core.c 178.6 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 *  kernel/sched/core.c
L
Linus Torvalds 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 *
 *  Kernel scheduler and related syscalls
 *
 *  Copyright (C) 1991-2002  Linus Torvalds
 *
 *  1996-12-23  Modified by Dave Grothe to fix bugs in semaphores and
 *		make semaphores SMP safe
 *  1998-11-19	Implemented schedule_timeout() and related stuff
 *		by Andrea Arcangeli
 *  2002-01-04	New ultra-scalable O(1) scheduler by Ingo Molnar:
 *		hybrid priority-list and round-robin design with
 *		an array-switch method of distributing timeslices
 *		and per-CPU runqueues.  Cleanups and useful suggestions
 *		by Davide Libenzi, preemptible kernel bits by Robert Love.
 *  2003-09-03	Interactivity tuning by Con Kolivas.
 *  2004-04-02	Scheduler domains code by Nick Piggin
I
Ingo Molnar 已提交
19 20 21 22 23 24
 *  2007-04-15  Work begun on replacing all interactivity tuning with a
 *              fair scheduling design by Con Kolivas.
 *  2007-05-05  Load balancing (smp-nice) and other improvements
 *              by Peter Williams
 *  2007-05-06  Interactivity improvements to CFS by Mike Galbraith
 *  2007-07-01  Group scheduling enhancements by Srivatsa Vaddagiri
25 26
 *  2007-11-29  RT balancing improvements by Steven Rostedt, Gregory Haskins,
 *              Thomas Gleixner, Mike Kravetz
L
Linus Torvalds 已提交
27 28 29 30 31 32
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/nmi.h>
#include <linux/init.h>
33
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
34 35 36
#include <linux/highmem.h>
#include <asm/mmu_context.h>
#include <linux/interrupt.h>
37
#include <linux/capability.h>
L
Linus Torvalds 已提交
38 39
#include <linux/completion.h>
#include <linux/kernel_stat.h>
40
#include <linux/debug_locks.h>
41
#include <linux/perf_event.h>
L
Linus Torvalds 已提交
42 43 44
#include <linux/security.h>
#include <linux/notifier.h>
#include <linux/profile.h>
45
#include <linux/freezer.h>
46
#include <linux/vmalloc.h>
L
Linus Torvalds 已提交
47 48
#include <linux/blkdev.h>
#include <linux/delay.h>
49
#include <linux/pid_namespace.h>
L
Linus Torvalds 已提交
50 51 52 53 54 55 56
#include <linux/smp.h>
#include <linux/threads.h>
#include <linux/timer.h>
#include <linux/rcupdate.h>
#include <linux/cpu.h>
#include <linux/cpuset.h>
#include <linux/percpu.h>
57
#include <linux/proc_fs.h>
L
Linus Torvalds 已提交
58
#include <linux/seq_file.h>
59
#include <linux/sysctl.h>
L
Linus Torvalds 已提交
60 61
#include <linux/syscalls.h>
#include <linux/times.h>
62
#include <linux/tsacct_kern.h>
63
#include <linux/kprobes.h>
64
#include <linux/delayacct.h>
65
#include <linux/unistd.h>
J
Jens Axboe 已提交
66
#include <linux/pagemap.h>
P
Peter Zijlstra 已提交
67
#include <linux/hrtimer.h>
R
Reynes Philippe 已提交
68
#include <linux/tick.h>
P
Peter Zijlstra 已提交
69 70
#include <linux/debugfs.h>
#include <linux/ctype.h>
71
#include <linux/ftrace.h>
72
#include <linux/slab.h>
73
#include <linux/init_task.h>
A
Al Viro 已提交
74
#include <linux/binfmts.h>
75
#include <linux/context_tracking.h>
L
Linus Torvalds 已提交
76

77
#include <asm/switch_to.h>
78
#include <asm/tlb.h>
79
#include <asm/irq_regs.h>
80
#include <asm/mutex.h>
G
Glauber Costa 已提交
81 82 83
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
#endif
L
Linus Torvalds 已提交
84

85
#include "sched.h"
86
#include "../workqueue_internal.h"
87
#include "../smpboot.h"
88

89
#define CREATE_TRACE_POINTS
90
#include <trace/events/sched.h>
91

92
void start_bandwidth_timer(struct hrtimer *period_timer, ktime_t period)
93
{
94 95
	unsigned long delta;
	ktime_t soft, hard, now;
96

97 98 99 100 101 102
	for (;;) {
		if (hrtimer_active(period_timer))
			break;

		now = hrtimer_cb_get_time(period_timer);
		hrtimer_forward(period_timer, now, period);
103

104 105 106 107 108 109 110 111
		soft = hrtimer_get_softexpires(period_timer);
		hard = hrtimer_get_expires(period_timer);
		delta = ktime_to_ns(ktime_sub(hard, soft));
		__hrtimer_start_range_ns(period_timer, soft, delta,
					 HRTIMER_MODE_ABS_PINNED, 0);
	}
}

112 113
DEFINE_MUTEX(sched_domains_mutex);
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
114

115
static void update_rq_clock_task(struct rq *rq, s64 delta);
116

117
void update_rq_clock(struct rq *rq)
118
{
119
	s64 delta;
120

121
	if (rq->skip_clock_update > 0)
122
		return;
123

124 125 126
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
127 128
}

I
Ingo Molnar 已提交
129 130 131
/*
 * Debugging: various feature bits
 */
P
Peter Zijlstra 已提交
132 133 134 135

#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |

I
Ingo Molnar 已提交
136
const_debug unsigned int sysctl_sched_features =
137
#include "features.h"
P
Peter Zijlstra 已提交
138 139 140 141 142 143 144 145
	0;

#undef SCHED_FEAT

#ifdef CONFIG_SCHED_DEBUG
#define SCHED_FEAT(name, enabled)	\
	#name ,

146
static const char * const sched_feat_names[] = {
147
#include "features.h"
P
Peter Zijlstra 已提交
148 149 150 151
};

#undef SCHED_FEAT

L
Li Zefan 已提交
152
static int sched_feat_show(struct seq_file *m, void *v)
P
Peter Zijlstra 已提交
153 154 155
{
	int i;

156
	for (i = 0; i < __SCHED_FEAT_NR; i++) {
L
Li Zefan 已提交
157 158 159
		if (!(sysctl_sched_features & (1UL << i)))
			seq_puts(m, "NO_");
		seq_printf(m, "%s ", sched_feat_names[i]);
P
Peter Zijlstra 已提交
160
	}
L
Li Zefan 已提交
161
	seq_puts(m, "\n");
P
Peter Zijlstra 已提交
162

L
Li Zefan 已提交
163
	return 0;
P
Peter Zijlstra 已提交
164 165
}

166 167
#ifdef HAVE_JUMP_LABEL

168 169
#define jump_label_key__true  STATIC_KEY_INIT_TRUE
#define jump_label_key__false STATIC_KEY_INIT_FALSE
170 171 172 173

#define SCHED_FEAT(name, enabled)	\
	jump_label_key__##enabled ,

174
struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
175 176 177 178 179 180 181
#include "features.h"
};

#undef SCHED_FEAT

static void sched_feat_disable(int i)
{
182 183
	if (static_key_enabled(&sched_feat_keys[i]))
		static_key_slow_dec(&sched_feat_keys[i]);
184 185 186 187
}

static void sched_feat_enable(int i)
{
188 189
	if (!static_key_enabled(&sched_feat_keys[i]))
		static_key_slow_inc(&sched_feat_keys[i]);
190 191 192 193 194 195
}
#else
static void sched_feat_disable(int i) { };
static void sched_feat_enable(int i) { };
#endif /* HAVE_JUMP_LABEL */

196
static int sched_feat_set(char *cmp)
P
Peter Zijlstra 已提交
197 198
{
	int i;
199
	int neg = 0;
P
Peter Zijlstra 已提交
200

H
Hillf Danton 已提交
201
	if (strncmp(cmp, "NO_", 3) == 0) {
P
Peter Zijlstra 已提交
202 203 204 205
		neg = 1;
		cmp += 3;
	}

206
	for (i = 0; i < __SCHED_FEAT_NR; i++) {
207
		if (strcmp(cmp, sched_feat_names[i]) == 0) {
208
			if (neg) {
P
Peter Zijlstra 已提交
209
				sysctl_sched_features &= ~(1UL << i);
210 211
				sched_feat_disable(i);
			} else {
P
Peter Zijlstra 已提交
212
				sysctl_sched_features |= (1UL << i);
213 214
				sched_feat_enable(i);
			}
P
Peter Zijlstra 已提交
215 216 217 218
			break;
		}
	}

219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239
	return i;
}

static ssize_t
sched_feat_write(struct file *filp, const char __user *ubuf,
		size_t cnt, loff_t *ppos)
{
	char buf[64];
	char *cmp;
	int i;

	if (cnt > 63)
		cnt = 63;

	if (copy_from_user(&buf, ubuf, cnt))
		return -EFAULT;

	buf[cnt] = 0;
	cmp = strstrip(buf);

	i = sched_feat_set(cmp);
240
	if (i == __SCHED_FEAT_NR)
P
Peter Zijlstra 已提交
241 242
		return -EINVAL;

243
	*ppos += cnt;
P
Peter Zijlstra 已提交
244 245 246 247

	return cnt;
}

L
Li Zefan 已提交
248 249 250 251 252
static int sched_feat_open(struct inode *inode, struct file *filp)
{
	return single_open(filp, sched_feat_show, NULL);
}

253
static const struct file_operations sched_feat_fops = {
L
Li Zefan 已提交
254 255 256 257 258
	.open		= sched_feat_open,
	.write		= sched_feat_write,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
P
Peter Zijlstra 已提交
259 260 261 262 263 264 265 266 267 268
};

static __init int sched_init_debug(void)
{
	debugfs_create_file("sched_features", 0644, NULL, NULL,
			&sched_feat_fops);

	return 0;
}
late_initcall(sched_init_debug);
269
#endif /* CONFIG_SCHED_DEBUG */
I
Ingo Molnar 已提交
270

271 272 273 274 275 276
/*
 * Number of tasks to iterate in a single balance run.
 * Limited because this is done with IRQs disabled.
 */
const_debug unsigned int sysctl_sched_nr_migrate = 32;

277 278 279 280 281 282 283 284
/*
 * period over which we average the RT time consumption, measured
 * in ms.
 *
 * default: 1s
 */
const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;

P
Peter Zijlstra 已提交
285
/*
P
Peter Zijlstra 已提交
286
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
287 288
 * default: 1s
 */
P
Peter Zijlstra 已提交
289
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
290

291
__read_mostly int scheduler_running;
292

P
Peter Zijlstra 已提交
293 294 295 296 297
/*
 * part of the period that we allow rt tasks to run in us.
 * default: 0.95s
 */
int sysctl_sched_rt_runtime = 950000;
P
Peter Zijlstra 已提交
298 299


L
Linus Torvalds 已提交
300

301
/*
302
 * __task_rq_lock - lock the rq @p resides on.
303
 */
304
static inline struct rq *__task_rq_lock(struct task_struct *p)
305 306
	__acquires(rq->lock)
{
307 308
	struct rq *rq;

309 310
	lockdep_assert_held(&p->pi_lock);

311
	for (;;) {
312
		rq = task_rq(p);
313
		raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
314
		if (likely(rq == task_rq(p)))
315
			return rq;
316
		raw_spin_unlock(&rq->lock);
317 318 319
	}
}

L
Linus Torvalds 已提交
320
/*
321
 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
L
Linus Torvalds 已提交
322
 */
323
static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags)
324
	__acquires(p->pi_lock)
L
Linus Torvalds 已提交
325 326
	__acquires(rq->lock)
{
327
	struct rq *rq;
L
Linus Torvalds 已提交
328

329
	for (;;) {
330
		raw_spin_lock_irqsave(&p->pi_lock, *flags);
331
		rq = task_rq(p);
332
		raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
333
		if (likely(rq == task_rq(p)))
334
			return rq;
335 336
		raw_spin_unlock(&rq->lock);
		raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
L
Linus Torvalds 已提交
337 338 339
	}
}

A
Alexey Dobriyan 已提交
340
static void __task_rq_unlock(struct rq *rq)
341 342
	__releases(rq->lock)
{
343
	raw_spin_unlock(&rq->lock);
344 345
}

346 347
static inline void
task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags)
L
Linus Torvalds 已提交
348
	__releases(rq->lock)
349
	__releases(p->pi_lock)
L
Linus Torvalds 已提交
350
{
351 352
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&p->pi_lock, *flags);
L
Linus Torvalds 已提交
353 354 355
}

/*
356
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
357
 */
A
Alexey Dobriyan 已提交
358
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
359 360
	__acquires(rq->lock)
{
361
	struct rq *rq;
L
Linus Torvalds 已提交
362 363 364

	local_irq_disable();
	rq = this_rq();
365
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
366 367 368 369

	return rq;
}

P
Peter Zijlstra 已提交
370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397
#ifdef CONFIG_SCHED_HRTICK
/*
 * Use HR-timers to deliver accurate preemption points.
 *
 * Its all a bit involved since we cannot program an hrt while holding the
 * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a
 * reschedule event.
 *
 * When we get rescheduled we reprogram the hrtick_timer outside of the
 * rq->lock.
 */

static void hrtick_clear(struct rq *rq)
{
	if (hrtimer_active(&rq->hrtick_timer))
		hrtimer_cancel(&rq->hrtick_timer);
}

/*
 * High-resolution timer tick.
 * Runs from hardirq context with interrupts disabled.
 */
static enum hrtimer_restart hrtick(struct hrtimer *timer)
{
	struct rq *rq = container_of(timer, struct rq, hrtick_timer);

	WARN_ON_ONCE(cpu_of(rq) != smp_processor_id());

398
	raw_spin_lock(&rq->lock);
399
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
400
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
401
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
402 403 404 405

	return HRTIMER_NORESTART;
}

406
#ifdef CONFIG_SMP
407 408 409 410
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
411
{
412
	struct rq *rq = arg;
413

414
	raw_spin_lock(&rq->lock);
415 416
	hrtimer_restart(&rq->hrtick_timer);
	rq->hrtick_csd_pending = 0;
417
	raw_spin_unlock(&rq->lock);
418 419
}

420 421 422 423 424
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
425
void hrtick_start(struct rq *rq, u64 delay)
426
{
427 428
	struct hrtimer *timer = &rq->hrtick_timer;
	ktime_t time = ktime_add_ns(timer->base->get_time(), delay);
429

430
	hrtimer_set_expires(timer, time);
431 432 433 434

	if (rq == this_rq()) {
		hrtimer_restart(timer);
	} else if (!rq->hrtick_csd_pending) {
435
		__smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
436 437
		rq->hrtick_csd_pending = 1;
	}
438 439 440 441 442 443 444 445 446 447 448 449 450 451
}

static int
hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
{
	int cpu = (int)(long)hcpu;

	switch (action) {
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
	case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
452
		hrtick_clear(cpu_rq(cpu));
453 454 455 456 457 458
		return NOTIFY_OK;
	}

	return NOTIFY_DONE;
}

459
static __init void init_hrtick(void)
460 461 462
{
	hotcpu_notifier(hotplug_hrtick, 0);
}
463 464 465 466 467 468
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
469
void hrtick_start(struct rq *rq, u64 delay)
470
{
471
	__hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0,
472
			HRTIMER_MODE_REL_PINNED, 0);
473
}
474

A
Andrew Morton 已提交
475
static inline void init_hrtick(void)
P
Peter Zijlstra 已提交
476 477
{
}
478
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
479

480
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
481
{
482 483
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
484

485 486 487 488
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
489

490 491
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
492
}
A
Andrew Morton 已提交
493
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
494 495 496 497 498 499 500 501
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}

502 503 504
static inline void init_hrtick(void)
{
}
A
Andrew Morton 已提交
505
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
506

I
Ingo Molnar 已提交
507 508 509 510 511 512 513 514
/*
 * resched_task - mark a task 'to be rescheduled now'.
 *
 * On UP this means the setting of the need_resched flag, on SMP it
 * might also involve a cross-CPU call to trigger the scheduler on
 * the target CPU.
 */
#ifdef CONFIG_SMP
515
void resched_task(struct task_struct *p)
I
Ingo Molnar 已提交
516 517 518
{
	int cpu;

519
	assert_raw_spin_locked(&task_rq(p)->lock);
I
Ingo Molnar 已提交
520

521
	if (test_tsk_need_resched(p))
I
Ingo Molnar 已提交
522 523
		return;

524
	set_tsk_need_resched(p);
I
Ingo Molnar 已提交
525 526 527 528 529 530 531 532 533 534 535

	cpu = task_cpu(p);
	if (cpu == smp_processor_id())
		return;

	/* NEED_RESCHED must be visible before we test polling */
	smp_mb();
	if (!tsk_is_polling(p))
		smp_send_reschedule(cpu);
}

536
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
537 538 539 540
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

541
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
542 543
		return;
	resched_task(cpu_curr(cpu));
544
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
545
}
546

547
#ifdef CONFIG_NO_HZ_COMMON
548 549 550 551 552 553 554 555 556 557 558 559 560 561
/*
 * In the semi idle case, use the nearest busy cpu for migrating timers
 * from an idle cpu.  This is good for power-savings.
 *
 * We don't do similar optimization for completely idle system, as
 * selecting an idle cpu will add more delays to the timers than intended
 * (as that cpu's timer base may not be uptodate wrt jiffies etc).
 */
int get_nohz_timer_target(void)
{
	int cpu = smp_processor_id();
	int i;
	struct sched_domain *sd;

562
	rcu_read_lock();
563
	for_each_domain(cpu, sd) {
564 565 566 567 568 569
		for_each_cpu(i, sched_domain_span(sd)) {
			if (!idle_cpu(i)) {
				cpu = i;
				goto unlock;
			}
		}
570
	}
571 572
unlock:
	rcu_read_unlock();
573 574
	return cpu;
}
575 576 577 578 579 580 581 582 583 584
/*
 * When add_timer_on() enqueues a timer into the timer wheel of an
 * idle CPU then this timer might expire before the next timer event
 * which is scheduled to wake up that CPU. In case of a completely
 * idle system the next event might even be infinite time into the
 * future. wake_up_idle_cpu() ensures that the CPU is woken up and
 * leaves the inner idle loop so the newly added timer is taken into
 * account when the CPU goes back to idle and evaluates the timer
 * wheel for the next timer event.
 */
585
static void wake_up_idle_cpu(int cpu)
586 587 588 589 590 591 592 593 594 595 596 597 598 599 600
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

	/*
	 * This is safe, as this function is called with the timer
	 * wheel base lock of (cpu) held. When the CPU is on the way
	 * to idle and has not yet set rq->curr to idle then it will
	 * be serialized on the timer wheel base lock and take the new
	 * timer into account automatically.
	 */
	if (rq->curr != rq->idle)
		return;
601 602

	/*
603 604 605
	 * We can set TIF_RESCHED on the idle task of the other CPU
	 * lockless. The worst case is that the other CPU runs the
	 * idle task through an additional NOOP schedule()
606
	 */
607
	set_tsk_need_resched(rq->idle);
608

609 610 611 612
	/* NEED_RESCHED must be visible before we test polling */
	smp_mb();
	if (!tsk_is_polling(rq->idle))
		smp_send_reschedule(cpu);
613 614
}

