_lb_stats *sgs = &tmp_sgs;
7870 7871
		int local_group;

7872
		local_group = cpumask_test_cpu(env->dst_cpu, sched_group_span(sg));
J
Joonsoo Kim 已提交
7873 7874
		if (local_group) {
			sds->local = sg;
7875
			sgs = local;
7876 7877

			if (env->idle != CPU_NEWLY_IDLE ||
7878 7879
			    time_after_eq(jiffies, sg->sgc->next_update))
				update_group_capacity(env->sd, env->dst_cpu);
J
Joonsoo Kim 已提交
7880
		}
7881

7882 7883
		update_sg_lb_stats(env, sg, load_idx, local_group, sgs,
						&overload);
7884

7885 7886 7887
		if (local_group)
			goto next_group;

7888 7889
		/*
		 * In case the child domain prefers tasks go to siblings
7890
		 * first, lower the sg capacity so that we'll try
7891 7892
		 * and move all the excess tasks away. We lower the capacity
		 * of a group only if the local group has the capacity to fit
7893 7894 7895 7896
		 * these excess tasks. The extra check prevents the case where
		 * you always pull from the heaviest group when it is already
		 * under-utilized (possible with a large weight task outweighs
		 * the tasks on the system).
7897
		 */
7898
		if (prefer_sibling && sds->local &&
7899 7900
		    group_has_capacity(env, local) &&
		    (sgs->sum_nr_running > local->sum_nr_running + 1)) {
7901
			sgs->group_no_capacity = 1;
7902
			sgs->group_type = group_classify(sg, sgs);
7903
		}
7904

7905
		if (update_sd_pick_busiest(env, sds, sg, sgs)) {
7906
			sds->busiest = sg;
J
Joonsoo Kim 已提交
7907
			sds->busiest_stat = *sgs;
7908 7909
		}

7910 7911
next_group:
		/* Now, start updating sd_lb_stats */
7912
		sds->total_running += sgs->sum_nr_running;
7913
		sds->total_load += sgs->group_load;
7914
		sds->total_capacity += sgs->group_capacity;
7915

7916
		sg = sg->next;
7917
	} while (sg != env->sd->groups);
7918 7919 7920

	if (env->sd->flags & SD_NUMA)
		env->fbq_type = fbq_classify_group(&sds->busiest_stat);
7921 7922 7923 7924 7925 7926 7927

	if (!env->sd->parent) {
		/* update overload indicator if we are at root domain */
		if (env->dst_rq->rd->overload != overload)
			env->dst_rq->rd->overload = overload;
	}

7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941 7942
	if (!shared)
		return;

	/*
	 * Since these are sums over groups they can contain some CPUs
	 * multiple times for the NUMA domains.
	 *
	 * Currently only wake_affine_llc() and find_busiest_group()
	 * uses these numbers, only the last is affected by this problem.
	 *
	 * XXX fix that.
	 */
	WRITE_ONCE(shared->nr_running,	sds->total_running);
	WRITE_ONCE(shared->load,	sds->total_load);
	WRITE_ONCE(shared->capacity,	sds->total_capacity);
7943 7944 7945 7946
}

/**
 * check_asym_packing - Check to see if the group is packed into the
7947
 *			sched domain.
7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961
 *
 * 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.
 *
7962
 * Return: 1 when packing is required and a task should be moved to
7963
 * this CPU.  The amount of the imbalance is returned in env->imbalance.
7964
 *
7965
 * @env: The load balancing environment.
7966 7967
 * @sds: Statistics of the sched_domain which is to be packed
 */
7968
static int check_asym_packing(struct lb_env *env, struct sd_lb_stats *sds)
7969 7970 7971
{
	int busiest_cpu;

7972
	if (!(env->sd->flags & SD_ASYM_PACKING))
7973 7974
		return 0;

7975 7976 7977
	if (env->idle == CPU_NOT_IDLE)
		return 0;

7978 7979 7980
	if (!sds->busiest)
		return 0;

T
Tim Chen 已提交
7981 7982
	busiest_cpu = sds->busiest->asym_prefer_cpu;
	if (sched_asym_prefer(busiest_cpu, env->dst_cpu))
7983 7984
		return 0;

7985
	env->imbalance = DIV_ROUND_CLOSEST(
7986
		sds->busiest_stat.avg_load * sds->busiest_stat.group_capacity,
7987
		SCHED_CAPACITY_SCALE);
7988

7989
	return 1;
7990 7991 7992 7993 7994 7995
}

/**
 * fix_small_imbalance - Calculate the minor imbalance that exists
 *			amongst the groups of a sched_domain, during
 *			load balancing.
7996
 * @env: The load balancing environment.
7997 7998
 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
 */
7999 8000
static inline
void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8001
{
8002
	unsigned long tmp, capa_now = 0, capa_move = 0;
8003
	unsigned int imbn = 2;
8004
	unsigned long scaled_busy_load_per_task;
J
Joonsoo Kim 已提交
8005
	struct sg_lb_stats *local, *busiest;
8006

J
Joonsoo Kim 已提交
8007 8008
	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8009

J
Joonsoo Kim 已提交
8010 8011 8012 8013
	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;
8014

J
Joonsoo Kim 已提交
8015
	scaled_busy_load_per_task =
8016
		(busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8017
		busiest->group_capacity;
J
Joonsoo Kim 已提交
8018

8019 8020
	if (busiest->avg_load + scaled_busy_load_per_task >=
	    local->avg_load + (scaled_busy_load_per_task * imbn)) {
J
Joonsoo Kim 已提交
8021
		env->imbalance = busiest->load_per_task;
8022 8023 8024 8025 8026
		return;
	}

	/*
	 * OK, we don't have enough imbalance to justify moving tasks,
8027
	 * however we may be able to increase total CPU capacity used by
8028 8029 8030
	 * moving them.
	 */

8031
	capa_now += busiest->group_capacity *
J
Joonsoo Kim 已提交
8032
			min(busiest->load_per_task, busiest->avg_load);
8033
	capa_now += local->group_capacity *
J
Joonsoo Kim 已提交
8034
			min(local->load_per_task, local->avg_load);
8035
	capa_now /= SCHED_CAPACITY_SCALE;
8036 8037

	/* Amount of load we'd subtract */
8038
	if (busiest->avg_load > scaled_busy_load_per_task) {
8039
		capa_move += busiest->group_capacity *
J
Joonsoo Kim 已提交
8040
			    min(busiest->load_per_task,
8041
				busiest->avg_load - scaled_busy_load_per_task);
J
Joonsoo Kim 已提交
8042
	}
8043 8044

	/* Amount of load we'd add */
8045
	if (busiest->avg_load * busiest->group_capacity <
8046
	    busiest->load_per_task * SCHED_CAPACITY_SCALE) {
8047 8048
		tmp = (busiest->avg_load * busiest->group_capacity) /
		      local->group_capacity;
J
Joonsoo Kim 已提交
8049
	} else {
8050
		tmp = (busiest->load_per_task * SCHED_CAPACITY_SCALE) /
8051
		      local->group_capacity;
J
Joonsoo Kim 已提交
8052
	}
8053
	capa_move += local->group_capacity *
8054
		    min(local->load_per_task, local->avg_load + tmp);
8055
	capa_move /= SCHED_CAPACITY_SCALE;
8056 8057

	/* Move if we gain throughput */
8058
	if (capa_move > capa_now)
J
Joonsoo Kim 已提交
8059
		env->imbalance = busiest->load_per_task;
8060 8061 8062 8063 8064
}

/**
 * calculate_imbalance - Calculate the amount of imbalance present within the
 *			 groups of a given sched_domain during load balance.
8065
 * @env: load balance environment
8066 8067
 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
 */
8068
static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
8069
{
8070
	unsigned long max_pull, load_above_capacity = ~0UL;
J
Joonsoo Kim 已提交
8071 8072 8073 8074
	struct sg_lb_stats *local, *busiest;

	local = &sds->local_stat;
	busiest = &sds->busiest_stat;
8075

8076
	if (busiest->group_type == group_imbalanced) {
8077 8078 8079 8080
		/*
		 * 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
8081 8082
		busiest->load_per_task =
			min(busiest->load_per_task, sds->avg_load);
8083 8084
	}