615
static bool wake_up_full_nohz_cpu(int cpu)
616
{
617
	if (tick_nohz_full_cpu(cpu)) {
618 619 620 621 622 623 624 625 626 627 628
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
			smp_send_reschedule(cpu);
		return true;
	}

	return false;
}

void wake_up_nohz_cpu(int cpu)
{
629
	if (!wake_up_full_nohz_cpu(cpu))
630 631 632
		wake_up_idle_cpu(cpu);
}

633
static inline bool got_nohz_idle_kick(void)
634
{
635
	int cpu = smp_processor_id();
636 637 638 639 640 641 642 643 644 645 646 647 648

	if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)))
		return false;

	if (idle_cpu(cpu) && !need_resched())
		return true;

	/*
	 * We can't run Idle Load Balance on this CPU for this time so we
	 * cancel it and clear NOHZ_BALANCE_KICK
	 */
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu));
	return false;
649 650
}

651
#else /* CONFIG_NO_HZ_COMMON */
652

653
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
654
{
655
	return false;
P
Peter Zijlstra 已提交
656 657
}

658
#endif /* CONFIG_NO_HZ_COMMON */
659

660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676
#ifdef CONFIG_NO_HZ_FULL
bool sched_can_stop_tick(void)
{
       struct rq *rq;

       rq = this_rq();

       /* Make sure rq->nr_running update is visible after the IPI */
       smp_rmb();

       /* More than one running task need preemption */
       if (rq->nr_running > 1)
               return false;

       return true;
}
#endif /* CONFIG_NO_HZ_FULL */
677

678
void sched_avg_update(struct rq *rq)
679
{
680 681
	s64 period = sched_avg_period();

682
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
683 684 685 686 687 688
		/*
		 * Inline assembly required to prevent the compiler
		 * optimising this loop into a divmod call.
		 * See __iter_div_u64_rem() for another example of this.
		 */
		asm("" : "+rm" (rq->age_stamp));
689 690 691
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
692 693
}

694
#else /* !CONFIG_SMP */
695
void resched_task(struct task_struct *p)
696
{
697
	assert_raw_spin_locked(&task_rq(p)->lock);
698
	set_tsk_need_resched(p);
699
}
700
#endif /* CONFIG_SMP */
701

702 703
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
704
/*
705 706 707 708
 * Iterate task_group tree rooted at *from, calling @down when first entering a
 * node and @up when leaving it for the final time.
 *
 * Caller must hold rcu_lock or sufficient equivalent.
709
 */
710
int walk_tg_tree_from(struct task_group *from,
711
			     tg_visitor down, tg_visitor up, void *data)
712 713
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
714
	int ret;
715

716 717
	parent = from;

718
down:
P
Peter Zijlstra 已提交
719 720
	ret = (*down)(parent, data);
	if (ret)
721
		goto out;
722 723 724 725 726 727 728
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
729
	ret = (*up)(parent, data);
730 731
	if (ret || parent == from)
		goto out;
732 733 734 735 736

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
737
out:
P
Peter Zijlstra 已提交
738
	return ret;
739 740
}

741
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
742
{
743
	return 0;
P
Peter Zijlstra 已提交
744
}
745 746
#endif

747 748
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
749 750 751
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
752 753 754 755
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
	if (p->policy == SCHED_IDLE) {
756
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
757
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
758 759
		return;
	}
760

761
	load->weight = scale_load(prio_to_weight[prio]);
N
Nikhil Rao 已提交
762
	load->inv_weight = prio_to_wmult[prio];
763 764
}

765
static void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
766
{
767
	update_rq_clock(rq);
I
Ingo Molnar 已提交
768
	sched_info_queued(p);
769
	p->sched_class->enqueue_task(rq, p, flags);
770 771
}

772
static void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
773
{
774
	update_rq_clock(rq);
775
	sched_info_dequeued(p);
776
	p->sched_class->dequeue_task(rq, p, flags);
777 778
}

779
void activate_task(struct rq *rq, struct task_struct *p, int flags)
780 781 782 783
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

784
	enqueue_task(rq, p, flags);
785 786
}

787
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
788 789 790 791
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

792
	dequeue_task(rq, p, flags);
793 794
}

795
static void update_rq_clock_task(struct rq *rq, s64 delta)
796
{
797 798 799 800 801 802 803 804
/*
 * In theory, the compile should just see 0 here, and optimize out the call
 * to sched_rt_avg_update. But I don't trust it...
 */
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
	s64 steal = 0, irq_delta = 0;
#endif
#ifdef CONFIG_IRQ_TIME_ACCOUNTING
805
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826

	/*
	 * Since irq_time is only updated on {soft,}irq_exit, we might run into
	 * this case when a previous update_rq_clock() happened inside a
	 * {soft,}irq region.
	 *
	 * When this happens, we stop ->clock_task and only update the
	 * prev_irq_time stamp to account for the part that fit, so that a next
	 * update will consume the rest. This ensures ->clock_task is
	 * monotonic.
	 *
	 * It does however cause some slight miss-attribution of {soft,}irq
	 * time, a more accurate solution would be to update the irq_time using
	 * the current rq->clock timestamp, except that would require using
	 * atomic ops.
	 */
	if (irq_delta > delta)
		irq_delta = delta;

	rq->prev_irq_time += irq_delta;
	delta -= irq_delta;
827 828
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
829
	if (static_key_false((&paravirt_steal_rq_enabled))) {
830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846
		u64 st;

		steal = paravirt_steal_clock(cpu_of(rq));
		steal -= rq->prev_steal_time_rq;

		if (unlikely(steal > delta))
			steal = delta;

		st = steal_ticks(steal);
		steal = st * TICK_NSEC;

		rq->prev_steal_time_rq += steal;

		delta -= steal;
	}
#endif

847 848
	rq->clock_task += delta;

849 850 851 852
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
	if ((irq_delta + steal) && sched_feat(NONTASK_POWER))
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
853 854
}

855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884
void sched_set_stop_task(int cpu, struct task_struct *stop)
{
	struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
	struct task_struct *old_stop = cpu_rq(cpu)->stop;

	if (stop) {
		/*
		 * Make it appear like a SCHED_FIFO task, its something
		 * userspace knows about and won't get confused about.
		 *
		 * Also, it will make PI more or less work without too
		 * much confusion -- but then, stop work should not
		 * rely on PI working anyway.
		 */
		sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);

		stop->sched_class = &stop_sched_class;
	}

	cpu_rq(cpu)->stop = stop;

	if (old_stop) {
		/*
		 * Reset it back to a normal scheduling class so that
		 * it can die in pieces.
		 */
		old_stop->sched_class = &rt_sched_class;
	}
}

885
/*
I
Ingo Molnar 已提交
886
 * __normal_prio - return the priority that is based on the static prio
887 888 889
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
890
	return p->static_prio;
891 892
}

893 894 895 896 897 898 899
/*
 * Calculate the expected normal priority: i.e. priority
 * without taking RT-inheritance into account. Might be
 * boosted by interactivity modifiers. Changes upon fork,
 * setprio syscalls, and whenever the interactivity
 * estimator recalculates.
 */
900
static inline int normal_prio(struct task_struct *p)
901 902 903
{
	int prio;

904
	if (task_has_rt_policy(p))
905 906 907 908 909 910 911 912 913 914 915 916 917
		prio = MAX_RT_PRIO-1 - p->rt_priority;
	else
		prio = __normal_prio(p);
	return prio;
}

/*
 * Calculate the current priority, i.e. the priority
 * taken into account by the scheduler. This value might
 * be boosted by RT tasks, or might be boosted by
 * interactivity modifiers. Will be RT if the task got
 * RT-boosted. If not then it returns p->normal_prio.
 */
918
static int effective_prio(struct task_struct *p)
919 920 921 922 923 924 925 926 927 928 929 930
{
	p->normal_prio = normal_prio(p);
	/*
	 * If we are RT tasks or we were boosted to RT priority,
	 * keep the priority unchanged. Otherwise, update priority
	 * to the normal priority:
	 */
	if (!rt_prio(p->prio))
		return p->normal_prio;
	return p->prio;
}

L
Linus Torvalds 已提交
931 932 933 934
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
 */
935
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
936 937 938 939
{
	return cpu_curr(task_cpu(p)) == p;
}

940 941
static inline void check_class_changed(struct rq *rq, struct task_struct *p,
				       const struct sched_class *prev_class,
P
Peter Zijlstra 已提交
942
				       int oldprio)
943 944 945
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
946 947 948 949
			prev_class->switched_from(rq, p);
		p->sched_class->switched_to(rq, p);
	} else if (oldprio != p->prio)
		p->sched_class->prio_changed(rq, p, oldprio);
950 951
}

952
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972
{
	const struct sched_class *class;

	if (p->sched_class == rq->curr->sched_class) {
		rq->curr->sched_class->check_preempt_curr(rq, p, flags);
	} else {
		for_each_class(class) {
			if (class == rq->curr->sched_class)
				break;
			if (class == p->sched_class) {
				resched_task(rq->curr);
				break;
			}
		}
	}

	/*
	 * A queue event has occurred, and we're going to schedule.  In
	 * this case, we can save a useless back to back clock update.
	 */
P
Peter Zijlstra 已提交
973
	if (rq->curr->on_rq && test_tsk_need_resched(rq->curr))
974 975 976
		rq->skip_clock_update = 1;
}

977 978 979 980 981 982 983
static ATOMIC_NOTIFIER_HEAD(task_migration_notifier);

void register_task_migration_notifier(struct notifier_block *n)
{
	atomic_notifier_chain_register(&task_migration_notifier, n);
}

L
Linus Torvalds 已提交
984
#ifdef CONFIG_SMP
I
Ingo Molnar 已提交
985
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
986
{
987 988 989 990 991
#ifdef CONFIG_SCHED_DEBUG
	/*
	 * We should never call set_task_cpu() on a blocked task,
	 * ttwu() will sort out the placement.
	 */
P
Peter Zijlstra 已提交
992 993
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
			!(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE));
994 995

#ifdef CONFIG_LOCKDEP
996 997 998 999 1000
	/*
	 * The caller should hold either p->pi_lock or rq->lock, when changing
	 * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
	 *
	 * sched_move_task() holds both and thus holding either pins the cgroup,
P
Peter Zijlstra 已提交
1001
	 * see task_group().
1002 1003 1004 1005
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1006 1007 1008
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1009 1010
#endif

1011
	trace_sched_migrate_task(p, new_cpu);
1012

1013
	if (task_cpu(p) != new_cpu) {
1014 1015
		struct task_migration_notifier tmn;

1016 1017
		if (p->sched_class->migrate_task_rq)
			p->sched_class->migrate_task_rq(p, new_cpu);
1018
		p->se.nr_migrations++;
1019
		perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, NULL, 0);
1020 1021 1022 1023 1024 1025

		tmn.task = p;
		tmn.from_cpu = task_cpu(p);
		tmn.to_cpu = new_cpu;

		atomic_notifier_call_chain(&task_migration_notifier, 0, &tmn);
1026
	}
I
Ingo Molnar 已提交
1027 1028

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1029 1030
}

1031
struct migration_arg {
1032
	struct task_struct *task;
L
Linus Torvalds 已提交
1033
	int dest_cpu;
1034
};
L
Linus Torvalds 已提交
1035

1036 1037
static int migration_cpu_stop(void *data);

L
Linus Torvalds 已提交
1038 1039 1040
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1041 1042 1043 1044 1045 1046 1047
 * If @match_state is nonzero, it's the @p->state value just checked and
 * not expected to change.  If it changes, i.e. @p might have woken up,
 * then return zero.  When we succeed in waiting for @p to be off its CPU,
 * we return a positive number (its total switch count).  If a second call
 * a short while later returns the same number, the caller can be sure that
 * @p has remained unscheduled the whole time.
 *
L
Linus Torvalds 已提交
1048 1049 1050 1051 1052 1053
 * The caller must ensure that the task *will* unschedule sometime soon,
 * else this function might spin for a *long* time. This function can't
 * be called with interrupts off, or it may introduce deadlock with
 * smp_call_function() if an IPI is sent by the same process we are
 * waiting to become inactive.
 */
R
Roland McGrath 已提交
1054
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1055 1056
{
	unsigned long flags;
I
Ingo Molnar 已提交
1057
	int running, on_rq;
R
Roland McGrath 已提交
1058
	unsigned long ncsw;
1059
	struct rq *rq;
L
Linus Torvalds 已提交
1060

1061 1062 1063 1064 1065 1066 1067 1068
	for (;;) {
		/*
		 * We do the initial early heuristics without holding
		 * any task-queue locks at all. We'll only try to get
		 * the runqueue lock when things look like they will
		 * work out!
		 */
		rq = task_rq(p);
1069

1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
		/*
		 * If the task is actively running on another CPU
		 * still, just relax and busy-wait without holding
		 * any locks.
		 *
		 * NOTE! Since we don't hold any locks, it's not
		 * even sure that "rq" stays as the right runqueue!
		 * But we don't care, since "task_running()" will
		 * return false if the runqueue has changed and p
		 * is actually now running somewhere else!
		 */
R
Roland McGrath 已提交
1081 1082 1083
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1084
			cpu_relax();
R
Roland McGrath 已提交
1085
		}
1086

1087 1088 1089 1090 1091 1092
		/*
		 * Ok, time to look more closely! We need the rq
		 * lock now, to be *sure*. If we're wrong, we'll
		 * just go back and repeat.
		 */
		rq = task_rq_lock(p, &flags);
1093
		trace_sched_wait_task(p);
1094
		running = task_running(rq, p);
P
Peter Zijlstra 已提交
1095
		on_rq = p->on_rq;
R
Roland McGrath 已提交
1096
		ncsw = 0;
1097
		if (!match_state || p->state == match_state)
1098
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1099
		task_rq_unlock(rq, p, &flags);
1100

R
Roland McGrath 已提交
1101 1102 1103 1104 1105 1106
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1107 1108 1109 1110 1111 1112 1113 1114 1115 1116
		/*
		 * Was it really running after all now that we
		 * checked with the proper locks actually held?
		 *
		 * Oops. Go back and try again..
		 */
		if (unlikely(running)) {
			cpu_relax();
			continue;
		}
1117

1118 1119 1120 1121 1122
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1123
		 * So if it was still runnable (but just not actively
1124 1125 1126 1127
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
		if (unlikely(on_rq)) {
1128 1129 1130 1131
			ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1132 1133
			continue;
		}
1134

1135 1136 1137 1138 1139 1140 1141
		/*
		 * Ahh, all good. It wasn't running, and it wasn't
		 * runnable, which means that it will never become
		 * running in the future either. We're all done!
		 */
		break;
	}
R
Roland McGrath 已提交
1142 1143

	return ncsw;
L
Linus Torvalds 已提交
1144 1145 1146 1147 1148 1149 1150 1151 1152
}

/***
 * kick_process - kick a running thread to enter/exit the kernel
 * @p: the to-be-kicked thread
 *
 * Cause a process which is running on another CPU to enter
 * kernel-mode, without any delay. (to get signals handled.)
 *
L
Lucas De Marchi 已提交
1153
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1154 1155 1156 1157 1158
 * because all it wants to ensure is that the remote task enters
 * the kernel. If the IPI races and the task has been migrated
 * to another CPU then no harm is done and the purpose has been
 * achieved as well.
 */
1159
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1160 1161 1162 1163 1164 1165 1166 1167 1168
{
	int cpu;

	preempt_disable();
	cpu = task_cpu(p);
	if ((cpu != smp_processor_id()) && task_curr(p))
		smp_send_reschedule(cpu);
	preempt_enable();
}
R
Rusty Russell 已提交
1169
EXPORT_SYMBOL_GPL(kick_process);
N
Nick Piggin 已提交
1170
#endif /* CONFIG_SMP */
L
Linus Torvalds 已提交
1171

1172
#ifdef CONFIG_SMP
1173
/*
1174
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1175
 */
1176 1177
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1178 1179
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1180 1181
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1182

1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199
	/*
	 * If the node that the cpu is on has been offlined, cpu_to_node()
	 * will return -1. There is no cpu on the node, and we should
	 * select the cpu on the other node.
	 */
	if (nid != -1) {
		nodemask = cpumask_of_node(nid);

		/* Look for allowed, online CPU in same node. */
		for_each_cpu(dest_cpu, nodemask) {
			if (!cpu_online(dest_cpu))
				continue;
			if (!cpu_active(dest_cpu))
				continue;
			if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
				return dest_cpu;
		}
1200
	}
1201

1202 1203
	for (;;) {
		/* Any allowed, online CPU? */
1204
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1205 1206 1207 1208 1209 1210
			if (!cpu_online(dest_cpu))
				continue;
			if (!cpu_active(dest_cpu))
				continue;
			goto out;
		}
1211

1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240
		switch (state) {
		case cpuset:
			/* No more Mr. Nice Guy. */
			cpuset_cpus_allowed_fallback(p);
			state = possible;
			break;

		case possible:
			do_set_cpus_allowed(p, cpu_possible_mask);
			state = fail;
			break;

		case fail:
			BUG();
			break;
		}
	}

out:
	if (state != cpuset) {
		/*
		 * Don't tell them about moving exiting tasks or
		 * kernel threads (both mm NULL), since they never
		 * leave kernel.
		 */
		if (p->mm && printk_ratelimit()) {
			printk_sched("process %d (%s) no longer affine to cpu%d\n",
					task_pid_nr(p), p->comm, cpu);
		}
1241 1242 1243 1244 1245
	}

	return dest_cpu;
}

1246
/*
1247
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1248
 */
1249
static inline
1250
int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags)
1251
{
1252
	int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags);
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263

	/*
	 * In order not to call set_task_cpu() on a blocking task we need
	 * to rely on ttwu() to place the task on a valid ->cpus_allowed
	 * cpu.
	 *
	 * Since this is common to all placement strategies, this lives here.
	 *
	 * [ this allows ->select_task() to simply return task_cpu(p) and
	 *   not worry about this generic constraint ]
	 */
1264
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1265
		     !cpu_online(cpu)))
1266
		cpu = select_fallback_rq(task_cpu(p), p);
1267 1268

	return cpu;
1269
}
1270 1271 1272 1273 1274 1275

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1276 1277
#endif

P
Peter Zijlstra 已提交
1278
static void
1279
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1280
{
P
Peter Zijlstra 已提交
1281
#ifdef CONFIG_SCHEDSTATS
1282 1283
	struct rq *rq = this_rq();

P
Peter Zijlstra 已提交
1284 1285 1286 1287 1288 1289 1290 1291 1292 1293
#ifdef CONFIG_SMP
	int this_cpu = smp_processor_id();

	if (cpu == this_cpu) {
		schedstat_inc(rq, ttwu_local);
		schedstat_inc(p, se.statistics.nr_wakeups_local);
	} else {
		struct sched_domain *sd;

		schedstat_inc(p, se.statistics.nr_wakeups_remote);
1294
		rcu_read_lock();
P
Peter Zijlstra 已提交
1295 1296 1297 1298 1299 1300
		for_each_domain(this_cpu, sd) {
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
				schedstat_inc(sd, ttwu_wake_remote);
				break;
			}
		}
1301
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1302
	}
1303 1304 1305 1306

	if (wake_flags & WF_MIGRATED)
		schedstat_inc(p, se.statistics.nr_wakeups_migrate);

P
Peter Zijlstra 已提交
1307 1308 1309
#endif /* CONFIG_SMP */

	schedstat_inc(rq, ttwu_count);
T
Tejun Heo 已提交
1310
	schedstat_inc(p, se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1311 1312

	if (wake_flags & WF_SYNC)
T
Tejun Heo 已提交
1313
		schedstat_inc(p, se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1314 1315 1316 1317 1318 1319

#endif /* CONFIG_SCHEDSTATS */
}

static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
{
T
Tejun Heo 已提交
1320
	activate_task(rq, p, en_flags);
P
Peter Zijlstra 已提交
1321
	p->on_rq = 1;
1322 1323 1324 1325

	/* if a worker is waking up, notify workqueue */
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1326 1327
}

1328 1329 1330
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1331
static void
1332
ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
T
Tejun Heo 已提交
1333 1334
{
	check_preempt_curr(rq, p, wake_flags);
1335
	trace_sched_wakeup(p, true);
T
Tejun Heo 已提交
1336 1337 1338 1339 1340 1341

	p->state = TASK_RUNNING;
#ifdef CONFIG_SMP
	if (p->sched_class->task_woken)
		p->sched_class->task_woken(rq, p);

1342
	if (rq->idle_stamp) {
1343
		u64 delta = rq_clock(rq) - rq->idle_stamp;
T
Tejun Heo 已提交
1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354
		u64 max = 2*sysctl_sched_migration_cost;

		if (delta > max)
			rq->avg_idle = max;
		else
			update_avg(&rq->avg_idle, delta);
		rq->idle_stamp = 0;
	}
#endif
}

1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379
static void
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
{
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
#endif

	ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING);
	ttwu_do_wakeup(rq, p, wake_flags);
}

/*
 * Called in case the task @p isn't fully descheduled from its runqueue,
 * in this case we must do a remote wakeup. Its a 'light' wakeup though,
 * since all we need to do is flip p->state to TASK_RUNNING, since
 * the task is still ->on_rq.
 */
static int ttwu_remote(struct task_struct *p, int wake_flags)
{
	struct rq *rq;
	int ret = 0;

	rq = __task_rq_lock(p);
	if (p->on_rq) {
1380 1381
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1382 1383 1384 1385 1386 1387 1388 1389
		ttwu_do_wakeup(rq, p, wake_flags);
		ret = 1;
	}
	__task_rq_unlock(rq);

	return ret;
}

1390
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
1391
static void sched_ttwu_pending(void)
1392 1393
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1394 1395
	struct llist_node *llist = llist_del_all(&rq->wake_list);
	struct task_struct *p;
1396 1397 1398

	raw_spin_lock(&rq->lock);

P
Peter Zijlstra 已提交
1399 1400 1401
	while (llist) {
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
1402 1403 1404 1405 1406 1407 1408 1409
		ttwu_do_activate(rq, p, 0);
	}

	raw_spin_unlock(&rq->lock);
}

void scheduler_ipi(void)
{
1410 1411 1412
	if (llist_empty(&this_rq()->wake_list)
			&& !tick_nohz_full_cpu(smp_processor_id())
			&& !got_nohz_idle_kick())
1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428
		return;

	/*
	 * Not all reschedule IPI handlers call irq_enter/irq_exit, since
	 * traditionally all their work was done from the interrupt return
	 * path. Now that we actually do some work, we need to make sure
	 * we do call them.
	 *
	 * Some archs already do call them, luckily irq_enter/exit nest
	 * properly.
	 *
	 * Arguably we should visit all archs and update all handlers,
	 * however a fair share of IPIs are still resched only so this would
	 * somewhat pessimize the simple resched case.
	 */
	irq_enter();
1429
	tick_nohz_full_check();
P
Peter Zijlstra 已提交
1430
	sched_ttwu_pending();
1431 1432 1433 1434

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1435
	if (unlikely(got_nohz_idle_kick())) {
1436
		this_rq()->idle_balance = 1;
1437
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1438
	}
1439
	irq_exit();
1440 1441 1442 1443
}

static void ttwu_queue_remote(struct task_struct *p, int cpu)
{
P
Peter Zijlstra 已提交
1444
	if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list))
1445 1446
		smp_send_reschedule(cpu);
}
1447

1448
bool cpus_share_cache(int this_cpu, int that_cpu)
1449 1450 1451
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1452
#endif /* CONFIG_SMP */
1453

1454 1455 1456 1457
static void ttwu_queue(struct task_struct *p, int cpu)
{
	struct rq *rq = cpu_rq(cpu);

1458
#if defined(CONFIG_SMP)
1459
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1460
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
1461 1462 1463 1464 1465
		ttwu_queue_remote(p, cpu);
		return;
	}
#endif

1466 1467 1468
	raw_spin_lock(&rq->lock);
	ttwu_do_activate(rq, p, 0);
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
1469 1470 1471
}

/**
L
Linus Torvalds 已提交
1472
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1473
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1474
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1475
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1476 1477 1478 1479 1480 1481 1482
 *
 * Put it on the run-queue if it's not already there. The "current"
 * thread is always on the run-queue (except when the actual
 * re-schedule is in progress), and as such you're allowed to do
 * the simpler "current->state = TASK_RUNNING" to mark yourself
 * runnable without the overhead of this.
 *
T
Tejun Heo 已提交
1483 1484
 * Returns %true if @p was woken up, %false if it was already running
 * or @state didn't match @p's state.
L
Linus Torvalds 已提交
1485
 */
1486 1487
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1488 1489
{
	unsigned long flags;
1490
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1491

1492
	smp_wmb();
1493
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
1494
	if (!(p->state & state))
L
Linus Torvalds 已提交
1495 1496
		goto out;

1497
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
1498 1499
	cpu = task_cpu(p);

1500 1501
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
1502 1503

#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
1504
	/*
1505 1506
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
P
Peter Zijlstra 已提交
1507
	 */
1508
	while (p->on_cpu)
1509
		cpu_relax();
1510
	/*
1511
	 * Pairs with the smp_wmb() in finish_lock_switch().
1512
	 */
1513
	smp_rmb();
L
Linus Torvalds 已提交
1514

1515
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
1516
	p->state = TASK_WAKING;
1517

1518
	if (p->sched_class->task_waking)
1519
		p->sched_class->task_waking(p);
1520

1521
	cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags);
1522 1523
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
1524
		set_task_cpu(p, cpu);
1525
	}
L
Linus Torvalds 已提交
1526 1527
#endif /* CONFIG_SMP */

1528 1529
	ttwu_queue(p, cpu);
stat:
1530
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
1531
out:
1532
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
1533 1534 1535 1536

	return success;
}

T
Tejun Heo 已提交
1537 1538 1539 1540
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
 *
1541
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
1542
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
1543
 * the current task.
T
Tejun Heo 已提交
1544 1545 1546 1547 1548
 */
static void try_to_wake_up_local(struct task_struct *p)
{
	struct rq *rq = task_rq(p);

1549 1550 1551 1552
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
1553 1554
	lockdep_assert_held(&rq->lock);

1555 1556 1557 1558 1559 1560
	if (!raw_spin_trylock(&p->pi_lock)) {
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
	}

T
Tejun Heo 已提交
1561
	if (!(p->state & TASK_NORMAL))
1562
		goto out;
T
Tejun Heo 已提交
1563

P
Peter Zijlstra 已提交
1564
	if (!p->on_rq)
P
Peter Zijlstra 已提交
1565 1566
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

1567
	ttwu_do_wakeup(rq, p, 0);
1568
	ttwu_stat(p, smp_processor_id(), 0);
1569 1570
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
1571 1572
}

1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583
/**
 * wake_up_process - Wake up a specific process
 * @p: The process to be woken up.
 *
 * Attempt to wake up the nominated process and move it to the set of runnable
 * processes.  Returns 1 if the process was woken up, 0 if it was already
 * running.
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
 */
1584
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
1585
{
1586 1587
	WARN_ON(task_is_stopped_or_traced(p));
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
1588 1589 1590
}
EXPORT_SYMBOL(wake_up_process);

1591
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
1592 1593 1594 1595 1596 1597 1598
{
	return try_to_wake_up(p, state, 0);
}

/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
1599 1600 1601 1602 1603
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
static void __sched_fork(struct task_struct *p)
{
P
Peter Zijlstra 已提交
1604 1605 1606
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
1607 1608
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
1609
	p->se.prev_sum_exec_runtime	= 0;
1610
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
1611
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
1612
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
1613 1614

#ifdef CONFIG_SCHEDSTATS
1615
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
1616
#endif
N
Nick Piggin 已提交
1617

P
Peter Zijlstra 已提交
1618
	INIT_LIST_HEAD(&p->rt.run_list);
N
Nick Piggin 已提交
1619

1620 1621 1622
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
1623 1624 1625 1626

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
		p->mm->numa_next_scan = jiffies;
1627
		p->mm->numa_next_reset = jiffies;
1628 1629 1630 1631 1632 1633
		p->mm->numa_scan_seq = 0;
	}

	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
	p->numa_migrate_seq = p->mm ? p->mm->numa_scan_seq - 1 : 0;
1634
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
1635 1636
	p->numa_work.next = &p->numa_work;
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
1637 1638
}

1639
#ifdef CONFIG_NUMA_BALANCING
1640
#ifdef CONFIG_SCHED_DEBUG
1641 1642 1643 1644 1645 1646 1647
void set_numabalancing_state(bool enabled)
{
	if (enabled)
		sched_feat_set("NUMA");
	else
		sched_feat_set("NO_NUMA");
}
1648 1649 1650 1651 1652 1653
#else
__read_mostly bool numabalancing_enabled;

void set_numabalancing_state(bool enabled)
{
	numabalancing_enabled = enabled;
I
Ingo Molnar 已提交
1654
}
1655
#endif /* CONFIG_SCHED_DEBUG */
1656
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
1657 1658 1659 1660

/*
 * fork()/clone()-time setup:
 */
1661
void sched_fork(struct task_struct *p)
I
Ingo Molnar 已提交
1662
{
1663
	unsigned long flags;
I
Ingo Molnar 已提交
1664 1665 1666
	int cpu = get_cpu();

	__sched_fork(p);
1667
	/*
1668
	 * We mark the process as running here. This guarantees that
1669 1670 1671
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
1672
	p->state = TASK_RUNNING;
I
Ingo Molnar 已提交
1673

1674 1675 1676 1677 1678
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

1679 1680 1681 1682
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
1683
		if (task_has_rt_policy(p)) {
1684
			p->policy = SCHED_NORMAL;
1685
			p->static_prio = NICE_TO_PRIO(0);
1686 1687 1688 1689 1690 1691
			p->rt_priority = 0;
		} else if (PRIO_TO_NICE(p->static_prio) < 0)
			p->static_prio = NICE_TO_PRIO(0);

		p->prio = p->normal_prio = __normal_prio(p);
		set_load_weight(p);
1692

1693 1694 1695 1696 1697 1698
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
1699

H
Hiroshi Shimamoto 已提交
1700 1701
	if (!rt_prio(p->prio))
		p->sched_class = &fair_sched_class;
1702

P
Peter Zijlstra 已提交
1703 1704 1705
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);

1706 1707 1708 1709 1710 1711 1712
	/*
	 * The child is not yet in the pid-hash so no cgroup attach races,
	 * and the cgroup is pinned to this child due to cgroup_fork()
	 * is ran before sched_fork().
	 *
	 * Silence PROVE_RCU.
	 */
1713
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1714
	set_task_cpu(p, cpu);
1715
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
1716

1717
#if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT)
I
Ingo Molnar 已提交
1718
	if (likely(sched_info_on()))
1719
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
1720
#endif
P
Peter Zijlstra 已提交
1721 1722
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
1723
#endif
1724
#ifdef CONFIG_PREEMPT_COUNT
1725
	/* Want to start with kernel preemption disabled. */
A
Al Viro 已提交
1726
	task_thread_info(p)->preempt_count = 1;
L
Linus Torvalds 已提交
1727
#endif
1728
#ifdef CONFIG_SMP
1729
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
1730
#endif
1731

N
Nick Piggin 已提交
1732
	put_cpu();
L
Linus Torvalds 已提交
1733 1734 1735 1736 1737 1738 1739 1740 1741
}

/*
 * wake_up_new_task - wake up a newly created task for the first time.
 *
 * This function will do some initial scheduler statistics housekeeping
 * that must be done for every newly created context, then puts the task
 * on the runqueue and wakes it.
 */
1742
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
1743 1744
{
	unsigned long flags;
I
Ingo Molnar 已提交
1745
	struct rq *rq;
1746

1747
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1748 1749 1750 1751 1752 1753
#ifdef CONFIG_SMP
	/*
	 * Fork balancing, do it here and not earlier because:
	 *  - cpus_allowed can change in the fork path
	 *  - any previously selected cpu might disappear through hotplug
	 */
1754
	set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0));
1755 1756
#endif

1757 1758
	/* Initialize new task's runnable average */
	init_task_runnable_average(p);
1759
	rq = __task_rq_lock(p);
P
Peter Zijlstra 已提交
1760
	activate_task(rq, p, 0);
P
Peter Zijlstra 已提交
1761
	p->on_rq = 1;
1762
	trace_sched_wakeup_new(p, true);
P
Peter Zijlstra 已提交
1763
	check_preempt_curr(rq, p, WF_FORK);
1764
#ifdef CONFIG_SMP
1765 1766
	if (p->sched_class->task_woken)
		p->sched_class->task_woken(rq, p);
1767
#endif
1768
	task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
1769 1770
}

1771 1772 1773
#ifdef CONFIG_PREEMPT_NOTIFIERS

/**
1774
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
1775
 * @notifier: notifier struct to register
1776 1777 1778 1779 1780 1781 1782 1783 1784
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
	hlist_add_head(&notifier->link, &current->preempt_notifiers);
}
EXPORT_SYMBOL_GPL(preempt_notifier_register);

/**
 * preempt_notifier_unregister - no longer interested in preemption notifications
R
Randy Dunlap 已提交
1785
 * @notifier: notifier struct to unregister
1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798
 *
 * This is safe to call from within a preemption notifier.
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
	struct preempt_notifier *notifier;

1799
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
1800 1801 1802 1803 1804 1805 1806 1807 1808
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

static void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
	struct preempt_notifier *notifier;

1809
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
1810 1811 1812
		notifier->ops->sched_out(notifier, next);
}

1813
#else /* !CONFIG_PREEMPT_NOTIFIERS */
1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824

static void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
}

static void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

1825
#endif /* CONFIG_PREEMPT_NOTIFIERS */
1826

1827 1828 1829
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
1830
 * @prev: the current task that is being switched out
1831 1832 1833 1834 1835 1836 1837 1838 1839
 * @next: the task we are going to switch to.
 *
 * This is called with the rq lock held and interrupts off. It must
 * be paired with a subsequent finish_task_switch after the context
 * switch.
 *
 * prepare_task_switch sets up locking and calls architecture specific
 * hooks.
 */
1840 1841 1842
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
1843
{
1844
	trace_sched_switch(prev, next);
1845 1846
	sched_info_switch(prev, next);
	perf_event_task_sched_out(prev, next);
1847
	fire_sched_out_preempt_notifiers(prev, next);
1848 1849 1850 1851
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
1852 1853
/**
 * finish_task_switch - clean up after a task-switch
1854
 * @rq: runqueue associated with task-switch
L
Linus Torvalds 已提交
1855 1856
 * @prev: the thread we just switched away from.
 *
1857 1858 1859 1860
 * finish_task_switch must be called after the context switch, paired
 * with a prepare_task_switch call before the context switch.
 * finish_task_switch will reconcile locking set up by prepare_task_switch,
 * and do any other architecture-specific cleanup actions.
L
Linus Torvalds 已提交
1861 1862
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
1863
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
1864 1865 1866
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
 */
A
Alexey Dobriyan 已提交
1867
static void finish_task_switch(struct rq *rq, struct task_struct *prev)
L
Linus Torvalds 已提交
1868 1869 1870
	__releases(rq->lock)
{
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
1871
	long prev_state;
L
Linus Torvalds 已提交
1872 1873 1874 1875 1876

	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
1877
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
1878 1879
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
1880
	 * The test for TASK_DEAD must occur while the runqueue locks are
L
Linus Torvalds 已提交
1881 1882 1883 1884 1885
	 * still held, otherwise prev could be scheduled on another cpu, die
	 * there before we look at prev->state, and then the reference would
	 * be dropped twice.
	 *		Manfred Spraul <manfred@colorfullife.com>
	 */
O
Oleg Nesterov 已提交
1886
	prev_state = prev->state;
1887
	vtime_task_switch(prev);
1888
	finish_arch_switch(prev);
1889
	perf_event_task_sched_in(prev, current);
1890
	finish_lock_switch(rq, prev);
1891
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
1892

1893
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
1894 1895
	if (mm)
		mmdrop(mm);
1896
	if (unlikely(prev_state == TASK_DEAD)) {
1897 1898 1899
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
1900
		 */
1901
		kprobe_flush_task(prev);
L
Linus Torvalds 已提交
1902
		put_task_struct(prev);
1903
	}
1904 1905

	tick_nohz_task_switch(current);
L
Linus Torvalds 已提交
1906 1907
}

1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922
#ifdef CONFIG_SMP

/* assumes rq->lock is held */
static inline void pre_schedule(struct rq *rq, struct task_struct *prev)
{
	if (prev->sched_class->pre_schedule)
		prev->sched_class->pre_schedule(rq, prev);
}

/* rq->lock is NOT held, but preemption is disabled */
static inline void post_schedule(struct rq *rq)
{
	if (rq->post_schedule) {
		unsigned long flags;

1923
		raw_spin_lock_irqsave(&rq->lock, flags);
1924 1925
		if (rq->curr->sched_class->post_schedule)
			rq->curr->sched_class->post_schedule(rq);
1926
		raw_spin_unlock_irqrestore(&rq->lock, flags);
1927 1928 1929 1930 1931 1932

		rq->post_schedule = 0;
	}
}

#else
1933

1934 1935 1936 1937 1938 1939
static inline void pre_schedule(struct rq *rq, struct task_struct *p)
{
}

static inline void post_schedule(struct rq *rq)
{
L
Linus Torvalds 已提交
1940 1941
}

1942 1943
#endif

L
Linus Torvalds 已提交
1944 1945 1946 1947
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
1948
asmlinkage void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
1949 1950
	__releases(rq->lock)
{
1951 1952
	struct rq *rq = this_rq();

1953
	finish_task_switch(rq, prev);
1954

1955 1956 1957 1958 1959
	/*
	 * FIXME: do we need to worry about rq being invalidated by the
	 * task_switch?
	 */
	post_schedule(rq);
1960

1961 1962 1963 1964
#ifdef __ARCH_WANT_UNLOCKED_CTXSW
	/* In this case, finish_task_switch does not reenable preemption */
	preempt_enable();
#endif
L
Linus Torvalds 已提交
1965
	if (current->set_child_tid)
1966
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
1967 1968 1969 1970 1971 1972
}

/*
 * context_switch - switch to the new MM and the new
 * thread's register state.
 */
I
Ingo Molnar 已提交
1973
static inline void
1974
context_switch(struct rq *rq, struct task_struct *prev,
1975
	       struct task_struct *next)
L
Linus Torvalds 已提交
1976
{
I
Ingo Molnar 已提交
1977
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
1978

1979
	prepare_task_switch(rq, prev, next);
1980

I
Ingo Molnar 已提交
1981 1982
	mm = next->mm;
	oldmm = prev->active_mm;
1983 1984 1985 1986 1987
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
1988
	arch_start_context_switch(prev);
1989

1990
	if (!mm) {
L
Linus Torvalds 已提交
1991 1992 1993 1994 1995 1996
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
		switch_mm(oldmm, mm, next);