8085
	/*
8086 8087 8088 8089
	 * Avg load of busiest sg can be less and avg load of local sg can
	 * be greater than avg load across all sgs of sd because avg load
	 * factors in sg capacity and sgs with smaller group_type are
	 * skipped when updating the busiest sg:
8090
	 */
8091 8092
	if (busiest->avg_load <= sds->avg_load ||
	    local->avg_load >= sds->avg_load) {
8093 8094
		env->imbalance = 0;
		return fix_small_imbalance(env, sds);
8095 8096
	}

8097 8098 8099 8100 8101
	/*
	 * If there aren't any idle cpus, avoid creating some.
	 */
	if (busiest->group_type == group_overloaded &&
	    local->group_type   == group_overloaded) {
8102
		load_above_capacity = busiest->sum_nr_running * SCHED_CAPACITY_SCALE;
8103
		if (load_above_capacity > busiest->group_capacity) {
8104
			load_above_capacity -= busiest->group_capacity;
8105
			load_above_capacity *= scale_load_down(NICE_0_LOAD);
8106 8107
			load_above_capacity /= busiest->group_capacity;
		} else
8108
			load_above_capacity = ~0UL;
8109 8110 8111 8112 8113 8114
	}

	/*
	 * 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,
8115 8116
	 * we also don't want to reduce the group load below the group
	 * capacity. Thus we look for the minimum possible imbalance.
8117
	 */
8118
	max_pull = min(busiest->avg_load - sds->avg_load, load_above_capacity);
8119 8120

	/* How much load to actually move to equalise the imbalance */
J
Joonsoo Kim 已提交
8121
	env->imbalance = min(
8122 8123
		max_pull * busiest->group_capacity,
		(sds->avg_load - local->avg_load) * local->group_capacity
8124
	) / SCHED_CAPACITY_SCALE;
8125 8126 8127

	/*
	 * if *imbalance is less than the average load per runnable task
L
Lucas De Marchi 已提交
8128
	 * there is no guarantee that any tasks will be moved so we'll have
8129 8130 8131
	 * a think about bumping its value to force at least one task to be
	 * moved
	 */
J
Joonsoo Kim 已提交
8132
	if (env->imbalance < busiest->load_per_task)
8133
		return fix_small_imbalance(env, sds);
8134
}
8135

8136 8137 8138 8139
/******* find_busiest_group() helpers end here *********************/

/**
 * find_busiest_group - Returns the busiest group within the sched_domain
8140
 * if there is an imbalance.
8141 8142 8143 8144
 *
 * Also calculates the amount of weighted load which should be moved
 * to restore balance.
 *
8145
 * @env: The load balancing environment.
8146
 *
8147
 * Return:	- The busiest group if imbalance exists.
8148
 */
J
Joonsoo Kim 已提交
8149
static struct sched_group *find_busiest_group(struct lb_env *env)
8150
{
J
Joonsoo Kim 已提交
8151
	struct sg_lb_stats *local, *busiest;
8152 8153
	struct sd_lb_stats sds;

8154
	init_sd_lb_stats(&sds);
8155 8156 8157 8158 8159

	/*
	 * Compute the various statistics relavent for load balancing at
	 * this level.
	 */
8160
	update_sd_lb_stats(env, &sds);
J
Joonsoo Kim 已提交
8161 8162
	local = &sds.local_stat;
	busiest = &sds.busiest_stat;
8163

8164
	/* ASYM feature bypasses nice load balance check */
8165
	if (check_asym_packing(env, &sds))
8166 8167
		return sds.busiest;

8168
	/* There is no busy sibling group to pull tasks from */
J
Joonsoo Kim 已提交
8169
	if (!sds.busiest || busiest->sum_nr_running == 0)
8170 8171
		goto out_balanced;

8172
	/* XXX broken for overlapping NUMA groups */
8173 8174
	sds.avg_load = (SCHED_CAPACITY_SCALE * sds.total_load)
						/ sds.total_capacity;
8175

P
Peter Zijlstra 已提交
8176 8177
	/*
	 * If the busiest group is imbalanced the below checks don't
8178
	 * work because they assume all things are equal, which typically
P
Peter Zijlstra 已提交
8179 8180
	 * isn't true due to cpus_allowed constraints and the like.
	 */
8181
	if (busiest->group_type == group_imbalanced)
P
Peter Zijlstra 已提交
8182 8183
		goto force_balance;

8184
	/* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
8185 8186
	if (env->idle == CPU_NEWLY_IDLE && group_has_capacity(env, local) &&
	    busiest->group_no_capacity)
8187 8188
		goto force_balance;

8189
	/*
8190
	 * If the local group is busier than the selected busiest group
8191 8192
	 * don't try and pull any tasks.
	 */
J
Joonsoo Kim 已提交
8193
	if (local->avg_load >= busiest->avg_load)
8194 8195
		goto out_balanced;

8196 8197 8198 8199
	/*
	 * Don't pull any tasks if this group is already above the domain
	 * average load.
	 */
J
Joonsoo Kim 已提交
8200
	if (local->avg_load >= sds.avg_load)
8201 8202
		goto out_balanced;

8203
	if (env->idle == CPU_IDLE) {
8204
		/*
8205 8206 8207 8208 8209
		 * This cpu is idle. If the busiest group is not overloaded
		 * and there is no imbalance between this and busiest group
		 * wrt idle cpus, it is balanced. The imbalance becomes
		 * significant if the diff is greater than 1 otherwise we
		 * might end up to just move the imbalance on another group
8210
		 */
8211 8212
		if ((busiest->group_type != group_overloaded) &&
				(local->idle_cpus <= (busiest->idle_cpus + 1)))
8213
			goto out_balanced;
8214 8215 8216 8217 8218
	} else {
		/*
		 * In the CPU_NEWLY_IDLE, CPU_NOT_IDLE cases, use
		 * imbalance_pct to be conservative.
		 */
J
Joonsoo Kim 已提交
8219 8220
		if (100 * busiest->avg_load <=
				env->sd->imbalance_pct * local->avg_load)
8221
			goto out_balanced;
8222
	}
8223

8224
force_balance:
8225
	/* Looks like there is an imbalance. Compute it */
8226
	calculate_imbalance(env, &sds);
8227 8228 8229
	return sds.busiest;

out_balanced:
8230
	env->imbalance = 0;
8231 8232 8233 8234 8235 8236
	return NULL;
}

/*
 * find_busiest_queue - find the busiest runqueue among the cpus in group.
 */
8237
static struct rq *find_busiest_queue(struct lb_env *env,
8238
				     struct sched_group *group)
8239 8240
{
	struct rq *busiest = NULL, *rq;
8241
	unsigned long busiest_load = 0, busiest_capacity = 1;
8242 8243
	int i;

8244
	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
8245
		unsigned long capacity, wl;
8246 8247 8248 8249
		enum fbq_type rt;

		rq = cpu_rq(i);
		rt = fbq_classify_rq(rq);
8250

8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272
		/*
		 * 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;

8273
		capacity = capacity_of(i);
8274

8275
		wl = weighted_cpuload(rq);
8276

8277 8278
		/*
		 * When comparing with imbalance, use weighted_cpuload()
8279
		 * which is not scaled with the cpu capacity.
8280
		 */
8281 8282 8283

		if (rq->nr_running == 1 && wl > env->imbalance &&
		    !check_cpu_capacity(rq, env->sd))
8284 8285
			continue;

8286 8287
		/*
		 * For the load comparisons with the other cpu's, consider
8288 8289 8290
		 * the weighted_cpuload() scaled with the cpu capacity, so
		 * that the load can be moved away from the cpu that is
		 * potentially running at a lower capacity.
8291
		 *
8292
		 * Thus we're looking for max(wl_i / capacity_i), crosswise
8293
		 * multiplication to rid ourselves of the division works out
8294 8295
		 * to: wl_i * capacity_j > wl_j * capacity_i;  where j is
		 * our previous maximum.
8296
		 */
8297
		if (wl * busiest_capacity > busiest_load * capacity) {
8298
			busiest_load = wl;
8299
			busiest_capacity = capacity;
8300 8301 8302 8303 8304 8305 8306 8307 8308 8309 8310 8311 8312
			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