1997
	if (!prev->mm) {
L
Linus Torvalds 已提交
1998 1999 2000
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2001 2002 2003 2004 2005 2006 2007
	/*
	 * Since the runqueue lock will be released by the next
	 * task (which is an invalid locking op but in the case
	 * of the scheduler it's an obvious special-case), so we
	 * do an early lockdep release here:
	 */
#ifndef __ARCH_WANT_UNLOCKED_CTXSW
2008
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2009
#endif
L
Linus Torvalds 已提交
2010

2011
	context_tracking_task_switch(prev, next);
L
Linus Torvalds 已提交
2012 2013 2014
	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);

I
Ingo Molnar 已提交
2015 2016 2017 2018 2019 2020 2021
	barrier();
	/*
	 * this_rq must be evaluated again because prev may have moved
	 * CPUs since it called schedule(), thus the 'rq' on its stack
	 * frame will be invalid.
	 */
	finish_task_switch(this_rq(), prev);
L
Linus Torvalds 已提交
2022 2023 2024
}

/*
2025
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2026 2027
 *
 * externally visible scheduler statistics: current number of runnable
2028
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2029 2030 2031 2032 2033 2034 2035 2036 2037
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

	for_each_online_cpu(i)
		sum += cpu_rq(i)->nr_running;

	return sum;
2038
}
L
Linus Torvalds 已提交
2039 2040

unsigned long long nr_context_switches(void)
2041
{
2042 2043
	int i;
	unsigned long long sum = 0;
2044

2045
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2046
		sum += cpu_rq(i)->nr_switches;
2047

L
Linus Torvalds 已提交
2048 2049
	return sum;
}
2050

L
Linus Torvalds 已提交
2051 2052 2053
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2054

2055
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2056
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2057

L
Linus Torvalds 已提交
2058 2059
	return sum;
}
2060

2061
unsigned long nr_iowait_cpu(int cpu)
2062
{
2063
	struct rq *this = cpu_rq(cpu);
2064 2065
	return atomic_read(&this->nr_iowait);
}
2066

I
Ingo Molnar 已提交
2067
#ifdef CONFIG_SMP
2068

2069
/*
P
Peter Zijlstra 已提交
2070 2071
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2072
 */
P
Peter Zijlstra 已提交
2073
void sched_exec(void)
2074
{
P
Peter Zijlstra 已提交
2075
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2076
	unsigned long flags;
2077
	int dest_cpu;
2078

2079
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2080
	dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0);
2081 2082
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2083

2084
	if (likely(cpu_active(dest_cpu))) {
2085
		struct migration_arg arg = { p, dest_cpu };
2086

2087 2088
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2089 2090
		return;
	}
2091
unlock:
2092
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2093
}
I
Ingo Molnar 已提交
2094

L
Linus Torvalds 已提交
2095 2096 2097
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2098
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2099 2100

EXPORT_PER_CPU_SYMBOL(kstat);
2101
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2102 2103

/*
2104
 * Return any ns on the sched_clock that have not yet been accounted in
2105
 * @p in case that task is currently running.
2106 2107
 *
 * Called with task_rq_lock() held on @rq.
L
Linus Torvalds 已提交
2108
 */
2109 2110 2111 2112 2113 2114
static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
{
	u64 ns = 0;

	if (task_current(rq, p)) {
		update_rq_clock(rq);
2115
		ns = rq_clock_task(rq) - p->se.exec_start;
2116 2117 2118 2119 2120 2121 2122
		if ((s64)ns < 0)
			ns = 0;
	}

	return ns;
}

2123
unsigned long long task_delta_exec(struct task_struct *p)
L
Linus Torvalds 已提交
2124 2125
{
	unsigned long flags;
2126
	struct rq *rq;
2127
	u64 ns = 0;
2128

2129
	rq = task_rq_lock(p, &flags);
2130
	ns = do_task_delta_exec(p, rq);
2131
	task_rq_unlock(rq, p, &flags);
2132

2133 2134
	return ns;
}
2135

2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148
/*
 * Return accounted runtime for the task.
 * In case the task is currently running, return the runtime plus current's
 * pending runtime that have not been accounted yet.
 */
unsigned long long task_sched_runtime(struct task_struct *p)
{
	unsigned long flags;
	struct rq *rq;
	u64 ns = 0;

	rq = task_rq_lock(p, &flags);
	ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq);
2149
	task_rq_unlock(rq, p, &flags);
2150 2151 2152

	return ns;
}
2153

2154 2155 2156 2157 2158 2159 2160 2161
/*
 * This function gets called by the timer code, with HZ frequency.
 * We call it with interrupts disabled.
 */
void scheduler_tick(void)
{
	int cpu = smp_processor_id();
	struct rq *rq = cpu_rq(cpu);
I
Ingo Molnar 已提交
2162
	struct task_struct *curr = rq->curr;
2163 2164

	sched_clock_tick();
I
Ingo Molnar 已提交
2165

2166
	raw_spin_lock(&rq->lock);
2167
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
2168
	curr->sched_class->task_tick(rq, curr, 0);
2169
	update_cpu_load_active(rq);
2170
	raw_spin_unlock(&rq->lock);
2171

2172
	perf_event_task_tick();
2173

2174
#ifdef CONFIG_SMP
2175
	rq->idle_balance = idle_cpu(cpu);
I
Ingo Molnar 已提交
2176
	trigger_load_balance(rq, cpu);
2177
#endif
2178
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
2179 2180
}

2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
#ifdef CONFIG_NO_HZ_FULL
/**
 * scheduler_tick_max_deferment
 *
 * Keep at least one tick per second when a single
 * active task is running because the scheduler doesn't
 * yet completely support full dynticks environment.
 *
 * This makes sure that uptime, CFS vruntime, load
 * balancing, etc... continue to move forward, even
 * with a very low granularity.
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
	unsigned long next, now = ACCESS_ONCE(jiffies);

	next = rq->last_sched_tick + HZ;

	if (time_before_eq(next, now))
		return 0;

	return jiffies_to_usecs(next - now) * NSEC_PER_USEC;
L
Linus Torvalds 已提交
2204
}
2205
#endif
L
Linus Torvalds 已提交
2206

2207
notrace unsigned long get_parent_ip(unsigned long addr)
2208 2209 2210 2211 2212 2213 2214 2215
{
	if (in_lock_functions(addr)) {
		addr = CALLER_ADDR2;
		if (in_lock_functions(addr))
			addr = CALLER_ADDR3;
	}
	return addr;
}
L
Linus Torvalds 已提交
2216

2217 2218 2219
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))

2220
void __kprobes add_preempt_count(int val)
L
Linus Torvalds 已提交
2221
{
2222
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2223 2224 2225
	/*
	 * Underflow?
	 */
2226 2227
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
2228
#endif
L
Linus Torvalds 已提交
2229
	preempt_count() += val;
2230
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2231 2232 2233
	/*
	 * Spinlock count overflowing soon?
	 */
2234 2235
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
2236 2237 2238
#endif
	if (preempt_count() == val)
		trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
L
Linus Torvalds 已提交
2239 2240 2241
}
EXPORT_SYMBOL(add_preempt_count);

2242
void __kprobes sub_preempt_count(int val)
L
Linus Torvalds 已提交
2243
{
2244
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2245 2246 2247
	/*
	 * Underflow?
	 */
2248
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
2249
		return;
L
Linus Torvalds 已提交
2250 2251 2252
	/*
	 * Is the spinlock portion underflowing?
	 */
2253 2254 2255
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
2256
#endif
2257

2258 2259
	if (preempt_count() == val)
		trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
L
Linus Torvalds 已提交
2260 2261 2262 2263 2264 2265 2266
	preempt_count() -= val;
}
EXPORT_SYMBOL(sub_preempt_count);

#endif

/*
I
Ingo Molnar 已提交
2267
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
2268
 */
I
Ingo Molnar 已提交
2269
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
2270
{
2271 2272 2273
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
2274 2275
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
2276

I
Ingo Molnar 已提交
2277
	debug_show_held_locks(prev);
2278
	print_modules();
I
Ingo Molnar 已提交
2279 2280
	if (irqs_disabled())
		print_irqtrace_events(prev);
2281
	dump_stack();
2282
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
2283
}
L
Linus Torvalds 已提交
2284

I
Ingo Molnar 已提交
2285 2286 2287 2288 2289
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
L
Linus Torvalds 已提交
2290
	/*
I
Ingo Molnar 已提交
2291
	 * Test if we are atomic. Since do_exit() needs to call into
L
Linus Torvalds 已提交
2292 2293 2294
	 * schedule() atomically, we ignore that path for now.
	 * Otherwise, whine if we are scheduling when we should not be.
	 */
2295
	if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
I
Ingo Molnar 已提交
2296
		__schedule_bug(prev);
2297
	rcu_sleep_check();
I
Ingo Molnar 已提交
2298

L
Linus Torvalds 已提交
2299 2300
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

2301
	schedstat_inc(this_rq(), sched_count);
I
Ingo Molnar 已提交
2302 2303
}

P
Peter Zijlstra 已提交
2304
static void put_prev_task(struct rq *rq, struct task_struct *prev)
M
Mike Galbraith 已提交
2305
{
2306
	if (prev->on_rq || rq->skip_clock_update < 0)
2307
		update_rq_clock(rq);
P
Peter Zijlstra 已提交
2308
	prev->sched_class->put_prev_task(rq, prev);
M
Mike Galbraith 已提交
2309 2310
}

I
Ingo Molnar 已提交
2311 2312 2313 2314
/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
2315
pick_next_task(struct rq *rq)
I
Ingo Molnar 已提交
2316
{
2317
	const struct sched_class *class;
I
Ingo Molnar 已提交
2318
	struct task_struct *p;
L
Linus Torvalds 已提交
2319 2320

	/*
I
Ingo Molnar 已提交
2321 2322
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
2323
	 */
2324
	if (likely(rq->nr_running == rq->cfs.h_nr_running)) {
2325
		p = fair_sched_class.pick_next_task(rq);
I
Ingo Molnar 已提交
2326 2327
		if (likely(p))
			return p;
L
Linus Torvalds 已提交
2328 2329
	}

2330
	for_each_class(class) {
2331
		p = class->pick_next_task(rq);
I
Ingo Molnar 已提交
2332 2333 2334
		if (p)
			return p;
	}
2335 2336

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
2337
}
L
Linus Torvalds 已提交
2338

I
Ingo Molnar 已提交
2339
/*
2340
 * __schedule() is the main scheduler function.
2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374
 *
 * The main means of driving the scheduler and thus entering this function are:
 *
 *   1. Explicit blocking: mutex, semaphore, waitqueue, etc.
 *
 *   2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return
 *      paths. For example, see arch/x86/entry_64.S.
 *
 *      To drive preemption between tasks, the scheduler sets the flag in timer
 *      interrupt handler scheduler_tick().
 *
 *   3. Wakeups don't really cause entry into schedule(). They add a
 *      task to the run-queue and that's it.
 *
 *      Now, if the new task added to the run-queue preempts the current
 *      task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets
 *      called on the nearest possible occasion:
 *
 *       - If the kernel is preemptible (CONFIG_PREEMPT=y):
 *
 *         - in syscall or exception context, at the next outmost
 *           preempt_enable(). (this might be as soon as the wake_up()'s
 *           spin_unlock()!)
 *
 *         - in IRQ context, return from interrupt-handler to
 *           preemptible context
 *
 *       - If the kernel is not preemptible (CONFIG_PREEMPT is not set)
 *         then at the next:
 *
 *          - cond_resched() call
 *          - explicit schedule() call
 *          - return from syscall or exception to user-space
 *          - return from interrupt-handler to user-space
I
Ingo Molnar 已提交
2375
 */
2376
static void __sched __schedule(void)
I
Ingo Molnar 已提交
2377 2378
{
	struct task_struct *prev, *next;
2379
	unsigned long *switch_count;
I
Ingo Molnar 已提交
2380
	struct rq *rq;
2381
	int cpu;
I
Ingo Molnar 已提交
2382

2383 2384
need_resched:
	preempt_disable();
I
Ingo Molnar 已提交
2385 2386
	cpu = smp_processor_id();
	rq = cpu_rq(cpu);
2387
	rcu_note_context_switch(cpu);
I
Ingo Molnar 已提交
2388 2389 2390
	prev = rq->curr;

	schedule_debug(prev);
L
Linus Torvalds 已提交
2391

2392
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
2393
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
2394

2395
	raw_spin_lock_irq(&rq->lock);
L
Linus Torvalds 已提交
2396

2397
	switch_count = &prev->nivcsw;
L
Linus Torvalds 已提交
2398
	if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
T
Tejun Heo 已提交
2399
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
2400
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
2401
		} else {
2402 2403 2404
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
2405
			/*
2406 2407 2408
			 * If a worker went to sleep, notify and ask workqueue
			 * whether it wants to wake up a task to maintain
			 * concurrency.
T
Tejun Heo 已提交
2409 2410 2411 2412 2413 2414 2415 2416 2417
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

				to_wakeup = wq_worker_sleeping(prev, cpu);
				if (to_wakeup)
					try_to_wake_up_local(to_wakeup);
			}
		}
I
Ingo Molnar 已提交
2418
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
2419 2420
	}

2421
	pre_schedule(rq, prev);
2422

I
Ingo Molnar 已提交
2423
	if (unlikely(!rq->nr_running))
L
Linus Torvalds 已提交
2424 2425
		idle_balance(cpu, rq);

M
Mike Galbraith 已提交
2426
	put_prev_task(rq, prev);
2427
	next = pick_next_task(rq);
2428 2429
	clear_tsk_need_resched(prev);
	rq->skip_clock_update = 0;
L
Linus Torvalds 已提交
2430 2431 2432 2433 2434 2435

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
		++*switch_count;

I
Ingo Molnar 已提交
2436
		context_switch(rq, prev, next); /* unlocks the rq */
P
Peter Zijlstra 已提交
2437
		/*
2438 2439 2440 2441
		 * The context switch have flipped the stack from under us
		 * and restored the local variables which were saved when
		 * this task called schedule() in the past. prev == current
		 * is still correct, but it can be moved to another cpu/rq.
P
Peter Zijlstra 已提交
2442 2443 2444
		 */
		cpu = smp_processor_id();
		rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
2445
	} else
2446
		raw_spin_unlock_irq(&rq->lock);
L
Linus Torvalds 已提交
2447

2448
	post_schedule(rq);
L
Linus Torvalds 已提交
2449

2450
	sched_preempt_enable_no_resched();
2451
	if (need_resched())
L
Linus Torvalds 已提交
2452 2453
		goto need_resched;
}
2454

2455 2456
static inline void sched_submit_work(struct task_struct *tsk)
{
2457
	if (!tsk->state || tsk_is_pi_blocked(tsk))
2458 2459 2460 2461 2462 2463 2464 2465 2466
		return;
	/*
	 * If we are going to sleep and we have plugged IO queued,
	 * make sure to submit it to avoid deadlocks.
	 */
	if (blk_needs_flush_plug(tsk))
		blk_schedule_flush_plug(tsk);
}

S
Simon Kirby 已提交
2467
asmlinkage void __sched schedule(void)
2468
{
2469 2470 2471
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
2472 2473
	__schedule();
}
L
Linus Torvalds 已提交
2474 2475
EXPORT_SYMBOL(schedule);

2476
#ifdef CONFIG_CONTEXT_TRACKING
2477 2478 2479 2480 2481 2482 2483 2484
asmlinkage void __sched schedule_user(void)
{
	/*
	 * If we come here after a random call to set_need_resched(),
	 * or we have been woken up remotely but the IPI has not yet arrived,
	 * we haven't yet exited the RCU idle mode. Do it here manually until
	 * we find a better solution.
	 */
2485
	user_exit();
2486
	schedule();
2487
	user_enter();
2488 2489 2490
}
#endif

2491 2492 2493 2494 2495 2496 2497
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
2498
	sched_preempt_enable_no_resched();
2499 2500 2501 2502
	schedule();
	preempt_disable();
}

L
Linus Torvalds 已提交
2503 2504
#ifdef CONFIG_PREEMPT
/*
2505
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
2506
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
2507 2508
 * occur there and call schedule directly.
 */
2509
asmlinkage void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
2510 2511
{
	struct thread_info *ti = current_thread_info();
2512

L
Linus Torvalds 已提交
2513 2514
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
2515
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
2516
	 */
N
Nick Piggin 已提交
2517
	if (likely(ti->preempt_count || irqs_disabled()))
L
Linus Torvalds 已提交
2518 2519
		return;

2520
	do {
2521
		add_preempt_count_notrace(PREEMPT_ACTIVE);
2522
		__schedule();
2523
		sub_preempt_count_notrace(PREEMPT_ACTIVE);
L
Linus Torvalds 已提交
2524

2525 2526 2527 2528 2529
		/*
		 * Check again in case we missed a preemption opportunity
		 * between schedule and now.
		 */
		barrier();
2530
	} while (need_resched());
L
Linus Torvalds 已提交
2531 2532 2533 2534
}
EXPORT_SYMBOL(preempt_schedule);

/*
2535
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
2536 2537 2538 2539 2540 2541 2542
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
asmlinkage void __sched preempt_schedule_irq(void)
{
	struct thread_info *ti = current_thread_info();
2543
	enum ctx_state prev_state;
2544

2545
	/* Catch callers which need to be fixed */
L
Linus Torvalds 已提交
2546 2547
	BUG_ON(ti->preempt_count || !irqs_disabled());

2548 2549
	prev_state = exception_enter();

2550 2551 2552
	do {
		add_preempt_count(PREEMPT_ACTIVE);
		local_irq_enable();
2553
		__schedule();
2554 2555
		local_irq_disable();
		sub_preempt_count(PREEMPT_ACTIVE);
L
Linus Torvalds 已提交
2556

2557 2558 2559 2560 2561
		/*
		 * Check again in case we missed a preemption opportunity
		 * between schedule and now.
		 */
		barrier();
2562
	} while (need_resched());
2563 2564

	exception_exit(prev_state);
L
Linus Torvalds 已提交
2565 2566 2567 2568
}

#endif /* CONFIG_PREEMPT */

P
Peter Zijlstra 已提交
2569
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
2570
			  void *key)
L
Linus Torvalds 已提交
2571
{
P
Peter Zijlstra 已提交
2572
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
2573 2574 2575 2576
}
EXPORT_SYMBOL(default_wake_function);

/*
I
Ingo Molnar 已提交
2577 2578
 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
L
Linus Torvalds 已提交
2579 2580 2581
 * number) then we wake all the non-exclusive tasks and one exclusive task.
 *
 * There are circumstances in which we can try to wake a task which has already
I
Ingo Molnar 已提交
2582
 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
L
Linus Torvalds 已提交
2583 2584
 * zero in this (rare) case, and we handle it by continuing to scan the queue.
 */
2585
static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
P
Peter Zijlstra 已提交
2586
			int nr_exclusive, int wake_flags, void *key)
L
Linus Torvalds 已提交
2587
{
2588
	wait_queue_t *curr, *next;
L
Linus Torvalds 已提交
2589

2590
	list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
2591 2592
		unsigned flags = curr->flags;

P
Peter Zijlstra 已提交
2593
		if (curr->func(curr, mode, wake_flags, key) &&
2594
				(flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
L
Linus Torvalds 已提交
2595 2596 2597 2598 2599 2600 2601 2602 2603
			break;
	}
}