8313
static int need_active_balance(struct lb_env *env)
8314
{
8315 8316 8317
	struct sched_domain *sd = env->sd;

	if (env->idle == CPU_NEWLY_IDLE) {
8318 8319 8320

		/*
		 * ASYM_PACKING needs to force migrate tasks from busy but
T
Tim Chen 已提交
8321 8322
		 * lower priority CPUs in order to pack all tasks in the
		 * highest priority CPUs.
8323
		 */
T
Tim Chen 已提交
8324 8325
		if ((sd->flags & SD_ASYM_PACKING) &&
		    sched_asym_prefer(env->dst_cpu, env->src_cpu))
8326
			return 1;
8327 8328
	}

8329 8330 8331 8332 8333 8334 8335 8336 8337 8338 8339 8340 8341
	/*
	 * The dst_cpu is idle and the src_cpu CPU has only 1 CFS task.
	 * It's worth migrating the task if the src_cpu's capacity is reduced
	 * because of other sched_class or IRQs if more capacity stays
	 * available on dst_cpu.
	 */
	if ((env->idle != CPU_NOT_IDLE) &&
	    (env->src_rq->cfs.h_nr_running == 1)) {
		if ((check_cpu_capacity(env->src_rq, sd)) &&
		    (capacity_of(env->src_cpu)*sd->imbalance_pct < capacity_of(env->dst_cpu)*100))
			return 1;
	}

8342 8343 8344
	return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
}

8345 8346
static int active_load_balance_cpu_stop(void *data);

8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359
static int should_we_balance(struct lb_env *env)
{
	struct sched_group *sg = env->sd->groups;
	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;

	/* Try to find first idle cpu */
8360
	for_each_cpu_and(cpu, group_balance_mask(sg), env->cpus) {
8361
		if (!idle_cpu(cpu))
8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374
			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.
	 */
8375
	return balance_cpu == env->dst_cpu;
8376 8377
}

8378 8379 8380 8381 8382 8383
/*
 * 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,
8384
			int *continue_balancing)
8385
{
8386
	int ld_moved, cur_ld_moved, active_balance = 0;
8387
	struct sched_domain *sd_parent = sd->parent;
8388 8389
	struct sched_group *group;
	struct rq *busiest;
8390
	struct rq_flags rf;
8391
	struct cpumask *cpus = this_cpu_cpumask_var_ptr(load_balance_mask);
8392

8393 8394
	struct lb_env env = {
		.sd		= sd,
8395 8396
		.dst_cpu	= this_cpu,
		.dst_rq		= this_rq,
8397
		.dst_grpmask    = sched_group_span(sd->groups),
8398
		.idle		= idle,
8399
		.loop_break	= sched_nr_migrate_break,
8400
		.cpus		= cpus,
8401
		.fbq_type	= all,
8402
		.tasks		= LIST_HEAD_INIT(env.tasks),
8403 8404
	};

8405
	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
8406

8407
	schedstat_inc(sd->lb_count[idle]);
8408 8409

redo:
8410 8411
	if (!should_we_balance(&env)) {
		*continue_balancing = 0;
8412
		goto out_balanced;
8413
	}
8414

8415
	group = find_busiest_group(&env);
8416
	if (!group) {
8417
		schedstat_inc(sd->lb_nobusyg[idle]);
8418 8419 8420
		goto out_balanced;
	}

8421
	busiest = find_busiest_queue(&env, group);
8422
	if (!busiest) {
8423
		schedstat_inc(sd->lb_nobusyq[idle]);
8424 8425 8426
		goto out_balanced;
	}

8427
	BUG_ON(busiest == env.dst_rq);
8428

8429
	schedstat_add(sd->lb_imbalance[idle], env.imbalance);
8430

8431 8432 8433
	env.src_cpu = busiest->cpu;
	env.src_rq = busiest;

8434 8435 8436 8437 8438 8439 8440 8441
	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.
		 */
8442
		env.flags |= LBF_ALL_PINNED;
8443
		env.loop_max  = min(sysctl_sched_nr_migrate, busiest->nr_running);
8444

8445
more_balance:
8446
		rq_lock_irqsave(busiest, &rf);
8447
		update_rq_clock(busiest);
8448 8449 8450 8451 8452

		/*
		 * cur_ld_moved - load moved in current iteration
		 * ld_moved     - cumulative load moved across iterations
		 */
8453
		cur_ld_moved = detach_tasks(&env);
8454 8455

		/*
8456 8457 8458 8459 8460
		 * We've detached some tasks from busiest_rq. Every
		 * task is masked "TASK_ON_RQ_MIGRATING", so we can safely
		 * unlock busiest->lock, and we are able to be sure
		 * that nobody can manipulate the tasks in parallel.
		 * See task_rq_lock() family for the details.
8461
		 */
8462

8463
		rq_unlock(busiest, &rf);
8464 8465 8466 8467 8468 8469

		if (cur_ld_moved) {
			attach_tasks(&env);
			ld_moved += cur_ld_moved;
		}

8470
		local_irq_restore(rf.flags);
8471

8472 8473 8474 8475 8476
		if (env.flags & LBF_NEED_BREAK) {
			env.flags &= ~LBF_NEED_BREAK;
			goto more_balance;
		}

8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495
		/*
		 * 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.
		 */
8496
		if ((env.flags & LBF_DST_PINNED) && env.imbalance > 0) {
8497

8498 8499 8500
			/* Prevent to re-select dst_cpu via env's cpus */
			cpumask_clear_cpu(env.dst_cpu, env.cpus);

8501
			env.dst_rq	 = cpu_rq(env.new_dst_cpu);
8502
			env.dst_cpu	 = env.new_dst_cpu;
8503
			env.flags	&= ~LBF_DST_PINNED;
8504 8505
			env.loop	 = 0;
			env.loop_break	 = sched_nr_migrate_break;
8506

8507 8508 8509 8510 8511 8512
			/*
			 * Go back to "more_balance" rather than "redo" since we
			 * need to continue with same src_cpu.
			 */
			goto more_balance;
		}
8513

8514 8515 8516 8517
		/*
		 * We failed to reach balance because of affinity.
		 */
		if (sd_parent) {
8518
			int *group_imbalance = &sd_parent->groups->sgc->imbalance;
8519

8520
			if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0)
8521 8522 8523
				*group_imbalance = 1;
		}

8524
		/* All tasks on this runqueue were pinned by CPU affinity */
8525
		if (unlikely(env.flags & LBF_ALL_PINNED)) {
8526
			cpumask_clear_cpu(cpu_of(busiest), cpus);
8527 8528 8529 8530 8531 8532 8533 8534 8535
			/*
			 * Attempting to continue load balancing at the current
			 * sched_domain level only makes sense if there are
			 * active CPUs remaining as possible busiest CPUs to
			 * pull load from which are not contained within the
			 * destination group that is receiving any migrated
			 * load.
			 */
			if (!cpumask_subset(cpus, env.dst_grpmask)) {
8536 8537
				env.loop = 0;
				env.loop_break = sched_nr_migrate_break;
8538
				goto redo;
8539
			}
8540
			goto out_all_pinned;
8541 8542 8543 8544
		}
	}

	if (!ld_moved) {
8545
		schedstat_inc(sd->lb_failed[idle]);
8546 8547 8548 8549 8550 8551 8552 8553
		/*
		 * 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++;
8554

8555
		if (need_active_balance(&env)) {
8556 8557
			unsigned long flags;

8558 8559
			raw_spin_lock_irqsave(&busiest->lock, flags);

8560 8561 8562
			/* don't kick the active_load_balance_cpu_stop,
			 * if the curr task on busiest cpu can't be
			 * moved to this_cpu
8563
			 */
8564
			if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) {
8565 8566
				raw_spin_unlock_irqrestore(&busiest->lock,
							    flags);
8567
				env.flags |= LBF_ALL_PINNED;
8568 8569 8570
				goto out_one_pinned;
			}

8571 8572 8573 8574 8575
			/*
			 * ->active_balance synchronizes accesses to
			 * ->active_balance_work.  Once set, it's cleared
			 * only after active load balance is finished.
			 */
8576 8577 8578 8579 8580 8581
			if (!busiest->active_balance) {
				busiest->active_balance = 1;
				busiest->push_cpu = this_cpu;
				active_balance = 1;
			}
			raw_spin_unlock_irqrestore(&busiest->lock, flags);
8582

8583
			if (active_balance) {
8584 8585 8586
				stop_one_cpu_nowait(cpu_of(busiest),
					active_load_balance_cpu_stop, busiest,
					&busiest->active_balance_work);
8587
			}
8588

8589
			/* We've kicked active balancing, force task migration. */
8590 8591 8592 8593 8594 8595 8596 8597 8598 8599 8600 8601 8602
			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
8603
		 * detach_tasks).
8604 8605 8606 8607 8608 8609 8610 8611
		 */
		if (sd->balance_interval < sd->max_interval)
			sd->balance_interval *= 2;
	}

	goto out;

out_balanced:
8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628
	/*
	 * We reach balance although we may have faced some affinity
	 * constraints. Clear the imbalance flag if it was set.
	 */
	if (sd_parent) {
		int *group_imbalance = &sd_parent->groups->sgc->imbalance;

		if (*group_imbalance)
			*group_imbalance = 0;
	}

out_all_pinned:
	/*
	 * We reach balance because all tasks are pinned at this level so
	 * we can't migrate them. Let the imbalance flag set so parent level
	 * can try to migrate them.
	 */
8629
	schedstat_inc(sd->lb_balanced[idle]);
8630 8631 8632 8633 8634

	sd->nr_balance_failed = 0;

out_one_pinned:
	/* tune up the balancing interval */
8635
	if (((env.flags & LBF_ALL_PINNED) &&
8636
			sd->balance_interval < MAX_PINNED_INTERVAL) ||
8637 8638 8639
			(sd->balance_interval < sd->max_interval))
		sd->balance_interval *= 2;

8640
	ld_moved = 0;
8641 8642 8643 8644
out:
	return ld_moved;
}

8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660
static inline unsigned long
get_sd_balance_interval(struct sched_domain *sd, int cpu_busy)
{
	unsigned long interval = sd->balance_interval;

	if (cpu_busy)
		interval *= sd->busy_factor;

	/* scale ms to jiffies */
	interval = msecs_to_jiffies(interval);
	interval = clamp(interval, 1UL, max_load_balance_interval);

	return interval;
}

static inline void
8661
update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
8662 8663 8664
{
	unsigned long interval, next;

8665 8666
	/* used by idle balance, so cpu_busy = 0 */
	interval = get_sd_balance_interval(sd, 0);
8667 8668 8669 8670 8671 8672
	next = sd->last_balance + interval;

	if (time_after(*next_balance, next))
		*next_balance = next;
}

8673 8674 8675 8676
/*
 * idle_balance is called by schedule() if this_cpu is about to become
 * idle. Attempts to pull tasks from other CPUs.
 */
8677
static int idle_balance(struct rq *this_rq, struct rq_flags *rf)
8678
{
8679 8680
	unsigned long next_balance = jiffies + HZ;
	int this_cpu = this_rq->cpu;
8681 8682
	struct sched_domain *sd;
	int pulled_task = 0;
8683
	u64 curr_cost = 0;
8684

8685 8686 8687 8688 8689 8690
	/*
	 * We must set idle_stamp _before_ calling idle_balance(), such that we
	 * measure the duration of idle_balance() as idle time.
	 */
	this_rq->idle_stamp = rq_clock(this_rq);

8691 8692 8693 8694 8695 8696
	/*
	 * Do not pull tasks towards !active CPUs...
	 */
	if (!cpu_active(this_cpu))
		return 0;

8697 8698 8699 8700 8701 8702 8703 8704
	/*
	 * This is OK, because current is on_cpu, which avoids it being picked
	 * for load-balance and preemption/IRQs are still disabled avoiding
	 * further scheduler activity on it and we're being very careful to
	 * re-start the picking loop.
	 */
	rq_unpin_lock(this_rq, rf);

8705 8706
	if (this_rq->avg_idle < sysctl_sched_migration_cost ||
	    !this_rq->rd->overload) {
8707 8708 8709
		rcu_read_lock();
		sd = rcu_dereference_check_sched_domain(this_rq->sd);
		if (sd)
8710
			update_next_balance(sd, &next_balance);
8711 8712
		rcu_read_unlock();

8713
		goto out;
8714
	}
8715

8716 8717
	raw_spin_unlock(&this_rq->lock);

8718
	update_blocked_averages(this_cpu);
8719
	rcu_read_lock();
8720
	for_each_domain(this_cpu, sd) {
8721
		int continue_balancing = 1;
8722
		u64 t0, domain_cost;
8723 8724 8725 8726

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

8727
		if (this_rq->avg_idle < curr_cost + sd->max_newidle_lb_cost) {
8728
			update_next_balance(sd, &next_balance);
8729
			break;
8730
		}
8731

8732
		if (sd->flags & SD_BALANCE_NEWIDLE) {
8733 8734
			t0 = sched_clock_cpu(this_cpu);

8735
			pulled_task = load_balance(this_cpu, this_rq,
8736 8737
						   sd, CPU_NEWLY_IDLE,
						   &continue_balancing);
8738 8739 8740 8741 8742 8743

			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;
8744
		}
8745

8746
		update_next_balance(sd, &next_balance);
8747 8748 8749 8750 8751 8752

		/*
		 * Stop searching for tasks to pull if there are
		 * now runnable tasks on this rq.
		 */
		if (pulled_task || this_rq->nr_running > 0)
8753 8754
			break;
	}
8755
	rcu_read_unlock();
8756 8757 8758

	raw_spin_lock(&this_rq->lock);

8759 8760 8761
	if (curr_cost > this_rq->max_idle_balance_cost)
		this_rq->max_idle_balance_cost = curr_cost;

8762
	/*
8763 8764 8765
	 * While browsing the domains, we released the rq lock, a task could
	 * have been enqueued in the meantime. Since we're not going idle,
	 * pretend we pulled a task.
8766
	 */
8767
	if (this_rq->cfs.h_nr_running && !pulled_task)
8768
		pulled_task = 1;
8769

8770 8771 8772
out:
	/* Move the next balance forward */
	if (time_after(this_rq->next_balance, next_balance))
8773
		this_rq->next_balance = next_balance;
8774

8775
	/* Is there a task of a high priority class? */
8776
	if (this_rq->nr_running != this_rq->cfs.h_nr_running)
8777 8778
		pulled_task = -1;

8779
	if (pulled_task)
8780 8781
		this_rq->idle_stamp = 0;

8782 8783
	rq_repin_lock(this_rq, rf);

8784
	return pulled_task;
8785 8786 8787
}

/*
8788 8789 8790 8791
 * 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.
8792
 */
8793
static int active_load_balance_cpu_stop(void *data)
8794
{
8795 8796
	struct rq *busiest_rq = data;
	int busiest_cpu = cpu_of(busiest_rq);
8797
	int target_cpu = busiest_rq->push_cpu;
8798
	struct rq *target_rq = cpu_rq(target_cpu);
8799
	struct sched_domain *sd;
8800
	struct task_struct *p = NULL;
8801
	struct rq_flags rf;
8802

8803
	rq_lock_irq(busiest_rq, &rf);
8804 8805 8806 8807 8808 8809 8810
	/*
	 * Between queueing the stop-work and running it is a hole in which
	 * CPUs can become inactive. We should not move tasks from or to
	 * inactive CPUs.
	 */
	if (!cpu_active(busiest_cpu) || !cpu_active(target_cpu))
		goto out_unlock;
8811 8812 8813 8814 8815

	/* 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;
8816 8817 8818

	/* Is there any task to move? */
	if (busiest_rq->nr_running <= 1)
8819
		goto out_unlock;
8820 8821 8822 8823 8824 8825 8826 8827 8828

	/*
	 * This condition is "impossible", if it occurs
	 * we need to fix it. Originally reported by
	 * Bjorn Helgaas on a 128-cpu setup.
	 */
	BUG_ON(busiest_rq == target_rq);

	/* Search for an sd spanning us and the target CPU. */
8829
	rcu_read_lock();
8830 8831 8832 8833 8834 8835 8836
	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)) {
8837 8838
		struct lb_env env = {
			.sd		= sd,
8839 8840 8841 8842
			.dst_cpu	= target_cpu,
			.dst_rq		= target_rq,
			.src_cpu	= busiest_rq->cpu,
			.src_rq		= busiest_rq,
8843
			.idle		= CPU_IDLE,
8844 8845 8846 8847 8848 8849 8850
			/*
			 * can_migrate_task() doesn't need to compute new_dst_cpu
			 * for active balancing. Since we have CPU_IDLE, but no
			 * @dst_grpmask we need to make that test go away with lying
			 * about DST_PINNED.
			 */
			.flags		= LBF_DST_PINNED,
8851 8852
		};