/**
 * __wake_up - wake up threads blocked on a waitqueue.
 * @q: the waitqueue
 * @mode: which threads
 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2604
 * @key: is directly passed to the wakeup function
2605 2606 2607
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
L
Linus Torvalds 已提交
2608
 */
2609
void __wake_up(wait_queue_head_t *q, unsigned int mode,
I
Ingo Molnar 已提交
2610
			int nr_exclusive, void *key)
L
Linus Torvalds 已提交
2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622
{
	unsigned long flags;

	spin_lock_irqsave(&q->lock, flags);
	__wake_up_common(q, mode, nr_exclusive, 0, key);
	spin_unlock_irqrestore(&q->lock, flags);
}
EXPORT_SYMBOL(__wake_up);

/*
 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
 */
2623
void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr)
L
Linus Torvalds 已提交
2624
{
2625
	__wake_up_common(q, mode, nr, 0, NULL);
L
Linus Torvalds 已提交
2626
}
2627
EXPORT_SYMBOL_GPL(__wake_up_locked);
L
Linus Torvalds 已提交
2628

2629 2630 2631 2632
void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
{
	__wake_up_common(q, mode, 1, 0, key);
}
2633
EXPORT_SYMBOL_GPL(__wake_up_locked_key);
2634

L
Linus Torvalds 已提交
2635
/**
2636
 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
L
Linus Torvalds 已提交
2637 2638 2639
 * @q: the waitqueue
 * @mode: which threads
 * @nr_exclusive: how many wake-one or wake-many threads to wake up
2640
 * @key: opaque value to be passed to wakeup targets
L
Linus Torvalds 已提交
2641 2642 2643 2644 2645 2646 2647
 *
 * The sync wakeup differs that the waker knows that it will schedule
 * away soon, so while the target thread will be woken up, it will not
 * be migrated to another CPU - ie. the two threads are 'synchronized'
 * with each other. This can prevent needless bouncing between CPUs.
 *
 * On UP it can prevent extra preemption.
2648 2649 2650
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
L
Linus Torvalds 已提交
2651
 */
2652 2653
void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
			int nr_exclusive, void *key)
L
Linus Torvalds 已提交
2654 2655
{
	unsigned long flags;
P
Peter Zijlstra 已提交
2656
	int wake_flags = WF_SYNC;
L
Linus Torvalds 已提交
2657 2658 2659 2660 2661

	if (unlikely(!q))
		return;

	if (unlikely(!nr_exclusive))
P
Peter Zijlstra 已提交
2662
		wake_flags = 0;
L
Linus Torvalds 已提交
2663 2664

	spin_lock_irqsave(&q->lock, flags);
P
Peter Zijlstra 已提交
2665
	__wake_up_common(q, mode, nr_exclusive, wake_flags, key);
L
Linus Torvalds 已提交
2666 2667
	spin_unlock_irqrestore(&q->lock, flags);
}
2668 2669 2670 2671 2672 2673 2674 2675 2676
EXPORT_SYMBOL_GPL(__wake_up_sync_key);

/*
 * __wake_up_sync - see __wake_up_sync_key()
 */
void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
{
	__wake_up_sync_key(q, mode, nr_exclusive, NULL);
}
L
Linus Torvalds 已提交
2677 2678
EXPORT_SYMBOL_GPL(__wake_up_sync);	/* For internal use only */

2679 2680 2681 2682 2683 2684 2685 2686
/**
 * complete: - signals a single thread waiting on this completion
 * @x:  holds the state of this particular completion
 *
 * This will wake up a single thread waiting on this completion. Threads will be
 * awakened in the same order in which they were queued.
 *
 * See also complete_all(), wait_for_completion() and related routines.
2687 2688 2689
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
2690
 */
2691
void complete(struct completion *x)
L
Linus Torvalds 已提交
2692 2693 2694 2695 2696
{
	unsigned long flags;

	spin_lock_irqsave(&x->wait.lock, flags);
	x->done++;
2697
	__wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL);
L
Linus Torvalds 已提交
2698 2699 2700 2701
	spin_unlock_irqrestore(&x->wait.lock, flags);
}
EXPORT_SYMBOL(complete);

2702 2703 2704 2705 2706
/**
 * complete_all: - signals all threads waiting on this completion
 * @x:  holds the state of this particular completion
 *
 * This will wake up all threads waiting on this particular completion event.
2707 2708 2709
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
2710
 */
2711
void complete_all(struct completion *x)
L
Linus Torvalds 已提交
2712 2713 2714 2715 2716
{
	unsigned long flags;

	spin_lock_irqsave(&x->wait.lock, flags);
	x->done += UINT_MAX/2;
2717
	__wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL);
L
Linus Torvalds 已提交
2718 2719 2720 2721
	spin_unlock_irqrestore(&x->wait.lock, flags);
}
EXPORT_SYMBOL(complete_all);

2722
static inline long __sched
2723 2724
do_wait_for_common(struct completion *x,
		   long (*action)(long), long timeout, int state)
L
Linus Torvalds 已提交
2725 2726 2727 2728
{
	if (!x->done) {
		DECLARE_WAITQUEUE(wait, current);

C
Changli Gao 已提交
2729
		__add_wait_queue_tail_exclusive(&x->wait, &wait);
L
Linus Torvalds 已提交
2730
		do {
2731
			if (signal_pending_state(state, current)) {
2732 2733
				timeout = -ERESTARTSYS;
				break;
2734 2735
			}
			__set_current_state(state);
L
Linus Torvalds 已提交
2736
			spin_unlock_irq(&x->wait.lock);
2737
			timeout = action(timeout);
L
Linus Torvalds 已提交
2738
			spin_lock_irq(&x->wait.lock);
2739
		} while (!x->done && timeout);
L
Linus Torvalds 已提交
2740
		__remove_wait_queue(&x->wait, &wait);
2741 2742
		if (!x->done)
			return timeout;
L
Linus Torvalds 已提交
2743 2744
	}
	x->done--;
2745
	return timeout ?: 1;
L
Linus Torvalds 已提交
2746 2747
}

2748 2749 2750
static inline long __sched
__wait_for_common(struct completion *x,
		  long (*action)(long), long timeout, int state)
L
Linus Torvalds 已提交
2751 2752 2753 2754
{
	might_sleep();

	spin_lock_irq(&x->wait.lock);
2755
	timeout = do_wait_for_common(x, action, timeout, state);
L
Linus Torvalds 已提交
2756
	spin_unlock_irq(&x->wait.lock);
2757 2758
	return timeout;
}
L
Linus Torvalds 已提交
2759

2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771
static long __sched
wait_for_common(struct completion *x, long timeout, int state)
{
	return __wait_for_common(x, schedule_timeout, timeout, state);
}

static long __sched
wait_for_common_io(struct completion *x, long timeout, int state)
{
	return __wait_for_common(x, io_schedule_timeout, timeout, state);
}

2772 2773 2774 2775 2776 2777 2778 2779 2780 2781
/**
 * wait_for_completion: - waits for completion of a task
 * @x:  holds the state of this particular completion
 *
 * This waits to be signaled for completion of a specific task. It is NOT
 * interruptible and there is no timeout.
 *
 * See also similar routines (i.e. wait_for_completion_timeout()) with timeout
 * and interrupt capability. Also see complete().
 */
2782
void __sched wait_for_completion(struct completion *x)
2783 2784
{
	wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
L
Linus Torvalds 已提交
2785
}
2786
EXPORT_SYMBOL(wait_for_completion);
L
Linus Torvalds 已提交
2787

2788 2789 2790 2791 2792 2793 2794 2795
/**
 * wait_for_completion_timeout: - waits for completion of a task (w/timeout)
 * @x:  holds the state of this particular completion
 * @timeout:  timeout value in jiffies
 *
 * This waits for either a completion of a specific task to be signaled or for a
 * specified timeout to expire. The timeout is in jiffies. It is not
 * interruptible.
2796 2797 2798
 *
 * The return value is 0 if timed out, and positive (at least 1, or number of
 * jiffies left till timeout) if completed.
2799
 */
2800
unsigned long __sched
2801
wait_for_completion_timeout(struct completion *x, unsigned long timeout)
L
Linus Torvalds 已提交
2802
{
2803
	return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE);
L
Linus Torvalds 已提交
2804
}
2805
EXPORT_SYMBOL(wait_for_completion_timeout);
L
Linus Torvalds 已提交
2806

2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839
/**
 * wait_for_completion_io: - waits for completion of a task
 * @x:  holds the state of this particular completion
 *
 * This waits to be signaled for completion of a specific task. It is NOT
 * interruptible and there is no timeout. The caller is accounted as waiting
 * for IO.
 */
void __sched wait_for_completion_io(struct completion *x)
{
	wait_for_common_io(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE);
}
EXPORT_SYMBOL(wait_for_completion_io);

/**
 * wait_for_completion_io_timeout: - waits for completion of a task (w/timeout)
 * @x:  holds the state of this particular completion
 * @timeout:  timeout value in jiffies
 *
 * This waits for either a completion of a specific task to be signaled or for a
 * specified timeout to expire. The timeout is in jiffies. It is not
 * interruptible. The caller is accounted as waiting for IO.
 *
 * The return value is 0 if timed out, and positive (at least 1, or number of
 * jiffies left till timeout) if completed.
 */
unsigned long __sched
wait_for_completion_io_timeout(struct completion *x, unsigned long timeout)
{
	return wait_for_common_io(x, timeout, TASK_UNINTERRUPTIBLE);
}
EXPORT_SYMBOL(wait_for_completion_io_timeout);

2840 2841 2842 2843 2844 2845
/**
 * wait_for_completion_interruptible: - waits for completion of a task (w/intr)
 * @x:  holds the state of this particular completion
 *
 * This waits for completion of a specific task to be signaled. It is
 * interruptible.
2846 2847
 *
 * The return value is -ERESTARTSYS if interrupted, 0 if completed.
2848
 */
2849
int __sched wait_for_completion_interruptible(struct completion *x)
I
Ingo Molnar 已提交
2850
{
2851 2852 2853 2854
	long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE);
	if (t == -ERESTARTSYS)
		return t;
	return 0;
I
Ingo Molnar 已提交
2855
}
2856
EXPORT_SYMBOL(wait_for_completion_interruptible);
L
Linus Torvalds 已提交
2857

2858 2859 2860 2861 2862 2863 2864
/**
 * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr))
 * @x:  holds the state of this particular completion
 * @timeout:  timeout value in jiffies
 *
 * This waits for either a completion of a specific task to be signaled or for a
 * specified timeout to expire. It is interruptible. The timeout is in jiffies.
2865 2866 2867
 *
 * The return value is -ERESTARTSYS if interrupted, 0 if timed out,
 * positive (at least 1, or number of jiffies left till timeout) if completed.
2868
 */
2869
long __sched
2870 2871
wait_for_completion_interruptible_timeout(struct completion *x,
					  unsigned long timeout)
I
Ingo Molnar 已提交
2872
{
2873
	return wait_for_common(x, timeout, TASK_INTERRUPTIBLE);
I
Ingo Molnar 已提交
2874
}
2875
EXPORT_SYMBOL(wait_for_completion_interruptible_timeout);
L
Linus Torvalds 已提交
2876

2877 2878 2879 2880 2881 2882
/**
 * wait_for_completion_killable: - waits for completion of a task (killable)
 * @x:  holds the state of this particular completion
 *
 * This waits to be signaled for completion of a specific task. It can be
 * interrupted by a kill signal.
2883 2884
 *
 * The return value is -ERESTARTSYS if interrupted, 0 if completed.
2885
 */
M
Matthew Wilcox 已提交
2886 2887 2888 2889 2890 2891 2892 2893 2894
int __sched wait_for_completion_killable(struct completion *x)
{
	long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE);
	if (t == -ERESTARTSYS)
		return t;
	return 0;
}
EXPORT_SYMBOL(wait_for_completion_killable);

2895 2896 2897 2898 2899 2900 2901 2902
/**
 * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable))
 * @x:  holds the state of this particular completion
 * @timeout:  timeout value in jiffies
 *
 * This waits for either a completion of a specific task to be
 * signaled or for a specified timeout to expire. It can be
 * interrupted by a kill signal. The timeout is in jiffies.
2903 2904 2905
 *
 * The return value is -ERESTARTSYS if interrupted, 0 if timed out,
 * positive (at least 1, or number of jiffies left till timeout) if completed.
2906
 */
2907
long __sched
2908 2909 2910 2911 2912 2913 2914
wait_for_completion_killable_timeout(struct completion *x,
				     unsigned long timeout)
{
	return wait_for_common(x, timeout, TASK_KILLABLE);
}
EXPORT_SYMBOL(wait_for_completion_killable_timeout);

2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928
/**
 *	try_wait_for_completion - try to decrement a completion without blocking
 *	@x:	completion structure
 *
 *	Returns: 0 if a decrement cannot be done without blocking
 *		 1 if a decrement succeeded.
 *
 *	If a completion is being used as a counting completion,
 *	attempt to decrement the counter without blocking. This
 *	enables us to avoid waiting if the resource the completion
 *	is protecting is not available.
 */
bool try_wait_for_completion(struct completion *x)
{
2929
	unsigned long flags;
2930 2931
	int ret = 1;

2932
	spin_lock_irqsave(&x->wait.lock, flags);
2933 2934 2935 2936
	if (!x->done)
		ret = 0;
	else
		x->done--;
2937
	spin_unlock_irqrestore(&x->wait.lock, flags);
2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951
	return ret;
}
EXPORT_SYMBOL(try_wait_for_completion);

/**
 *	completion_done - Test to see if a completion has any waiters
 *	@x:	completion structure
 *
 *	Returns: 0 if there are waiters (wait_for_completion() in progress)
 *		 1 if there are no waiters.
 *
 */
bool completion_done(struct completion *x)
{
2952
	unsigned long flags;
2953 2954
	int ret = 1;

2955
	spin_lock_irqsave(&x->wait.lock, flags);
2956 2957
	if (!x->done)
		ret = 0;
2958
	spin_unlock_irqrestore(&x->wait.lock, flags);
2959 2960 2961 2962
	return ret;
}
EXPORT_SYMBOL(completion_done);

2963 2964
static long __sched
sleep_on_common(wait_queue_head_t *q, int state, long timeout)
L
Linus Torvalds 已提交
2965
{
I
Ingo Molnar 已提交
2966 2967 2968 2969
	unsigned long flags;
	wait_queue_t wait;

	init_waitqueue_entry(&wait, current);
L
Linus Torvalds 已提交
2970

2971
	__set_current_state(state);
L
Linus Torvalds 已提交
2972

2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986
	spin_lock_irqsave(&q->lock, flags);
	__add_wait_queue(q, &wait);
	spin_unlock(&q->lock);
	timeout = schedule_timeout(timeout);
	spin_lock_irq(&q->lock);
	__remove_wait_queue(q, &wait);
	spin_unlock_irqrestore(&q->lock, flags);

	return timeout;
}

void __sched interruptible_sleep_on(wait_queue_head_t *q)
{
	sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
L
Linus Torvalds 已提交
2987 2988 2989
}
EXPORT_SYMBOL(interruptible_sleep_on);

I
Ingo Molnar 已提交
2990
long __sched
I
Ingo Molnar 已提交
2991
interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout)
L
Linus Torvalds 已提交
2992
{
2993
	return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout);
L
Linus Torvalds 已提交
2994 2995 2996
}
EXPORT_SYMBOL(interruptible_sleep_on_timeout);

I
Ingo Molnar 已提交
2997
void __sched sleep_on(wait_queue_head_t *q)
L
Linus Torvalds 已提交
2998
{
2999
	sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
L
Linus Torvalds 已提交
3000 3001 3002
}
EXPORT_SYMBOL(sleep_on);

I
Ingo Molnar 已提交
3003
long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout)
L
Linus Torvalds 已提交
3004
{
3005
	return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout);
L
Linus Torvalds 已提交
3006 3007 3008
}
EXPORT_SYMBOL(sleep_on_timeout);

3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
#ifdef CONFIG_RT_MUTEXES

/*
 * rt_mutex_setprio - set the current priority of a task
 * @p: task
 * @prio: prio value (kernel-internal form)
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
 * Used by the rt_mutex code to implement priority inheritance logic.
 */
3021
void rt_mutex_setprio(struct task_struct *p, int prio)
3022
{
3023
	int oldprio, on_rq, running;
3024
	struct rq *rq;
3025
	const struct sched_class *prev_class;
3026 3027 3028

	BUG_ON(prio < 0 || prio > MAX_PRIO);

3029
	rq = __task_rq_lock(p);
3030

3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048
	/*
	 * Idle task boosting is a nono in general. There is one
	 * exception, when PREEMPT_RT and NOHZ is active:
	 *
	 * The idle task calls get_next_timer_interrupt() and holds
	 * the timer wheel base->lock on the CPU and another CPU wants
	 * to access the timer (probably to cancel it). We can safely
	 * ignore the boosting request, as the idle CPU runs this code
	 * with interrupts disabled and will complete the lock
	 * protected section without being interrupted. So there is no
	 * real need to boost.
	 */
	if (unlikely(p == rq->idle)) {
		WARN_ON(p != rq->curr);
		WARN_ON(p->pi_blocked_on);
		goto out_unlock;
	}

3049
	trace_sched_pi_setprio(p, prio);
3050
	oldprio = p->prio;
3051
	prev_class = p->sched_class;
P
Peter Zijlstra 已提交
3052
	on_rq = p->on_rq;
3053
	running = task_current(rq, p);
3054
	if (on_rq)
3055
		dequeue_task(rq, p, 0);
3056 3057
	if (running)
		p->sched_class->put_prev_task(rq, p);
I
Ingo Molnar 已提交
3058 3059 3060 3061 3062 3063

	if (rt_prio(prio))
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;

3064 3065
	p->prio = prio;

3066 3067
	if (running)
		p->sched_class->set_curr_task(rq);
P
Peter Zijlstra 已提交
3068
	if (on_rq)
3069
		enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0);
3070

P
Peter Zijlstra 已提交
3071
	check_class_changed(rq, p, prev_class, oldprio);
3072
out_unlock:
3073
	__task_rq_unlock(rq);
3074 3075
}
#endif
3076
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3077
{
I
Ingo Molnar 已提交
3078
	int old_prio, delta, on_rq;
L
Linus Torvalds 已提交
3079
	unsigned long flags;
3080
	struct rq *rq;
L
Linus Torvalds 已提交
3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092

	if (TASK_NICE(p) == nice || nice < -20 || nice > 19)
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
	rq = task_rq_lock(p, &flags);
	/*
	 * The RT priorities are set via sched_setscheduler(), but we still
	 * allow the 'normal' nice value to be set - but as expected
	 * it wont have any effect on scheduling until the task is
I
Ingo Molnar 已提交
3093
	 * SCHED_FIFO/SCHED_RR:
L
Linus Torvalds 已提交
3094
	 */
3095
	if (task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3096 3097 3098
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
P
Peter Zijlstra 已提交
3099
	on_rq = p->on_rq;
3100
	if (on_rq)
3101
		dequeue_task(rq, p, 0);
L
Linus Torvalds 已提交
3102 3103

	p->static_prio = NICE_TO_PRIO(nice);
3104
	set_load_weight(p);
3105 3106 3107
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3108

I
Ingo Molnar 已提交
3109
	if (on_rq) {
3110
		enqueue_task(rq, p, 0);
L
Linus Torvalds 已提交
3111
		/*
3112 3113
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3114
		 */
3115
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
L
Linus Torvalds 已提交
3116 3117 3118
			resched_task(rq->curr);
	}
out_unlock:
3119
	task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
3120 3121 3122
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3123 3124 3125 3126 3127
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3128
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3129
{
3130 3131
	/* convert nice value [19,-20] to rlimit style value [1,40] */
	int nice_rlim = 20 - nice;
3132

3133
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3134 3135 3136
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3137 3138 3139 3140 3141 3142 3143 3144 3145
#ifdef __ARCH_WANT_SYS_NICE