8853
		schedstat_inc(sd->alb_count);
8854
		update_rq_clock(busiest_rq);
8855

8856
		p = detach_one_task(&env);
8857
		if (p) {
8858
			schedstat_inc(sd->alb_pushed);
8859 8860 8861
			/* Active balancing done, reset the failure counter. */
			sd->nr_balance_failed = 0;
		} else {
8862
			schedstat_inc(sd->alb_failed);
8863
		}
8864
	}
8865
	rcu_read_unlock();
8866 8867
out_unlock:
	busiest_rq->active_balance = 0;
8868
	rq_unlock(busiest_rq, &rf);
8869 8870 8871 8872 8873 8874

	if (p)
		attach_one_task(target_rq, p);

	local_irq_enable();

8875
	return 0;
8876 8877
}

8878 8879 8880 8881 8882
static inline int on_null_domain(struct rq *rq)
{
	return unlikely(!rcu_dereference_sched(rq->sd));
}

8883
#ifdef CONFIG_NO_HZ_COMMON
8884 8885 8886 8887 8888 8889
/*
 * 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.
 */
8890
static struct {
8891
	cpumask_var_t idle_cpus_mask;
8892
	atomic_t nr_cpus;
8893 8894
	unsigned long next_balance;     /* in jiffy units */
} nohz ____cacheline_aligned;
8895

8896
static inline int find_new_ilb(void)
8897
{
8898
	int ilb = cpumask_first(nohz.idle_cpus_mask);
8899

8900 8901 8902 8903
	if (ilb < nr_cpu_ids && idle_cpu(ilb))
		return ilb;

	return nr_cpu_ids;
8904 8905
}

8906 8907 8908 8909 8910
/*
 * 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).
 */
8911
static void nohz_balancer_kick(void)
8912 8913 8914 8915 8916
{
	int ilb_cpu;

	nohz.next_balance++;

8917
	ilb_cpu = find_new_ilb();
8918

8919 8920
	if (ilb_cpu >= nr_cpu_ids)
		return;
8921

8922
	if (test_and_set_bit(NOHZ_BALANCE_KICK, nohz_flags(ilb_cpu)))
8923 8924 8925 8926 8927 8928 8929 8930
		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);
8931 8932 8933
	return;
}

8934
void nohz_balance_exit_idle(unsigned int cpu)
8935 8936
{
	if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
8937 8938 8939 8940 8941 8942 8943
		/*
		 * Completely isolated CPUs don't ever set, so we must test.
		 */
		if (likely(cpumask_test_cpu(cpu, nohz.idle_cpus_mask))) {
			cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
			atomic_dec(&nohz.nr_cpus);
		}
8944 8945 8946 8947
		clear_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
	}
}

8948 8949 8950
static inline void set_cpu_sd_state_busy(void)
{
	struct sched_domain *sd;
8951
	int cpu = smp_processor_id();
8952 8953

	rcu_read_lock();
8954
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8955 8956 8957 8958 8959

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

8960
	atomic_inc(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8961
unlock:
8962 8963 8964 8965 8966 8967
	rcu_read_unlock();
}

void set_cpu_sd_state_idle(void)
{
	struct sched_domain *sd;
8968
	int cpu = smp_processor_id();
8969 8970

	rcu_read_lock();
8971
	sd = rcu_dereference(per_cpu(sd_llc, cpu));
V
Vincent Guittot 已提交
8972 8973 8974 8975 8976

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

8977
	atomic_dec(&sd->shared->nr_busy_cpus);
V
Vincent Guittot 已提交
8978
unlock:
8979 8980 8981
	rcu_read_unlock();
}

8982
/*
8983
 * This routine will record that the cpu is going idle with tick stopped.
8984
 * This info will be used in performing idle load balancing in the future.
8985
 */
8986
void nohz_balance_enter_idle(int cpu)
8987
{
8988 8989 8990 8991 8992 8993
	/*
	 * If this cpu is going down, then nothing needs to be done.
	 */
	if (!cpu_active(cpu))
		return;

8994 8995 8996 8997
	/* Spare idle load balancing on CPUs that don't want to be disturbed: */
	if (!is_housekeeping_cpu(cpu))
		return;

8998 8999
	if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
		return;
9000

9001 9002 9003 9004 9005 9006
	/*
	 * If we're a completely isolated CPU, we don't play.
	 */
	if (on_null_domain(cpu_rq(cpu)))
		return;

9007 9008 9009
	cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
	atomic_inc(&nohz.nr_cpus);
	set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
9010 9011 9012 9013 9014
}
#endif

static DEFINE_SPINLOCK(balancing);

9015 9016 9017 9018
/*
 * 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.
 */
9019
void update_max_interval(void)
9020 9021 9022 9023
{
	max_load_balance_interval = HZ*num_online_cpus()/10;
}

9024 9025 9026 9027
/*
 * It checks each scheduling domain to see if it is due to be balanced,
 * and initiates a balancing operation if so.
 *
9028
 * Balancing parameters are set up in init_sched_domains.
9029
 */
9030
static void rebalance_domains(struct rq *rq, enum cpu_idle_type idle)
9031
{
9032
	int continue_balancing = 1;
9033
	int cpu = rq->cpu;
9034
	unsigned long interval;
9035
	struct sched_domain *sd;
9036 9037 9038
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
9039 9040
	int need_serialize, need_decay = 0;
	u64 max_cost = 0;
9041

9042
	update_blocked_averages(cpu);
P
Peter Zijlstra 已提交
9043

9044
	rcu_read_lock();
9045
	for_each_domain(cpu, sd) {
9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057
		/*
		 * 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;

9058 9059 9060
		if (!(sd->flags & SD_LOAD_BALANCE))
			continue;

9061 9062 9063 9064 9065 9066 9067 9068 9069 9070 9071
		/*
		 * 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;
		}

9072
		interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
9073 9074 9075 9076 9077 9078 9079 9080

		need_serialize = sd->flags & SD_SERIALIZE;
		if (need_serialize) {
			if (!spin_trylock(&balancing))
				goto out;
		}

		if (time_after_eq(jiffies, sd->last_balance + interval)) {
9081
			if (load_balance(cpu, rq, sd, idle, &continue_balancing)) {
9082
				/*
9083
				 * The LBF_DST_PINNED logic could have changed
9084 9085
				 * env->dst_cpu, so we can't know our idle
				 * state even if we migrated tasks. Update it.
9086
				 */
9087
				idle = idle_cpu(cpu) ? CPU_IDLE : CPU_NOT_IDLE;
9088 9089
			}
			sd->last_balance = jiffies;
9090
			interval = get_sd_balance_interval(sd, idle != CPU_IDLE);
9091 9092 9093 9094 9095 9096 9097 9098
		}
		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;
		}
9099 9100
	}
	if (need_decay) {
9101
		/*
9102 9103
		 * Ensure the rq-wide value also decays but keep it at a
		 * reasonable floor to avoid funnies with rq->avg_idle.
9104
		 */
9105 9106
		rq->max_idle_balance_cost =
			max((u64)sysctl_sched_migration_cost, max_cost);
9107
	}
9108
	rcu_read_unlock();
9109 9110 9111 9112 9113 9114

	/*
	 * 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.
	 */
9115
	if (likely(update_next_balance)) {
9116
		rq->next_balance = next_balance;
9117 9118 9119 9120 9121 9122 9123 9124 9125 9126 9127 9128 9129 9130

#ifdef CONFIG_NO_HZ_COMMON
		/*
		 * If this CPU has been elected to perform the nohz idle
		 * balance. Other idle CPUs have already rebalanced with
		 * nohz_idle_balance() and nohz.next_balance has been
		 * updated accordingly. This CPU is now running the idle load
		 * balance for itself and we need to update the
		 * nohz.next_balance accordingly.
		 */
		if ((idle == CPU_IDLE) && time_after(nohz.next_balance, rq->next_balance))
			nohz.next_balance = rq->next_balance;
#endif
	}
9131 9132
}

9133
#ifdef CONFIG_NO_HZ_COMMON
9134
/*
9135
 * In CONFIG_NO_HZ_COMMON case, the idle balance kickee will do the
9136 9137
 * rebalancing for all the cpus for whom scheduler ticks are stopped.
 */
9138
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle)
9139
{
9140
	int this_cpu = this_rq->cpu;
9141 9142
	struct rq *rq;
	int balance_cpu;
9143 9144 9145
	/* Earliest time when we have to do rebalance again */
	unsigned long next_balance = jiffies + 60*HZ;
	int update_next_balance = 0;
9146

9147 9148 9149
	if (idle != CPU_IDLE ||
	    !test_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu)))
		goto end;
9150 9151

	for_each_cpu(balance_cpu, nohz.idle_cpus_mask) {
9152
		if (balance_cpu == this_cpu || !idle_cpu(balance_cpu))
9153 9154 9155 9156 9157 9158 9159
			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.
		 */
9160
		if (need_resched())
9161 9162
			break;

V
Vincent Guittot 已提交
9163 9164
		rq = cpu_rq(balance_cpu);

9165 9166 9167 9168 9169
		/*
		 * If time for next balance is due,
		 * do the balance.
		 */
		if (time_after_eq(jiffies, rq->next_balance)) {
9170 9171 9172
			struct rq_flags rf;

			rq_lock_irq(rq, &rf);
9173
			update_rq_clock(rq);
9174
			cpu_load_update_idle(rq);
9175 9176
			rq_unlock_irq(rq, &rf);

9177 9178
			rebalance_domains(rq, CPU_IDLE);
		}
9179

9180 9181 9182 9183
		if (time_after(next_balance, rq->next_balance)) {
			next_balance = rq->next_balance;
			update_next_balance = 1;
		}
9184
	}
9185 9186 9187 9188 9189 9190 9191 9192

	/*
	 * next_balance will be updated only when there is a need.
	 * When the CPU is attached to null domain for ex, it will not be
	 * updated.
	 */
	if (likely(update_next_balance))
		nohz.next_balance = next_balance;
9193 9194
end:
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(this_cpu));
9195 9196 9197
}

/*
9198
 * Current heuristic for kicking the idle load balancer in the presence
9199
 * of an idle cpu in the system.
9200
 *   - This rq has more than one task.
9201 9202 9203 9204
 *   - This rq has at least one CFS task and the capacity of the CPU is
 *     significantly reduced because of RT tasks or IRQs.
 *   - At parent of LLC scheduler domain level, this cpu's scheduler group has
 *     multiple busy cpu.
9205 9206
 *   - For SD_ASYM_PACKING, if the lower numbered cpu's in the scheduler
 *     domain span are idle.
9207
 */
9208
static inline bool nohz_kick_needed(struct rq *rq)
9209 9210
{
	unsigned long now = jiffies;
9211
	struct sched_domain_shared *sds;
9212
	struct sched_domain *sd;
T
Tim Chen 已提交
9213
	int nr_busy, i, cpu = rq->cpu;
9214
	bool kick = false;
9215

9216
	if (unlikely(rq->idle_balance))
9217
		return false;
9218

9219 9220 9221 9222
       /*
	* 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.
	*/
9223
	set_cpu_sd_state_busy();
9224
	nohz_balance_exit_idle(cpu);
9225 9226 9227 9228 9229 9230

	/*
	 * None are in tickless mode and hence no need for NOHZ idle load
	 * balancing.
	 */
	if (likely(!atomic_read(&nohz.nr_cpus)))
9231
		return false;
9232 9233

	if (time_before(now, nohz.next_balance))
9234
		return false;
9235

9236
	if (rq->nr_running >= 2)
9237
		return true;
9238

9239
	rcu_read_lock();
9240 9241 9242 9243 9244 9245 9246
	sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
	if (sds) {
		/*
		 * XXX: write a coherent comment on why we do this.
		 * See also: http://lkml.kernel.org/r/20111202010832.602203411@sbsiddha-desk.sc.intel.com
		 */
		nr_busy = atomic_read(&sds->nr_busy_cpus);
9247 9248 9249 9250 9251
		if (nr_busy > 1) {
			kick = true;
			goto unlock;
		}

9252
	}
9253

9254 9255 9256 9257 9258 9259 9260 9261
	sd = rcu_dereference(rq->sd);
	if (sd) {
		if ((rq->cfs.h_nr_running >= 1) &&
				check_cpu_capacity(rq, sd)) {
			kick = true;
			goto unlock;
		}
	}
9262

9263
	sd = rcu_dereference(per_cpu(sd_asym, cpu));
T
Tim Chen 已提交
9264 9265 9266 9267 9268
	if (sd) {
		for_each_cpu(i, sched_domain_span(sd)) {
			if (i == cpu ||
			    !cpumask_test_cpu(i, nohz.idle_cpus_mask))
				continue;
9269

T
Tim Chen 已提交
9270 9271 9272 9273 9274 9275
			if (sched_asym_prefer(i, cpu)) {
				kick = true;
				goto unlock;
			}
		}
	}
9276
unlock:
9277
	rcu_read_unlock();
9278
	return kick;
9279 9280
}
#else
9281
static void nohz_idle_balance(struct rq *this_rq, enum cpu_idle_type idle) { }
9282 9283 9284 9285 9286 9287
#endif

/*
 * run_rebalance_domains is triggered when needed from the scheduler tick.
 * Also triggered for nohz idle balancing (with nohz_balancing_kick set).
 */
9288
static __latent_entropy void run_rebalance_domains(struct softirq_action *h)
9289
{
9290
	struct rq *this_rq = this_rq();
9291
	enum cpu_idle_type idle = this_rq->idle_balance ?
9292 9293 9294
						CPU_IDLE : CPU_NOT_IDLE;

	/*
9295
	 * If this cpu has a pending nohz_balance_kick, then do the
9296
	 * balancing on behalf of the other idle cpus whose ticks are
9297 9298 9299 9300
	 * stopped. Do nohz_idle_balance *before* rebalance_domains to
	 * give the idle cpus a chance to load balance. Else we may
	 * load balance only within the local sched_domain hierarchy
	 * and abort nohz_idle_balance altogether if we pull some load.
9301
	 */
9302
	nohz_idle_balance(this_rq, idle);
9303
	rebalance_domains(this_rq, idle);
9304 9305 9306 9307 9308
}

/*
 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
 */
9309
void trigger_load_balance(struct rq *rq)
9310 9311
{
	/* Don't need to rebalance while attached to NULL domain */
9312 9313 9314 9315
	if (unlikely(on_null_domain(rq)))
		return;

	if (time_after_eq(jiffies, rq->next_balance))
9316
		raise_softirq(SCHED_SOFTIRQ);
9317
#ifdef CONFIG_NO_HZ_COMMON
9318
	if (nohz_kick_needed(rq))
9319
		nohz_balancer_kick();
9320
#endif
9321 9322
}

9323 9324 9325
static void rq_online_fair(struct rq *rq)
{
	update_sysctl();
9326 9327

	update_runtime_enabled(rq);
9328 9329 9330 9331 9332
}

static void rq_offline_fair(struct rq *rq)
{
	update_sysctl();
9333 9334 9335

	/* Ensure any throttled groups are reachable by pick_next_task */
	unthrottle_offline_cfs_rqs(rq);
9336 9337
}

9338
#endif /* CONFIG_SMP */
9339

9340 9341 9342
/*
 * scheduler tick hitting a task of our scheduling class:
 */
P
Peter Zijlstra 已提交
9343
static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
9344 9345 9346 9347 9348 9349
{
	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
9350
		entity_tick(cfs_rq, se, queued);
9351
	}
9352

9353
	if (static_branch_unlikely(&sched_numa_balancing))
9354
		task_tick_numa(rq, curr);
9355 9356 9357
}

/*
P
Peter Zijlstra 已提交
9358 9359 9360
 * called on fork with the child task as argument from the parent's context
 *  - child not yet on the tasklist
 *  - preemption disabled
9361
 */
P
Peter Zijlstra 已提交
9362
static void task_fork_fair(struct task_struct *p)
9363
{
9364 9365
	struct cfs_rq *cfs_rq;
	struct sched_entity *se = &p->se, *curr;
P
Peter Zijlstra 已提交
9366
	struct rq *rq = this_rq();
9367
	struct rq_flags rf;
9368

9369
	rq_lock(rq, &rf);
9370 9371
	update_rq_clock(rq);

9372 9373
	cfs_rq = task_cfs_rq(current);
	curr = cfs_rq->curr;
9374 9375
	if (curr) {
		update_curr(cfs_rq);
9376
		se->vruntime = curr->vruntime;
9377
	}
9378
	place_entity(cfs_rq, se, 1);
9379