/*
 * sys_nice - change the priority of the current process.
 * @increment: priority increment
 *
 * sys_setpriority is a more generic, but much slower function that
 * does similar things.
 */
3146
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3147
{
3148
	long nice, retval;
L
Linus Torvalds 已提交
3149 3150 3151 3152 3153 3154

	/*
	 * Setpriority might change our priority at the same moment.
	 * We don't have to worry. Conceptually one call occurs first
	 * and we have a single winner.
	 */
M
Matt Mackall 已提交
3155 3156
	if (increment < -40)
		increment = -40;
L
Linus Torvalds 已提交
3157 3158 3159
	if (increment > 40)
		increment = 40;

3160
	nice = TASK_NICE(current) + increment;
L
Linus Torvalds 已提交
3161 3162 3163 3164 3165
	if (nice < -20)
		nice = -20;
	if (nice > 19)
		nice = 19;

M
Matt Mackall 已提交
3166 3167 3168
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186
	retval = security_task_setnice(current, nice);
	if (retval)
		return retval;

	set_user_nice(current, nice);
	return 0;
}

#endif

/**
 * task_prio - return the priority value of a given task.
 * @p: the task in question.
 *
 * This is the priority value as seen by users in /proc.
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3187
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3188 3189 3190 3191 3192 3193 3194 3195
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * task_nice - return the nice value of a given task.
 * @p: the task in question.
 */
3196
int task_nice(const struct task_struct *p)
L
Linus Torvalds 已提交
3197 3198 3199
{
	return TASK_NICE(p);
}
P
Pavel Roskin 已提交
3200
EXPORT_SYMBOL(task_nice);
L
Linus Torvalds 已提交
3201 3202 3203 3204 3205 3206 3207

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221
	struct rq *rq = cpu_rq(cpu);

	if (rq->curr != rq->idle)
		return 0;

	if (rq->nr_running)
		return 0;

#ifdef CONFIG_SMP
	if (!llist_empty(&rq->wake_list))
		return 0;
#endif

	return 1;
L
Linus Torvalds 已提交
3222 3223 3224 3225 3226 3227
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
 */
3228
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3229 3230 3231 3232 3233 3234 3235 3236
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
 */
A
Alexey Dobriyan 已提交
3237
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3238
{
3239
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3240 3241 3242
}

/* Actually do priority change: must hold rq lock. */
I
Ingo Molnar 已提交
3243 3244
static void
__setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio)
L
Linus Torvalds 已提交
3245 3246 3247
{
	p->policy = policy;
	p->rt_priority = prio;
3248 3249 3250
	p->normal_prio = normal_prio(p);
	/* we are holding p->pi_lock already */
	p->prio = rt_mutex_getprio(p);
3251 3252 3253 3254
	if (rt_prio(p->prio))
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
3255
	set_load_weight(p);
L
Linus Torvalds 已提交
3256 3257
}

3258 3259 3260 3261 3262 3263 3264 3265 3266 3267
/*
 * check the target process has a UID that matches the current process's
 */
static bool check_same_owner(struct task_struct *p)
{
	const struct cred *cred = current_cred(), *pcred;
	bool match;

	rcu_read_lock();
	pcred = __task_cred(p);
3268 3269
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
3270 3271 3272 3273
	rcu_read_unlock();
	return match;
}

3274
static int __sched_setscheduler(struct task_struct *p, int policy,
3275
				const struct sched_param *param, bool user)
L
Linus Torvalds 已提交
3276
{
3277
	int retval, oldprio, oldpolicy = -1, on_rq, running;
L
Linus Torvalds 已提交
3278
	unsigned long flags;
3279
	const struct sched_class *prev_class;
3280
	struct rq *rq;
3281
	int reset_on_fork;
L
Linus Torvalds 已提交
3282

3283 3284
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
3285 3286
recheck:
	/* double check policy once rq lock held */
3287 3288
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
3289
		policy = oldpolicy = p->policy;
3290 3291 3292 3293 3294 3295 3296 3297 3298 3299
	} else {
		reset_on_fork = !!(policy & SCHED_RESET_ON_FORK);
		policy &= ~SCHED_RESET_ON_FORK;

		if (policy != SCHED_FIFO && policy != SCHED_RR &&
				policy != SCHED_NORMAL && policy != SCHED_BATCH &&
				policy != SCHED_IDLE)
			return -EINVAL;
	}

L
Linus Torvalds 已提交
3300 3301
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
3302 3303
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
3304 3305
	 */
	if (param->sched_priority < 0 ||
I
Ingo Molnar 已提交
3306
	    (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) ||
3307
	    (!p->mm && param->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
3308
		return -EINVAL;
3309
	if (rt_policy(policy) != (param->sched_priority != 0))
L
Linus Torvalds 已提交
3310 3311
		return -EINVAL;

3312 3313 3314
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
3315
	if (user && !capable(CAP_SYS_NICE)) {
3316
		if (rt_policy(policy)) {
3317 3318
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
3319 3320 3321 3322 3323 3324 3325 3326 3327 3328

			/* can't set/change the rt policy */
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

			/* can't increase priority */
			if (param->sched_priority > p->rt_priority &&
			    param->sched_priority > rlim_rtprio)
				return -EPERM;
		}
3329

I
Ingo Molnar 已提交
3330
		/*
3331 3332
		 * Treat SCHED_IDLE as nice 20. Only allow a switch to
		 * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
I
Ingo Molnar 已提交
3333
		 */
3334 3335 3336 3337
		if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) {
			if (!can_nice(p, TASK_NICE(p)))
				return -EPERM;
		}
3338

3339
		/* can't change other user's priorities */
3340
		if (!check_same_owner(p))
3341
			return -EPERM;
3342 3343 3344 3345

		/* Normal users shall not reset the sched_reset_on_fork flag */
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
3346
	}
L
Linus Torvalds 已提交
3347

3348
	if (user) {
3349
		retval = security_task_setscheduler(p);
3350 3351 3352 3353
		if (retval)
			return retval;
	}

3354 3355 3356
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
3357
	 *
L
Lucas De Marchi 已提交
3358
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
3359 3360
	 * runqueue lock must be held.
	 */
3361
	rq = task_rq_lock(p, &flags);
3362

3363 3364 3365 3366
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
3367
		task_rq_unlock(rq, p, &flags);
3368 3369 3370
		return -EINVAL;
	}

3371 3372 3373 3374 3375
	/*
	 * If not changing anything there's no need to proceed further:
	 */
	if (unlikely(policy == p->policy && (!rt_policy(policy) ||
			param->sched_priority == p->rt_priority))) {
3376
		task_rq_unlock(rq, p, &flags);
3377 3378 3379
		return 0;
	}

3380 3381 3382 3383 3384 3385 3386
#ifdef CONFIG_RT_GROUP_SCHED
	if (user) {
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
3387 3388
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
3389
			task_rq_unlock(rq, p, &flags);
3390 3391 3392 3393 3394
			return -EPERM;
		}
	}
#endif

L
Linus Torvalds 已提交
3395 3396 3397
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
3398
		task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
3399 3400
		goto recheck;
	}
P
Peter Zijlstra 已提交
3401
	on_rq = p->on_rq;
3402
	running = task_current(rq, p);
3403
	if (on_rq)
3404
		dequeue_task(rq, p, 0);
3405 3406
	if (running)
		p->sched_class->put_prev_task(rq, p);
3407

3408 3409
	p->sched_reset_on_fork = reset_on_fork;

L
Linus Torvalds 已提交
3410
	oldprio = p->prio;
3411
	prev_class = p->sched_class;
I
Ingo Molnar 已提交
3412
	__setscheduler(rq, p, policy, param->sched_priority);
3413

3414 3415
	if (running)
		p->sched_class->set_curr_task(rq);
P
Peter Zijlstra 已提交
3416
	if (on_rq)
3417
		enqueue_task(rq, p, 0);
3418

P
Peter Zijlstra 已提交
3419
	check_class_changed(rq, p, prev_class, oldprio);
3420
	task_rq_unlock(rq, p, &flags);
3421

3422 3423
	rt_mutex_adjust_pi(p);

L
Linus Torvalds 已提交
3424 3425
	return 0;
}
3426 3427 3428 3429 3430 3431 3432 3433 3434 3435

/**
 * sched_setscheduler - change the scheduling policy and/or RT priority of a thread.
 * @p: the task in question.
 * @policy: new policy.
 * @param: structure containing the new RT priority.
 *
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
3436
		       const struct sched_param *param)
3437 3438 3439
{
	return __sched_setscheduler(p, policy, param, true);
}
L
Linus Torvalds 已提交
3440 3441
EXPORT_SYMBOL_GPL(sched_setscheduler);

3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453
/**
 * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace.
 * @p: the task in question.
 * @policy: new policy.
 * @param: structure containing the new RT priority.
 *
 * Just like sched_setscheduler, only don't bother checking if the
 * current context has permission.  For example, this is needed in
 * stop_machine(): we create temporary high priority worker threads,
 * but our caller might not have that capability.
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
3454
			       const struct sched_param *param)
3455 3456 3457 3458
{
	return __sched_setscheduler(p, policy, param, false);
}

I
Ingo Molnar 已提交
3459 3460
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
3461 3462 3463
{
	struct sched_param lparam;
	struct task_struct *p;
3464
	int retval;
L
Linus Torvalds 已提交
3465 3466 3467 3468 3469

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
3470 3471 3472

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
3473
	p = find_process_by_pid(pid);
3474 3475 3476
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
3477

L
Linus Torvalds 已提交
3478 3479 3480 3481 3482 3483 3484 3485 3486
	return retval;
}

/**
 * sys_sched_setscheduler - set/change the scheduler policy and RT priority
 * @pid: the pid in question.
 * @policy: new policy.
 * @param: structure containing the new RT priority.
 */
3487 3488
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
3489
{
3490 3491 3492 3493
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
3494 3495 3496 3497 3498 3499 3500 3501
	return do_sched_setscheduler(pid, policy, param);
}

/**
 * sys_sched_setparam - set/change the RT priority of a thread
 * @pid: the pid in question.
 * @param: structure containing the new RT priority.
 */
3502
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
3503 3504 3505 3506 3507 3508 3509 3510
{
	return do_sched_setscheduler(pid, -1, param);
}

/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
 */
3511
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
3512
{
3513
	struct task_struct *p;
3514
	int retval;
L
Linus Torvalds 已提交
3515 3516

	if (pid < 0)
3517
		return -EINVAL;
L
Linus Torvalds 已提交
3518 3519

	retval = -ESRCH;
3520
	rcu_read_lock();
L
Linus Torvalds 已提交
3521 3522 3523 3524
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
3525 3526
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
3527
	}
3528
	rcu_read_unlock();
L
Linus Torvalds 已提交
3529 3530 3531 3532
	return retval;
}

/**
3533
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
3534 3535 3536
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
 */
3537
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
3538 3539
{
	struct sched_param lp;
3540
	struct task_struct *p;
3541
	int retval;
L
Linus Torvalds 已提交
3542 3543

	if (!param || pid < 0)
3544
		return -EINVAL;
L
Linus Torvalds 已提交
3545

3546
	rcu_read_lock();
L
Linus Torvalds 已提交
3547 3548 3549 3550 3551 3552 3553 3554 3555 3556
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

	lp.sched_priority = p->rt_priority;
3557
	rcu_read_unlock();
L
Linus Torvalds 已提交
3558 3559 3560 3561 3562 3563 3564 3565 3566

	/*
	 * This one might sleep, we cannot do it with a spinlock held ...
	 */
	retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0;

	return retval;

out_unlock:
3567
	rcu_read_unlock();
L
Linus Torvalds 已提交
3568 3569 3570
	return retval;
}

3571
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
3572
{
3573
	cpumask_var_t cpus_allowed, new_mask;
3574 3575
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
3576

3577
	get_online_cpus();
3578
	rcu_read_lock();
L
Linus Torvalds 已提交
3579 3580 3581

	p = find_process_by_pid(pid);
	if (!p) {
3582
		rcu_read_unlock();
3583
		put_online_cpus();
L
Linus Torvalds 已提交
3584 3585 3586
		return -ESRCH;
	}

3587
	/* Prevent p going away */
L
Linus Torvalds 已提交
3588
	get_task_struct(p);
3589
	rcu_read_unlock();
L
Linus Torvalds 已提交
3590

3591 3592 3593 3594
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
3595 3596 3597 3598 3599 3600 3601 3602
	if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
		retval = -ENOMEM;
		goto out_put_task;
	}
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) {
		retval = -ENOMEM;
		goto out_free_cpus_allowed;
	}
L
Linus Torvalds 已提交
3603
	retval = -EPERM;
E
Eric W. Biederman 已提交
3604 3605 3606 3607 3608 3609 3610 3611
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
			goto out_unlock;
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
3612

3613
	retval = security_task_setscheduler(p);
3614 3615 3616
	if (retval)
		goto out_unlock;

3617 3618
	cpuset_cpus_allowed(p, cpus_allowed);
	cpumask_and(new_mask, in_mask, cpus_allowed);
P
Peter Zijlstra 已提交
3619
again:
3620
	retval = set_cpus_allowed_ptr(p, new_mask);
L
Linus Torvalds 已提交
3621

P
Paul Menage 已提交
3622
	if (!retval) {
3623 3624
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
3625 3626 3627 3628 3629
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
3630
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
3631 3632 3633
			goto again;
		}
	}
L
Linus Torvalds 已提交
3634
out_unlock:
3635 3636 3637 3638
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
3639
	put_task_struct(p);
3640
	put_online_cpus();
L
Linus Torvalds 已提交
3641 3642 3643 3644
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
3645
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
3646
{
3647 3648 3649 3650 3651
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
3652 3653 3654 3655 3656 3657 3658 3659 3660
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
 * sys_sched_setaffinity - set the cpu affinity of a process
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
 * @user_mask_ptr: user-space pointer to the new cpu mask
 */
3661 3662
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
3663
{
3664
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
3665 3666
	int retval;

3667 3668
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
3669

3670 3671 3672 3673 3674
	retval = get_user_cpu_mask(user_mask_ptr, len, new_mask);
	if (retval == 0)
		retval = sched_setaffinity(pid, new_mask);
	free_cpumask_var(new_mask);
	return retval;
L
Linus Torvalds 已提交
3675 3676
}

3677
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
3678
{
3679
	struct task_struct *p;
3680
	unsigned long flags;
L
Linus Torvalds 已提交
3681 3682
	int retval;

3683
	get_online_cpus();
3684
	rcu_read_lock();
L
Linus Torvalds 已提交
3685 3686 3687 3688 3689 3690

	retval = -ESRCH;
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

3691 3692 3693 3694
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

3695
	raw_spin_lock_irqsave(&p->pi_lock, flags);
3696
	cpumask_and(mask, &p->cpus_allowed, cpu_online_mask);
3697
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
3698 3699

out_unlock:
3700
	rcu_read_unlock();
3701
	put_online_cpus();
L
Linus Torvalds 已提交
3702

3703
	return retval;
L
Linus Torvalds 已提交
3704 3705 3706 3707 3708 3709 3710 3711
}

/**
 * sys_sched_getaffinity - get the cpu affinity of a process
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
 * @user_mask_ptr: user-space pointer to hold the current cpu mask
 */
3712 3713
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
3714 3715
{
	int ret;
3716
	cpumask_var_t mask;
L
Linus Torvalds 已提交
3717

A
Anton Blanchard 已提交
3718
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
3719 3720
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
3721 3722
		return -EINVAL;

3723 3724
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
3725

3726 3727
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
3728
		size_t retlen = min_t(size_t, len, cpumask_size());
3729 3730

		if (copy_to_user(user_mask_ptr, mask, retlen))
3731 3732
			ret = -EFAULT;
		else
3733
			ret = retlen;
3734 3735
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
3736

3737
	return ret;
L
Linus Torvalds 已提交
3738 3739 3740 3741 3742
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
3743 3744
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
L
Linus Torvalds 已提交
3745
 */
3746
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
3747
{
3748
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
3749

3750
	schedstat_inc(rq, yld_count);
3751
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
3752 3753 3754 3755 3756 3757

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
3758
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
3759
	do_raw_spin_unlock(&rq->lock);
3760
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
3761 3762 3763 3764 3765 3766

	schedule();

	return 0;
}

P
Peter Zijlstra 已提交
3767 3768 3769 3770 3771
static inline int should_resched(void)
{
	return need_resched() && !(preempt_count() & PREEMPT_ACTIVE);
}

A
Andrew Morton 已提交
3772
static void __cond_resched(void)
L
Linus Torvalds 已提交
3773
{
3774
	add_preempt_count(PREEMPT_ACTIVE);
3775
	__schedule();
3776
	sub_preempt_count(PREEMPT_ACTIVE);
L
Linus Torvalds 已提交
3777 3778
}

3779
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
3780
{
P
Peter Zijlstra 已提交
3781
	if (should_resched()) {
L
Linus Torvalds 已提交
3782 3783 3784 3785 3786
		__cond_resched();
		return 1;
	}
	return 0;
}
3787
EXPORT_SYMBOL(_cond_resched);
L
Linus Torvalds 已提交
3788 3789

/*
3790
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
3791 3792
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
3793
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
3794 3795 3796
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
3797
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
3798
{
P
Peter Zijlstra 已提交
3799
	int resched = should_resched();
J
Jan Kara 已提交
3800 3801
	int ret = 0;

3802 3803
	lockdep_assert_held(lock);

N
Nick Piggin 已提交
3804
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
3805
		spin_unlock(lock);
P
Peter Zijlstra 已提交
3806
		if (resched)
N
Nick Piggin 已提交
3807 3808 3809
			__cond_resched();
		else
			cpu_relax();
J
Jan Kara 已提交
3810
		ret = 1;
L
Linus Torvalds 已提交
3811 3812
		spin_lock(lock);
	}
J
Jan Kara 已提交
3813
	return ret;
L
Linus Torvalds 已提交
3814
}
3815
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
3816

3817
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
3818 3819 3820
{
	BUG_ON(!in_softirq());

P
Peter Zijlstra 已提交
3821
	if (should_resched()) {
3822
		local_bh_enable();
L
Linus Torvalds 已提交
3823 3824 3825 3826 3827 3828
		__cond_resched();
		local_bh_disable();
		return 1;
	}
	return 0;
}
3829
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
3830 3831 3832 3833

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851
 * Do not ever use this function, there's a 99% chance you're doing it wrong.
 *
 * The scheduler is at all times free to pick the calling task as the most
 * eligible task to run, if removing the yield() call from your code breaks
 * it, its already broken.
 *
 * Typical broken usage is:
 *
 * while (!event)
 * 	yield();
 *
 * where one assumes that yield() will let 'the other' process run that will
 * make event true. If the current task is a SCHED_FIFO task that will never
 * happen. Never use yield() as a progress guarantee!!
 *
 * If you want to use yield() to wait for something, use wait_event().
 * If you want to use yield() to be 'nice' for others, use cond_resched().
 * If you still want to use yield(), do not!
L
Linus Torvalds 已提交
3852 3853 3854 3855 3856 3857 3858 3859
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