P
Peter Zijlstra 已提交
9380
	if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
D
Dmitry Adamushko 已提交
9381
		/*
9382 9383 9384
		 * Upon rescheduling, sched_class::put_prev_task() will place
		 * 'current' within the tree based on its new key value.
		 */
9385
		swap(curr->vruntime, se->vruntime);
9386
		resched_curr(rq);
9387
	}
9388

9389
	se->vruntime -= cfs_rq->min_vruntime;
9390
	rq_unlock(rq, &rf);
9391 9392
}

9393 9394 9395 9396
/*
 * Priority of the task has changed. Check to see if we preempt
 * the current task.
 */
P
Peter Zijlstra 已提交
9397 9398
static void
prio_changed_fair(struct rq *rq, struct task_struct *p, int oldprio)
9399
{
9400
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
9401 9402
		return;

9403 9404 9405 9406 9407
	/*
	 * 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
9408
	if (rq->curr == p) {
9409
		if (p->prio > oldprio)
9410
			resched_curr(rq);
9411
	} else
9412
		check_preempt_curr(rq, p, 0);
9413 9414
}

9415
static inline bool vruntime_normalized(struct task_struct *p)
P
Peter Zijlstra 已提交
9416 9417 9418 9419
{
	struct sched_entity *se = &p->se;

	/*
9420 9421 9422 9423 9424 9425 9426 9427 9428 9429
	 * In both the TASK_ON_RQ_QUEUED and TASK_ON_RQ_MIGRATING cases,
	 * the dequeue_entity(.flags=0) will already have normalized the
	 * vruntime.
	 */
	if (p->on_rq)
		return true;

	/*
	 * When !on_rq, vruntime of the task has usually NOT been normalized.
	 * But there are some cases where it has already been normalized:
P
Peter Zijlstra 已提交
9430
	 *
9431 9432 9433 9434
	 * - A forked child which is waiting for being woken up by
	 *   wake_up_new_task().
	 * - A task which has been woken up by try_to_wake_up() and
	 *   waiting for actually being woken up by sched_ttwu_pending().
P
Peter Zijlstra 已提交
9435
	 */
9436 9437 9438 9439 9440 9441
	if (!se->sum_exec_runtime || p->state == TASK_WAKING)
		return true;

	return false;
}

9442 9443 9444 9445 9446 9447 9448 9449 9450 9451 9452 9453 9454 9455 9456 9457 9458 9459
#ifdef CONFIG_FAIR_GROUP_SCHED
/*
 * Propagate the changes of the sched_entity across the tg tree to make it
 * visible to the root
 */
static void propagate_entity_cfs_rq(struct sched_entity *se)
{
	struct cfs_rq *cfs_rq;

	/* Start to propagate at parent */
	se = se->parent;

	for_each_sched_entity(se) {
		cfs_rq = cfs_rq_of(se);

		if (cfs_rq_throttled(cfs_rq))
			break;

9460
		update_load_avg(cfs_rq, se, UPDATE_TG);
9461 9462 9463 9464 9465 9466
	}
}
#else
static void propagate_entity_cfs_rq(struct sched_entity *se) { }
#endif

9467
static void detach_entity_cfs_rq(struct sched_entity *se)
9468 9469 9470
{
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

9471
	/* Catch up with the cfs_rq and remove our load when we leave */
9472
	update_load_avg(cfs_rq, se, 0);
9473
	detach_entity_load_avg(cfs_rq, se);
9474
	update_tg_load_avg(cfs_rq, false);
9475
	propagate_entity_cfs_rq(se);
P
Peter Zijlstra 已提交
9476 9477
}

9478
static void attach_entity_cfs_rq(struct sched_entity *se)
9479
{
9480
	struct cfs_rq *cfs_rq = cfs_rq_of(se);
9481 9482

#ifdef CONFIG_FAIR_GROUP_SCHED
9483 9484 9485 9486 9487 9488
	/*
	 * Since the real-depth could have been changed (only FAIR
	 * class maintain depth value), reset depth properly.
	 */
	se->depth = se->parent ? se->parent->depth + 1 : 0;
#endif
9489

9490
	/* Synchronize entity with its cfs_rq */
9491
	update_load_avg(cfs_rq, se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
9492
	attach_entity_load_avg(cfs_rq, se);
9493
	update_tg_load_avg(cfs_rq, false);
9494
	propagate_entity_cfs_rq(se);
9495 9496 9497 9498 9499 9500 9501 9502 9503 9504 9505 9506 9507 9508 9509 9510 9511 9512 9513 9514 9515 9516 9517 9518 9519
}

static void detach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	if (!vruntime_normalized(p)) {
		/*
		 * Fix up our vruntime so that the current sleep doesn't
		 * cause 'unlimited' sleep bonus.
		 */
		place_entity(cfs_rq, se, 0);
		se->vruntime -= cfs_rq->min_vruntime;
	}

	detach_entity_cfs_rq(se);
}

static void attach_task_cfs_rq(struct task_struct *p)
{
	struct sched_entity *se = &p->se;
	struct cfs_rq *cfs_rq = cfs_rq_of(se);

	attach_entity_cfs_rq(se);
9520 9521 9522 9523

	if (!vruntime_normalized(p))
		se->vruntime += cfs_rq->min_vruntime;
}
9524

9525 9526 9527 9528 9529 9530 9531 9532
static void switched_from_fair(struct rq *rq, struct task_struct *p)
{
	detach_task_cfs_rq(p);
}

static void switched_to_fair(struct rq *rq, struct task_struct *p)
{
	attach_task_cfs_rq(p);
9533

9534
	if (task_on_rq_queued(p)) {
9535
		/*
9536 9537 9538
		 * 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.
9539
		 */
9540 9541 9542 9543
		if (rq->curr == p)
			resched_curr(rq);
		else
			check_preempt_curr(rq, p, 0);
9544
	}
9545 9546
}

9547 9548 9549 9550 9551 9552 9553 9554 9555
/* 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;

9556 9557 9558 9559 9560 9561 9562
	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);
	}
9563 9564
}

9565 9566
void init_cfs_rq(struct cfs_rq *cfs_rq)
{
9567
	cfs_rq->tasks_timeline = RB_ROOT_CACHED;
9568 9569 9570 9571
	cfs_rq->min_vruntime = (u64)(-(1LL << 20));
#ifndef CONFIG_64BIT
	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
#endif
9572
#ifdef CONFIG_SMP
9573
	raw_spin_lock_init(&cfs_rq->removed.lock);
9574
#endif
9575 9576
}

P
Peter Zijlstra 已提交
9577
#ifdef CONFIG_FAIR_GROUP_SCHED
9578 9579 9580 9581 9582 9583 9584 9585
static void task_set_group_fair(struct task_struct *p)
{
	struct sched_entity *se = &p->se;

	set_task_rq(p, task_cpu(p));
	se->depth = se->parent ? se->parent->depth + 1 : 0;
}

9586
static void task_move_group_fair(struct task_struct *p)
P
Peter Zijlstra 已提交
9587
{
9588
	detach_task_cfs_rq(p);
9589
	set_task_rq(p, task_cpu(p));
9590 9591 9592 9593 9594

#ifdef CONFIG_SMP
	/* Tell se's cfs_rq has been changed -- migrated */
	p->se.avg.last_update_time = 0;
#endif
9595
	attach_task_cfs_rq(p);
P
Peter Zijlstra 已提交
9596
}
9597

9598 9599 9600 9601 9602 9603 9604 9605 9606 9607 9608 9609 9610
static void task_change_group_fair(struct task_struct *p, int type)
{
	switch (type) {
	case TASK_SET_GROUP:
		task_set_group_fair(p);
		break;

	case TASK_MOVE_GROUP:
		task_move_group_fair(p);
		break;
	}
}

9611 9612 9613 9614 9615 9616 9617 9618 9619
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]);
9620
		if (tg->se)
9621 9622 9623 9624 9625 9626 9627 9628 9629 9630
			kfree(tg->se[i]);
	}

	kfree(tg->cfs_rq);
	kfree(tg->se);
}

int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
{
	struct sched_entity *se;
9631
	struct cfs_rq *cfs_rq;
9632 9633 9634 9635 9636 9637 9638 9639 9640 9641 9642 9643 9644 9645 9646 9647 9648 9649 9650 9651 9652 9653 9654 9655 9656 9657
	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]);
9658
		init_entity_runnable_average(se);
9659 9660 9661 9662 9663 9664 9665 9666 9667 9668
	}

	return 1;

err_free_rq:
	kfree(cfs_rq);
err:
	return 0;
}

9669 9670 9671 9672 9673 9674 9675 9676 9677 9678 9679
void online_fair_sched_group(struct task_group *tg)
{
	struct sched_entity *se;
	struct rq *rq;
	int i;

	for_each_possible_cpu(i) {
		rq = cpu_rq(i);
		se = tg->se[i];

		raw_spin_lock_irq(&rq->lock);
9680
		update_rq_clock(rq);
9681
		attach_entity_cfs_rq(se);
9682
		sync_throttle(tg, i);
9683 9684 9685 9686
		raw_spin_unlock_irq(&rq->lock);
	}
}

9687
void unregister_fair_sched_group(struct task_group *tg)
9688 9689
{
	unsigned long flags;
9690 9691
	struct rq *rq;
	int cpu;
9692

9693 9694 9695
	for_each_possible_cpu(cpu) {
		if (tg->se[cpu])
			remove_entity_load_avg(tg->se[cpu]);
9696

9697 9698 9699 9700 9701 9702 9703 9704 9705 9706 9707 9708 9709
		/*
		 * Only empty task groups can be destroyed; so we can speculatively
		 * check on_list without danger of it being re-added.
		 */
		if (!tg->cfs_rq[cpu]->on_list)
			continue;

		rq = cpu_rq(cpu);

		raw_spin_lock_irqsave(&rq->lock, flags);
		list_del_leaf_cfs_rq(tg->cfs_rq[cpu]);
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}
9710 9711 9712 9713 9714 9715 9716 9717 9718 9719 9720 9721 9722 9723 9724 9725 9726 9727 9728
}

void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
			struct sched_entity *se, int cpu,
			struct sched_entity *parent)
{
	struct rq *rq = cpu_rq(cpu);

	cfs_rq->tg = tg;
	cfs_rq->rq = rq;
	init_cfs_rq_runtime(cfs_rq);

	tg->cfs_rq[cpu] = cfs_rq;
	tg->se[cpu] = se;

	/* se could be NULL for root_task_group */
	if (!se)
		return;

P
Peter Zijlstra 已提交
9729
	if (!parent) {
9730
		se->cfs_rq = &rq->cfs;
P
Peter Zijlstra 已提交
9731 9732
		se->depth = 0;
	} else {
9733
		se->cfs_rq = parent->my_q;
P
Peter Zijlstra 已提交
9734 9735
		se->depth = parent->depth + 1;
	}
9736 9737

	se->my_q = cfs_rq;
9738 9739
	/* guarantee group entities always have weight */
	update_load_set(&se->load, NICE_0_LOAD);
9740 9741 9742 9743 9744 9745 9746 9747 9748 9749 9750 9751 9752 9753 9754 9755 9756 9757 9758 9759 9760 9761 9762 9763
	se->parent = parent;
}

static DEFINE_MUTEX(shares_mutex);

int sched_group_set_shares(struct task_group *tg, unsigned long shares)
{
	int i;

	/*
	 * We can't change the weight of the root cgroup.
	 */
	if (!tg->se[0])
		return -EINVAL;

	shares = clamp(shares, scale_load(MIN_SHARES), scale_load(MAX_SHARES));

	mutex_lock(&shares_mutex);
	if (tg->shares == shares)
		goto done;

	tg->shares = shares;
	for_each_possible_cpu(i) {
		struct rq *rq = cpu_rq(i);
9764 9765
		struct sched_entity *se = tg->se[i];
		struct rq_flags rf;
9766 9767

		/* Propagate contribution to hierarchy */
9768
		rq_lock_irqsave(rq, &rf);
9769
		update_rq_clock(rq);
9770
		for_each_sched_entity(se) {
9771
			update_load_avg(cfs_rq_of(se), se, UPDATE_TG);
9772
			update_cfs_group(se);
9773
		}
9774
		rq_unlock_irqrestore(rq, &rf);
9775 9776 9777 9778 9779 9780 9781 9782 9783 9784 9785 9786 9787 9788 9789
	}

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;
}

9790 9791
void online_fair_sched_group(struct task_group *tg) { }

9792
void unregister_fair_sched_group(struct task_group *tg) { }
9793 9794 9795

#endif /* CONFIG_FAIR_GROUP_SCHED */

P
Peter Zijlstra 已提交
9796

9797
static unsigned int get_rr_interval_fair(struct rq *rq, struct task_struct *task)
9798 9799 9800 9801 9802 9803 9804 9805 9806
{
	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)
9807
		rr_interval = NS_TO_JIFFIES(sched_slice(cfs_rq_of(se), se));
9808 9809 9810 9811

	return rr_interval;
}

9812 9813 9814
/*
 * All the scheduling class methods:
 */
9815
const struct sched_class fair_sched_class = {
9816
	.next			= &idle_sched_class,
9817 9818 9819
	.enqueue_task		= enqueue_task_fair,
	.dequeue_task		= dequeue_task_fair,
	.yield_task		= yield_task_fair,
9820
	.yield_to_task		= yield_to_task_fair,
9821

I
Ingo Molnar 已提交
9822
	.check_preempt_curr	= check_preempt_wakeup,
9823 9824 9825 9826

	.pick_next_task		= pick_next_task_fair,
	.put_prev_task		= put_prev_task_fair,

9827
#ifdef CONFIG_SMP
L
Li Zefan 已提交
9828
	.select_task_rq		= select_task_rq_fair,
9829
	.migrate_task_rq	= migrate_task_rq_fair,
9830

9831 9832
	.rq_online		= rq_online_fair,
	.rq_offline		= rq_offline_fair,
9833

9834
	.task_dead		= task_dead_fair,
9835
	.set_cpus_allowed	= set_cpus_allowed_common,
9836
#endif
9837

9838
	.set_curr_task          = set_curr_task_fair,
9839
	.task_tick		= task_tick_fair,
P
Peter Zijlstra 已提交
9840
	.task_fork		= task_fork_fair,
9841 9842

	.prio_changed		= prio_changed_fair,
P
Peter Zijlstra 已提交
9843
	.switched_from		= switched_from_fair,
9844
	.switched_to		= switched_to_fair,
P
Peter Zijlstra 已提交
9845

9846 9847
	.get_rr_interval	= get_rr_interval_fair,

9848 9849
	.update_curr		= update_curr_fair,

P
Peter Zijlstra 已提交
9850
#ifdef CONFIG_FAIR_GROUP_SCHED
9851
	.task_change_group	= task_change_group_fair,
P
Peter Zijlstra 已提交
9852
#endif
9853 9854 9855
};

#ifdef CONFIG_SCHED_DEBUG
9856
void print_cfs_stats(struct seq_file *m, int cpu)
9857
{
9858
	struct cfs_rq *cfs_rq, *pos;
9859

9860
	rcu_read_lock();
9861
	for_each_leaf_cfs_rq_safe(cpu_rq(cpu), cfs_rq, pos)
9862
		print_cfs_rq(m, cpu, cfs_rq);
9863
	rcu_read_unlock();
9864
}
9865 9866 9867 9868 9869 9870 9871 9872 9873 9874 9875 9876 9877 9878 9879 9880 9881 9882 9883 9884 9885

#ifdef CONFIG_NUMA_BALANCING
void show_numa_stats(struct task_struct *p, struct seq_file *m)
{
	int node;
	unsigned long tsf = 0, tpf = 0, gsf = 0, gpf = 0;

	for_each_online_node(node) {
		if (p->numa_faults) {
			tsf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 0)];
			tpf = p->numa_faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		if (p->numa_group) {
			gsf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 0)],
			gpf = p->numa_group->faults[task_faults_idx(NUMA_MEM, node, 1)];
		}
		print_numa_stats(m, node, tsf, tpf, gsf, gpf);
	}
}
#endif /* CONFIG_NUMA_BALANCING */
#endif /* CONFIG_SCHED_DEBUG */
9886 9887 9888 9889 9890 9891

__init void init_sched_fair_class(void)
{
#ifdef CONFIG_SMP
	open_softirq(SCHED_SOFTIRQ, run_rebalance_domains);

9892
#ifdef CONFIG_NO_HZ_COMMON
9893
	nohz.next_balance = jiffies;
9894 9895 9896 9897 9898
	zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT);
#endif
#endif /* SMP */

}