3860 3861 3862 3863
/**
 * yield_to - yield the current processor to another thread in
 * your thread group, or accelerate that thread toward the
 * processor it's on.
R
Randy Dunlap 已提交
3864 3865
 * @p: target task
 * @preempt: whether task preemption is allowed or not
3866 3867 3868 3869
 *
 * It's the caller's job to ensure that the target task struct
 * can't go away on us before we can do any checks.
 *
3870 3871 3872 3873
 * Returns:
 *	true (>0) if we indeed boosted the target task.
 *	false (0) if we failed to boost the target.
 *	-ESRCH if there's no task to yield to.
3874 3875 3876 3877 3878 3879
 */
bool __sched yield_to(struct task_struct *p, bool preempt)
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
3880
	int yielded = 0;
3881 3882 3883 3884 3885 3886

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
3887 3888 3889 3890 3891 3892 3893 3894 3895
	/*
	 * If we're the only runnable task on the rq and target rq also
	 * has only one task, there's absolutely no point in yielding.
	 */
	if (rq->nr_running == 1 && p_rq->nr_running == 1) {
		yielded = -ESRCH;
		goto out_irq;
	}

3896 3897 3898 3899 3900 3901 3902
	double_rq_lock(rq, p_rq);
	while (task_rq(p) != p_rq) {
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
3903
		goto out_unlock;
3904 3905

	if (curr->sched_class != p->sched_class)
3906
		goto out_unlock;
3907 3908

	if (task_running(p_rq, p) || p->state)
3909
		goto out_unlock;
3910 3911

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
3912
	if (yielded) {
3913
		schedstat_inc(rq, yld_count);
3914 3915 3916 3917 3918 3919 3920
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
			resched_task(p_rq->curr);
	}
3921

3922
out_unlock:
3923
	double_rq_unlock(rq, p_rq);
3924
out_irq:
3925 3926
	local_irq_restore(flags);

3927
	if (yielded > 0)
3928 3929 3930 3931 3932 3933
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
3934
/*
I
Ingo Molnar 已提交
3935
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
3936 3937 3938 3939
 * that process accounting knows that this is a task in IO wait state.
 */
void __sched io_schedule(void)
{
3940
	struct rq *rq = raw_rq();
L
Linus Torvalds 已提交
3941

3942
	delayacct_blkio_start();
L
Linus Torvalds 已提交
3943
	atomic_inc(&rq->nr_iowait);
3944
	blk_flush_plug(current);
3945
	current->in_iowait = 1;
L
Linus Torvalds 已提交
3946
	schedule();
3947
	current->in_iowait = 0;
L
Linus Torvalds 已提交
3948
	atomic_dec(&rq->nr_iowait);
3949
	delayacct_blkio_end();
L
Linus Torvalds 已提交
3950 3951 3952 3953 3954
}
EXPORT_SYMBOL(io_schedule);

long __sched io_schedule_timeout(long timeout)
{
3955
	struct rq *rq = raw_rq();
L
Linus Torvalds 已提交
3956 3957
	long ret;

3958
	delayacct_blkio_start();
L
Linus Torvalds 已提交
3959
	atomic_inc(&rq->nr_iowait);
3960
	blk_flush_plug(current);
3961
	current->in_iowait = 1;
L
Linus Torvalds 已提交
3962
	ret = schedule_timeout(timeout);
3963
	current->in_iowait = 0;
L
Linus Torvalds 已提交
3964
	atomic_dec(&rq->nr_iowait);
3965
	delayacct_blkio_end();
L
Linus Torvalds 已提交
3966 3967 3968 3969 3970 3971 3972 3973 3974 3975
	return ret;
}

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
 * this syscall returns the maximum rt_priority that can be used
 * by a given scheduling class.
 */
3976
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
3977 3978 3979 3980 3981 3982 3983 3984 3985
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
	case SCHED_NORMAL:
3986
	case SCHED_BATCH:
I
Ingo Molnar 已提交
3987
	case SCHED_IDLE:
L
Linus Torvalds 已提交
3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
 * this syscall returns the minimum rt_priority that can be used
 * by a given scheduling class.
 */
4001
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
4002 4003 4004 4005 4006 4007 4008 4009 4010
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
	case SCHED_NORMAL:
4011
	case SCHED_BATCH:
I
Ingo Molnar 已提交
4012
	case SCHED_IDLE:
L
Linus Torvalds 已提交
4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025
		ret = 0;
	}
	return ret;
}

/**
 * sys_sched_rr_get_interval - return the default timeslice of a process.
 * @pid: pid of the process.
 * @interval: userspace pointer to the timeslice value.
 *
 * this syscall writes the default timeslice value of a given process
 * into the user-space timespec buffer. A value of '0' means infinity.
 */
4026
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
4027
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
4028
{
4029
	struct task_struct *p;
D
Dmitry Adamushko 已提交
4030
	unsigned int time_slice;
4031 4032
	unsigned long flags;
	struct rq *rq;
4033
	int retval;
L
Linus Torvalds 已提交
4034 4035 4036
	struct timespec t;

	if (pid < 0)
4037
		return -EINVAL;
L
Linus Torvalds 已提交
4038 4039

	retval = -ESRCH;
4040
	rcu_read_lock();
L
Linus Torvalds 已提交
4041 4042 4043 4044 4045 4046 4047 4048
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4049 4050
	rq = task_rq_lock(p, &flags);
	time_slice = p->sched_class->get_rr_interval(rq, p);
4051
	task_rq_unlock(rq, p, &flags);
D
Dmitry Adamushko 已提交
4052

4053
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
4054
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
4055 4056
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
4057

L
Linus Torvalds 已提交
4058
out_unlock:
4059
	rcu_read_unlock();
L
Linus Torvalds 已提交
4060 4061 4062
	return retval;
}

4063
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
4064

4065
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
4066 4067
{
	unsigned long free = 0;
4068
	int ppid;
4069
	unsigned state;
L
Linus Torvalds 已提交
4070 4071

	state = p->state ? __ffs(p->state) + 1 : 0;
4072
	printk(KERN_INFO "%-15.15s %c", p->comm,
4073
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
4074
#if BITS_PER_LONG == 32
L
Linus Torvalds 已提交
4075
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
4076
		printk(KERN_CONT " running  ");
L
Linus Torvalds 已提交
4077
	else
P
Peter Zijlstra 已提交
4078
		printk(KERN_CONT " %08lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
4079 4080
#else
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
4081
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
4082
	else
P
Peter Zijlstra 已提交
4083
		printk(KERN_CONT " %016lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
4084 4085
#endif
#ifdef CONFIG_DEBUG_STACK_USAGE
4086
	free = stack_not_used(p);
L
Linus Torvalds 已提交
4087
#endif
4088 4089 4090
	rcu_read_lock();
	ppid = task_pid_nr(rcu_dereference(p->real_parent));
	rcu_read_unlock();
P
Peter Zijlstra 已提交
4091
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
4092
		task_pid_nr(p), ppid,
4093
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
4094

4095
	print_worker_info(KERN_INFO, p);
4096
	show_stack(p, NULL);
L
Linus Torvalds 已提交
4097 4098
}

I
Ingo Molnar 已提交
4099
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
4100
{
4101
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
4102

4103
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
4104 4105
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
4106
#else
P
Peter Zijlstra 已提交
4107 4108
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
4109
#endif
4110
	rcu_read_lock();
L
Linus Torvalds 已提交
4111 4112 4113
	do_each_thread(g, p) {
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
4114
		 * console might take a lot of time:
L
Linus Torvalds 已提交
4115 4116
		 */
		touch_nmi_watchdog();
I
Ingo Molnar 已提交
4117
		if (!state_filter || (p->state & state_filter))
4118
			sched_show_task(p);
L
Linus Torvalds 已提交
4119 4120
	} while_each_thread(g, p);

4121 4122
	touch_all_softlockup_watchdogs();

I
Ingo Molnar 已提交
4123 4124 4125
#ifdef CONFIG_SCHED_DEBUG
	sysrq_sched_debug_show();
#endif
4126
	rcu_read_unlock();
I
Ingo Molnar 已提交
4127 4128 4129
	/*
	 * Only show locks if all tasks are dumped:
	 */
4130
	if (!state_filter)
I
Ingo Molnar 已提交
4131
		debug_show_all_locks();
L
Linus Torvalds 已提交
4132 4133
}

I
Ingo Molnar 已提交
4134 4135
void __cpuinit init_idle_bootup_task(struct task_struct *idle)
{
I
Ingo Molnar 已提交
4136
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
4137 4138
}

4139 4140 4141 4142 4143 4144 4145 4146
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
 * @cpu: cpu the idle task belongs to
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
4147
void __cpuinit init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
4148
{
4149
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
4150 4151
	unsigned long flags;

4152
	raw_spin_lock_irqsave(&rq->lock, flags);
4153

I
Ingo Molnar 已提交
4154
	__sched_fork(idle);
4155
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
4156 4157
	idle->se.exec_start = sched_clock();

4158
	do_set_cpus_allowed(idle, cpumask_of(cpu));
4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169
	/*
	 * We're having a chicken and egg problem, even though we are
	 * holding rq->lock, the cpu isn't yet set to this cpu so the
	 * lockdep check in task_group() will fail.
	 *
	 * Similar case to sched_fork(). / Alternatively we could
	 * use task_rq_lock() here and obtain the other rq->lock.
	 *
	 * Silence PROVE_RCU
	 */
	rcu_read_lock();
I
Ingo Molnar 已提交
4170
	__set_task_cpu(idle, cpu);
4171
	rcu_read_unlock();
L
Linus Torvalds 已提交
4172 4173

	rq->curr = rq->idle = idle;
P
Peter Zijlstra 已提交
4174 4175
#if defined(CONFIG_SMP)
	idle->on_cpu = 1;
4176
#endif
4177
	raw_spin_unlock_irqrestore(&rq->lock, flags);
L
Linus Torvalds 已提交
4178 4179

	/* Set the preempt count _outside_ the spinlocks! */
A
Al Viro 已提交
4180
	task_thread_info(idle)->preempt_count = 0;
4181

I
Ingo Molnar 已提交
4182 4183 4184 4185
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
4186
	ftrace_graph_init_idle_task(idle, cpu);
4187
	vtime_init_idle(idle, cpu);
4188 4189 4190
#if defined(CONFIG_SMP)
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
4191 4192
}

L
Linus Torvalds 已提交
4193
#ifdef CONFIG_SMP
4194 4195 4196 4197
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
	if (p->sched_class && p->sched_class->set_cpus_allowed)
		p->sched_class->set_cpus_allowed(p, new_mask);
4198 4199

	cpumask_copy(&p->cpus_allowed, new_mask);
4200
	p->nr_cpus_allowed = cpumask_weight(new_mask);
4201 4202
}

L
Linus Torvalds 已提交
4203 4204 4205
/*
 * This is how migration works:
 *
4206 4207 4208 4209 4210 4211
 * 1) we invoke migration_cpu_stop() on the target CPU using
 *    stop_one_cpu().
 * 2) stopper starts to run (implicitly forcing the migrated thread
 *    off the CPU)
 * 3) it checks whether the migrated task is still in the wrong runqueue.
 * 4) if it's in the wrong runqueue then the migration thread removes
L
Linus Torvalds 已提交
4212
 *    it and puts it into the right queue.
4213 4214
 * 5) stopper completes and stop_one_cpu() returns and the migration
 *    is done.
L
Linus Torvalds 已提交
4215 4216 4217 4218 4219 4220 4221 4222
 */

/*
 * Change a given task's CPU affinity. Migrate the thread to a
 * proper CPU and schedule it away if the CPU it's executing on
 * is removed from the allowed bitmask.
 *
 * NOTE: the caller must have a valid reference to the task, the
I
Ingo Molnar 已提交
4223
 * task must not exit() & deallocate itself prematurely. The
L
Linus Torvalds 已提交
4224 4225
 * call is not atomic; no spinlocks may be held.
 */
4226
int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
L
Linus Torvalds 已提交
4227 4228
{
	unsigned long flags;
4229
	struct rq *rq;
4230
	unsigned int dest_cpu;
4231
	int ret = 0;
L
Linus Torvalds 已提交
4232 4233

	rq = task_rq_lock(p, &flags);
4234

4235 4236 4237
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

4238
	if (!cpumask_intersects(new_mask, cpu_active_mask)) {
L
Linus Torvalds 已提交
4239 4240 4241 4242
		ret = -EINVAL;
		goto out;
	}

4243
	do_set_cpus_allowed(p, new_mask);
4244

L
Linus Torvalds 已提交
4245
	/* Can the task run on the task's current CPU? If so, we're done */
4246
	if (cpumask_test_cpu(task_cpu(p), new_mask))
L
Linus Torvalds 已提交
4247 4248
		goto out;

4249
	dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
4250
	if (p->on_rq) {
4251
		struct migration_arg arg = { p, dest_cpu };
L
Linus Torvalds 已提交
4252
		/* Need help from migration thread: drop lock and wait. */
4253
		task_rq_unlock(rq, p, &flags);
4254
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
4255 4256 4257 4258
		tlb_migrate_finish(p->mm);
		return 0;
	}
out:
4259
	task_rq_unlock(rq, p, &flags);
4260

L
Linus Torvalds 已提交
4261 4262
	return ret;
}
4263
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
L
Linus Torvalds 已提交
4264 4265

/*
I
Ingo Molnar 已提交
4266
 * Move (not current) task off this cpu, onto dest cpu. We're doing
L
Linus Torvalds 已提交
4267 4268 4269 4270 4271 4272
 * this because either it can't run here any more (set_cpus_allowed()
 * away from this CPU, or CPU going down), or because we're
 * attempting to rebalance this task on exec (sched_exec).
 *
 * So we race with normal scheduler movements, but that's OK, as long
 * as the task is no longer on this CPU.
4273 4274
 *
 * Returns non-zero if task was successfully migrated.
L
Linus Torvalds 已提交
4275
 */
4276
static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
L
Linus Torvalds 已提交
4277
{
4278
	struct rq *rq_dest, *rq_src;
4279
	int ret = 0;
L
Linus Torvalds 已提交
4280

4281
	if (unlikely(!cpu_active(dest_cpu)))
4282
		return ret;
L
Linus Torvalds 已提交
4283 4284 4285 4286

	rq_src = cpu_rq(src_cpu);
	rq_dest = cpu_rq(dest_cpu);

4287
	raw_spin_lock(&p->pi_lock);
L
Linus Torvalds 已提交
4288 4289 4290
	double_rq_lock(rq_src, rq_dest);
	/* Already moved. */
	if (task_cpu(p) != src_cpu)
L
Linus Torvalds 已提交
4291
		goto done;
L
Linus Torvalds 已提交
4292
	/* Affinity changed (again). */
4293
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
L
Linus Torvalds 已提交
4294
		goto fail;
L
Linus Torvalds 已提交
4295

4296 4297 4298 4299
	/*
	 * If we're not on a rq, the next wake-up will ensure we're
	 * placed properly.
	 */
P
Peter Zijlstra 已提交
4300
	if (p->on_rq) {
4301
		dequeue_task(rq_src, p, 0);
4302
		set_task_cpu(p, dest_cpu);
4303
		enqueue_task(rq_dest, p, 0);
4304
		check_preempt_curr(rq_dest, p, 0);
L
Linus Torvalds 已提交
4305
	}
L
Linus Torvalds 已提交
4306
done:
4307
	ret = 1;
L
Linus Torvalds 已提交
4308
fail:
L
Linus Torvalds 已提交
4309
	double_rq_unlock(rq_src, rq_dest);
4310
	raw_spin_unlock(&p->pi_lock);
4311
	return ret;
L
Linus Torvalds 已提交
4312 4313 4314
}

/*
4315 4316 4317
 * migration_cpu_stop - this will be executed by a highprio stopper thread
 * and performs thread migration by bumping thread off CPU then
 * 'pushing' onto another runqueue.
L
Linus Torvalds 已提交
4318
 */
4319
static int migration_cpu_stop(void *data)
L
Linus Torvalds 已提交
4320
{
4321
	struct migration_arg *arg = data;
4322

4323 4324 4325 4326
	/*
	 * The original target cpu might have gone down and we might
	 * be on another cpu but it doesn't matter.
	 */
4327
	local_irq_disable();
4328
	__migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu);
4329
	local_irq_enable();
L
Linus Torvalds 已提交
4330
	return 0;
4331 4332
}

L
Linus Torvalds 已提交
4333
#ifdef CONFIG_HOTPLUG_CPU
4334

4335
/*
4336 4337
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
4338
 */
4339
void idle_task_exit(void)
L
Linus Torvalds 已提交
4340
{
4341
	struct mm_struct *mm = current->active_mm;
4342

4343
	BUG_ON(cpu_online(smp_processor_id()));
4344

4345 4346 4347
	if (mm != &init_mm)
		switch_mm(mm, &init_mm, current);
	mmdrop(mm);
L
Linus Torvalds 已提交
4348 4349 4350
}

/*
4351 4352 4353 4354 4355
 * Since this CPU is going 'away' for a while, fold any nr_active delta
 * we might have. Assumes we're called after migrate_tasks() so that the
 * nr_active count is stable.
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
4356
 */
4357
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
4358
{
4359 4360 4361
	long delta = calc_load_fold_active(rq);
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
4362 4363
}

4364
/*
4365 4366 4367 4368 4369 4370
 * Migrate all tasks from the rq, sleeping tasks will be migrated by
 * try_to_wake_up()->select_task_rq().
 *
 * Called with rq->lock held even though we'er in stop_machine() and
 * there's no concurrency possible, we hold the required locks anyway
 * because of lock validation efforts.
L
Linus Torvalds 已提交
4371
 */
4372
static void migrate_tasks(unsigned int dead_cpu)
L
Linus Torvalds 已提交
4373
{
4374
	struct rq *rq = cpu_rq(dead_cpu);
4375 4376
	struct task_struct *next, *stop = rq->stop;
	int dest_cpu;
L
Linus Torvalds 已提交
4377 4378

	/*
4379 4380 4381 4382 4383 4384 4385
	 * Fudge the rq selection such that the below task selection loop
	 * doesn't get stuck on the currently eligible stop task.
	 *
	 * We're currently inside stop_machine() and the rq is either stuck
	 * in the stop_machine_cpu_stop() loop, or we're executing this code,
	 * either way we should never end up calling schedule() until we're
	 * done here.
L
Linus Torvalds 已提交
4386
	 */
4387
	rq->stop = NULL;
4388

4389 4390 4391 4392 4393 4394 4395
	/*
	 * put_prev_task() and pick_next_task() sched
	 * class method both need to have an up-to-date
	 * value of rq->clock[_task]
	 */
	update_rq_clock(rq);

I
Ingo Molnar 已提交
4396
	for ( ; ; ) {
4397 4398 4399 4400 4401
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
4402
			break;
4403

4404
		next = pick_next_task(rq);
4405
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
4406
		next->sched_class->put_prev_task(rq, next);
4407

4408 4409 4410 4411 4412 4413 4414
		/* Find suitable destination for @next, with force if needed. */
		dest_cpu = select_fallback_rq(dead_cpu, next);
		raw_spin_unlock(&rq->lock);

		__migrate_task(next, dead_cpu, dest_cpu);

		raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
4415
	}
4416

4417
	rq->stop = stop;
4418
}
4419

L
Linus Torvalds 已提交
4420 4421
#endif /* CONFIG_HOTPLUG_CPU */

4422 4423 4424
#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)

static struct ctl_table sd_ctl_dir[] = {
4425 4426
	{
		.procname	= "sched_domain",
4427
		.mode		= 0555,
4428
	},
4429
	{}
4430 4431 4432
};

static struct ctl_table sd_ctl_root[] = {
4433 4434
	{
		.procname	= "kernel",
4435
		.mode		= 0555,
4436 4437
		.child		= sd_ctl_dir,
	},
4438
	{}
4439 4440 4441 4442 4443
};

static struct ctl_table *sd_alloc_ctl_entry(int n)
{
	struct ctl_table *entry =
4444
		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
4445 4446 4447 4448

	return entry;
}

4449 4450
static void sd_free_ctl_entry(struct ctl_table **tablep)
{
4451
	struct ctl_table *entry;
4452

4453 4454 4455
	/*
	 * In the intermediate directories, both the child directory and
	 * procname are dynamically allocated and could fail but the mode
I
Ingo Molnar 已提交
4456
	 * will always be set. In the lowest directory the names are
4457 4458 4459
	 * static strings and all have proc handlers.
	 */
	for (entry = *tablep; entry->mode; entry++) {
4460 4461
		if (entry->child)
			sd_free_ctl_entry(&entry->child);
4462 4463 4464
		if (entry->proc_handler == NULL)
			kfree(entry->procname);
	}
4465 4466 4467 4468 4469

	kfree(*tablep);
	*tablep = NULL;
}

4470
static int min_load_idx = 0;
4471
static int max_load_idx = CPU_LOAD_IDX_MAX-1;
4472

4473
static void
4474
set_table_entry(struct ctl_table *entry,
4475
		const char *procname, void *data, int maxlen,
4476 4477
		umode_t mode, proc_handler *proc_handler,
		bool load_idx)
4478 4479 4480 4481 4482 4483
{
	entry->procname = procname;
	entry->data = data;
	entry->maxlen = maxlen;
	entry->mode = mode;
	entry->proc_handler = proc_handler;
4484 4485 4486 4487 4488

	if (load_idx) {
		entry->extra1 = &min_load_idx;
		entry->extra2 = &max_load_idx;
	}
4489 4490 4491 4492 4493
}

static struct ctl_table *
sd_alloc_ctl_domain_table(struct sched_domain *sd)
{
4494
	struct ctl_table *table = sd_alloc_ctl_entry(13);
4495

4496 4497 4498
	if (table == NULL)
		return NULL;

4499
	set_table_entry(&table[0], "min_interval", &sd->min_interval,
4500
		sizeof(long), 0644, proc_doulongvec_minmax, false);
4501
	set_table_entry(&table[1], "max_interval", &sd->max_interval,
4502
		sizeof(long), 0644, proc_doulongvec_minmax, false);
4503
	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
4504
		sizeof(int), 0644, proc_dointvec_minmax, true);
4505
	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
4506
		sizeof(int), 0644, proc_dointvec_minmax, true);
4507
	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
4508
		sizeof(int), 0644, proc_dointvec_minmax, true);
4509
	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
4510
		sizeof(int), 0644, proc_dointvec_minmax, true);
4511
	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
4512
		sizeof(int), 0644, proc_dointvec_minmax, true);
4513
	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
4514
		sizeof(int), 0644, proc_dointvec_minmax, false);
4515
	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
4516
		sizeof(int), 0644, proc_dointvec_minmax, false);
4517
	set_table_entry(&table[9], "cache_nice_tries",
4518
		&sd->cache_nice_tries,
4519
		sizeof(int), 0644, proc_dointvec_minmax, false);
4520
	set_table_entry(&table[10], "flags", &sd->flags,
4521
		sizeof(int), 0644, proc_dointvec_minmax, false);
4522
	set_table_entry(&table[11], "name", sd->name,
4523
		CORENAME_MAX_SIZE, 0444, proc_dostring, false);
4524
	/* &table[12] is terminator */
4525 4526 4527 4528

	return table;
}

4529
static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
4530 4531 4532 4533 4534 4535 4536 4537 4538
{
	struct ctl_table *entry, *table;
	struct sched_domain *sd;
	int domain_num = 0, i;
	char buf[32];

	for_each_domain(cpu, sd)
		domain_num++;
	entry = table = sd_alloc_ctl_entry(domain_num + 1);
4539 4540
	if (table == NULL)
		return NULL;
4541 4542 4543 4544 4545

	i = 0;
	for_each_domain(cpu, sd) {
		snprintf(buf, 32, "domain%d", i);
		entry->procname = kstrdup(buf, GFP_KERNEL);
4546
		entry->mode = 0555;
4547 4548 4549 4550 4551 4552 4553 4554
		entry->child = sd_alloc_ctl_domain_table(sd);
		entry++;
		i++;
	}
	return table;
}

static struct ctl_table_header *sd_sysctl_header;
4555
static void register_sched_domain_sysctl(void)
4556
{
4557
	int i, cpu_num = num_possible_cpus();
4558 4559 4560
	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
	char buf[32];

4561 4562 4563
	WARN_ON(sd_ctl_dir[0].child);
	sd_ctl_dir[0].child = entry;

4564 4565 4566
	if (entry == NULL)
		return;

4567
	for_each_possible_cpu(i) {
4568 4569
		snprintf(buf, 32, "cpu%d", i);
		entry->procname = kstrdup(buf, GFP_KERNEL);
4570
		entry->mode = 0555;
4571
		entry->child = sd_alloc_ctl_cpu_table(i);
4572
		entry++;
4573
	}
4574 4575

	WARN_ON(sd_sysctl_header);
4576 4577
	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
}
4578

4579
/* may be called multiple times per register */
4580 4581
static void unregister_sched_domain_sysctl(void)
{
4582 4583
	if (sd_sysctl_header)
		unregister_sysctl_table(sd_sysctl_header);
4584
	sd_sysctl_header = NULL;
4585 4586
	if (sd_ctl_dir[0].child)
		sd_free_ctl_entry(&sd_ctl_dir[0].child);
4587
}
4588
#else
4589 4590 4591 4592
static void register_sched_domain_sysctl(void)
{
}
static void unregister_sched_domain_sysctl(void)
4593 4594 4595 4596
{
}
#endif

4597 4598 4599 4600 4601
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

4602
		cpumask_set_cpu(rq->cpu, rq->rd->online);
4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621
		rq->online = 1;

		for_each_class(class) {
			if (class->rq_online)
				class->rq_online(rq);
		}
	}
}

static void set_rq_offline(struct rq *rq)
{
	if (rq->online) {
		const struct sched_class *class;

		for_each_class(class) {
			if (class->rq_offline)
				class->rq_offline(rq);
		}

4622
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
4623 4624 4625 4626
		rq->online = 0;
	}
}

L
Linus Torvalds 已提交
4627 4628 4629 4630
/*
 * migration_call - callback that gets triggered when a CPU is added.
 * Here we can start up the necessary migration thread for the new CPU.
 */
4631 4632
static int __cpuinit
migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
L
Linus Torvalds 已提交
4633
{
4634
	int cpu = (long)hcpu;
L
Linus Torvalds 已提交
4635
	unsigned long flags;
4636
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
4637

4638
	switch (action & ~CPU_TASKS_FROZEN) {
4639

L
Linus Torvalds 已提交
4640
	case CPU_UP_PREPARE:
4641
		rq->calc_load_update = calc_load_update;
L
Linus Torvalds 已提交
4642
		break;
4643

L
Linus Torvalds 已提交
4644
	case CPU_ONLINE:
4645
		/* Update our root-domain */
4646
		raw_spin_lock_irqsave(&rq->lock, flags);
4647
		if (rq->rd) {
4648
			BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
4649 4650

			set_rq_online(rq);
4651
		}
4652
		raw_spin_unlock_irqrestore(&rq->lock, flags);
L
Linus Torvalds 已提交
4653
		break;
4654

L
Linus Torvalds 已提交
4655
#ifdef CONFIG_HOTPLUG_CPU
4656
	case CPU_DYING:
4657
		sched_ttwu_pending();
G
Gregory Haskins 已提交
4658
		/* Update our root-domain */
4659
		raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
4660
		if (rq->rd) {
4661
			BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
4662
			set_rq_offline(rq);
G
Gregory Haskins 已提交
4663
		}
4664 4665
		migrate_tasks(cpu);
		BUG_ON(rq->nr_running != 1); /* the migration thread */
4666
		raw_spin_unlock_irqrestore(&rq->lock, flags);
4667
		break;
4668

4669
	case CPU_DEAD:
4670
		calc_load_migrate(rq);
G
Gregory Haskins 已提交
4671
		break;
L
Linus Torvalds 已提交
4672 4673
#endif
	}
4674 4675 4676

	update_max_interval();

L
Linus Torvalds 已提交
4677 4678 4679
	return NOTIFY_OK;
}

4680 4681 4682
/*
 * Register at high priority so that task migration (migrate_all_tasks)
 * happens before everything else.  This has to be lower priority than
4683
 * the notifier in the perf_event subsystem, though.
L
Linus Torvalds 已提交
4684
 */
4685
static struct notifier_block __cpuinitdata migration_notifier = {
L
Linus Torvalds 已提交
4686
	.notifier_call = migration_call,
4687
	.priority = CPU_PRI_MIGRATION,
L
Linus Torvalds 已提交
4688 4689
};

4690 4691 4692 4693
static int __cpuinit sched_cpu_active(struct notifier_block *nfb,
				      unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
4694
	case CPU_STARTING:
4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714
	case CPU_DOWN_FAILED:
		set_cpu_active((long)hcpu, true);
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}

static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb,
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DOWN_PREPARE:
		set_cpu_active((long)hcpu, false);
		return NOTIFY_OK;
	default:
		return NOTIFY_DONE;
	}
}

4715
static int __init migration_init(void)
L
Linus Torvalds 已提交
4716 4717
{
	void *cpu = (void *)(long)smp_processor_id();
4718
	int err;
4719

4720
	/* Initialize migration for the boot CPU */
4721 4722
	err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
	BUG_ON(err == NOTIFY_BAD);
L
Linus Torvalds 已提交
4723 4724
	migration_call(&migration_notifier, CPU_ONLINE, cpu);
	register_cpu_notifier(&migration_notifier);
4725

4726 4727 4728 4729
	/* Register cpu active notifiers */
	cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
	cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);

4730
	return 0;
L
Linus Torvalds 已提交
4731
}
4732
early_initcall(migration_init);
L
Linus Torvalds 已提交
4733 4734 4735
#endif

#ifdef CONFIG_SMP
4736

4737 4738
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

4739
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
4740

4741
static __read_mostly int sched_debug_enabled;
4742

4743
static int __init sched_debug_setup(char *str)
4744
{
4745
	sched_debug_enabled = 1;
4746 4747 4748

	return 0;
}
4749 4750 4751 4752 4753 4754
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
4755

4756
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
4757
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
4758
{
I
Ingo Molnar 已提交
4759
	struct sched_group *group = sd->groups;
4760
	char str[256];
L
Linus Torvalds 已提交
4761

R
Rusty Russell 已提交
4762
	cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd));
4763
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
4764 4765 4766 4767

	printk(KERN_DEBUG "%*s domain %d: ", level, "", level);

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
4768
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
4769
		if (sd->parent)
P
Peter Zijlstra 已提交
4770 4771
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
4772
		return -1;
N
Nick Piggin 已提交
4773 4774
	}

P
Peter Zijlstra 已提交
4775
	printk(KERN_CONT "span %s level %s\n", str, sd->name);
I
Ingo Molnar 已提交
4776

4777
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
4778 4779
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
4780
	}
4781
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
4782 4783
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
4784
	}
L
Linus Torvalds 已提交
4785

I
Ingo Molnar 已提交
4786
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
4787
	do {
I
Ingo Molnar 已提交
4788
		if (!group) {
P
Peter Zijlstra 已提交
4789 4790
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
4791 4792 4793
			break;
		}

4794 4795 4796 4797 4798 4799
		/*
		 * Even though we initialize ->power to something semi-sane,
		 * we leave power_orig unset. This allows us to detect if
		 * domain iteration is still funny without causing /0 traps.
		 */
		if (!group->sgp->power_orig) {
P
Peter Zijlstra 已提交
4800 4801 4802
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: domain->cpu_power not "
					"set\n");
I
Ingo Molnar 已提交
4803 4804
			break;
		}
L
Linus Torvalds 已提交
4805

4806
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
4807 4808
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
4809 4810
			break;
		}
L
Linus Torvalds 已提交
4811

4812 4813
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
4814 4815
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
4816 4817
			break;
		}
L
Linus Torvalds 已提交
4818

4819
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
4820

R
Rusty Russell 已提交
4821
		cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group));
4822

P
Peter Zijlstra 已提交
4823
		printk(KERN_CONT " %s", str);
4824
		if (group->sgp->power != SCHED_POWER_SCALE) {
P
Peter Zijlstra 已提交
4825
			printk(KERN_CONT " (cpu_power = %d)",
4826
				group->sgp->power);
4827
		}
L
Linus Torvalds 已提交
4828

I
Ingo Molnar 已提交
4829 4830
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
4831
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
4832

4833
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
4834
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
4835

4836 4837
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
4838 4839
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
4840 4841
	return 0;
}
L
Linus Torvalds 已提交
4842

I
Ingo Molnar 已提交
4843 4844 4845
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
4846

4847
	if (!sched_debug_enabled)
4848 4849
		return;

I
Ingo Molnar 已提交
4850 4851 4852 4853
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
4854

I
Ingo Molnar 已提交
4855 4856 4857
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
4858
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
4859
			break;
L
Linus Torvalds 已提交
4860 4861
		level++;
		sd = sd->parent;
4862
		if (!sd)
I
Ingo Molnar 已提交
4863 4864
			break;
	}
L
Linus Torvalds 已提交
4865
}
4866
#else /* !CONFIG_SCHED_DEBUG */
4867
# define sched_domain_debug(sd, cpu) do { } while (0)
4868 4869 4870 4871
static inline bool sched_debug(void)
{
	return false;
}
4872
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
4873

4874
static int sd_degenerate(struct sched_domain *sd)
4875
{
4876
	if (cpumask_weight(sched_domain_span(sd)) == 1)
4877 4878 4879 4880 4881 4882
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
4883 4884 4885
			 SD_BALANCE_EXEC |
			 SD_SHARE_CPUPOWER |
			 SD_SHARE_PKG_RESOURCES)) {
4886 4887 4888 4889 4890
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
4891
	if (sd->flags & (SD_WAKE_AFFINE))
4892 4893 4894 4895 4896
		return 0;

	return 1;
}

4897 4898
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
4899 4900 4901 4902 4903 4904
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

4905
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
4906 4907 4908 4909 4910 4911 4912
		return 0;

	/* Flags needing groups don't count if only 1 group in parent */
	if (parent->groups == parent->groups->next) {
		pflags &= ~(SD_LOAD_BALANCE |
				SD_BALANCE_NEWIDLE |
				SD_BALANCE_FORK |
4913 4914 4915
				SD_BALANCE_EXEC |
				SD_SHARE_CPUPOWER |
				SD_SHARE_PKG_RESOURCES);
4916 4917
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
4918 4919 4920 4921 4922 4923 4924
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

4925
static void free_rootdomain(struct rcu_head *rcu)
4926
{
4927
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
4928

4929
	cpupri_cleanup(&rd->cpupri);
4930 4931 4932 4933 4934 4935
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
4936 4937
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
4938
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
4939 4940
	unsigned long flags;

4941
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
4942 4943

	if (rq->rd) {
I
Ingo Molnar 已提交
4944
		old_rd = rq->rd;
G
Gregory Haskins 已提交
4945

4946
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
4947
			set_rq_offline(rq);
G
Gregory Haskins 已提交
4948

4949
		cpumask_clear_cpu(rq->cpu, old_rd->span);
4950

I
Ingo Molnar 已提交
4951 4952 4953 4954 4955 4956 4957
		/*
		 * If we dont want to free the old_rt yet then
		 * set old_rd to NULL to skip the freeing later
		 * in this function:
		 */
		if (!atomic_dec_and_test(&old_rd->refcount))
			old_rd = NULL;
G
Gregory Haskins 已提交
4958 4959 4960 4961 4962
	}

	atomic_inc(&rd->refcount);
	rq->rd = rd;

4963
	cpumask_set_cpu(rq->cpu, rd->span);
4964
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
4965
		set_rq_online(rq);
G
Gregory Haskins 已提交
4966

4967
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
4968 4969

	if (old_rd)
4970
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
4971 4972
}

4973
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
4974 4975 4976
{
	memset(rd, 0, sizeof(*rd));

4977
	if (!alloc_cpumask_var(&rd->span, GFP_KERNEL))
4978
		goto out;
4979
	if (!alloc_cpumask_var(&rd->online, GFP_KERNEL))
4980
		goto free_span;
4981
	if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
4982
		goto free_online;
4983

4984
	if (cpupri_init(&rd->cpupri) != 0)
4985
		goto free_rto_mask;
4986
	return 0;
4987

4988 4989
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
4990 4991 4992 4993
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
4994
out:
4995
	return -ENOMEM;
G
Gregory Haskins 已提交
4996 4997
}

4998 4999 5000 5001 5002 5003
/*
 * By default the system creates a single root-domain with all cpus as
 * members (mimicking the global state we have today).
 */
struct root_domain def_root_domain;

G
Gregory Haskins 已提交
5004 5005
static void init_defrootdomain(void)
{
5006
	init_rootdomain(&def_root_domain);
5007

G
Gregory Haskins 已提交
5008 5009 5010
	atomic_set(&def_root_domain.refcount, 1);
}

5011
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5012 5013 5014 5015 5016 5017 5018
{
	struct root_domain *rd;

	rd = kmalloc(sizeof(*rd), GFP_KERNEL);
	if (!rd)
		return NULL;

5019
	if (init_rootdomain(rd) != 0) {
5020 5021 5022
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5023 5024 5025 5026

	return rd;
}

5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045
static void free_sched_groups(struct sched_group *sg, int free_sgp)
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

	first = sg;
	do {
		tmp = sg->next;

		if (free_sgp && atomic_dec_and_test(&sg->sgp->ref))
			kfree(sg->sgp);

		kfree(sg);
		sg = tmp;
	} while (sg != first);
}

5046 5047 5048
static void free_sched_domain(struct rcu_head *rcu)
{
	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
5049 5050 5051 5052 5053 5054 5055 5056

	/*
	 * If its an overlapping domain it has private groups, iterate and
	 * nuke them all.
	 */
	if (sd->flags & SD_OVERLAP) {
		free_sched_groups(sd->groups, 1);
	} else if (atomic_dec_and_test(&sd->groups->ref)) {
5057
		kfree(sd->groups->sgp);
5058
		kfree(sd->groups);
5059
	}
5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073
	kfree(sd);
}

static void destroy_sched_domain(struct sched_domain *sd, int cpu)
{
	call_rcu(&sd->rcu, free_sched_domain);
}

static void destroy_sched_domains(struct sched_domain *sd, int cpu)
{
	for (; sd; sd = sd->parent)
		destroy_sched_domain(sd, cpu);
}

5074 5075 5076 5077 5078 5079 5080
/*
 * Keep a special pointer to the highest sched_domain that has
 * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
 * allows us to avoid some pointer chasing select_idle_sibling().
 *
 * Also keep a unique ID per domain (we use the first cpu number in
 * the cpumask of the domain), this allows us to quickly tell if
5081
 * two cpus are in the same cache domain, see cpus_share_cache().
5082 5083 5084 5085 5086 5087 5088 5089 5090 5091
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
DEFINE_PER_CPU(int, sd_llc_id);

static void update_top_cache_domain(int cpu)
{
	struct sched_domain *sd;
	int id = cpu;

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
5092
	if (sd)
5093 5094 5095 5096 5097 5098
		id = cpumask_first(sched_domain_span(sd));

	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
	per_cpu(sd_llc_id, cpu) = id;
}

L
Linus Torvalds 已提交
5099
/*
I
Ingo Molnar 已提交
5100
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
5101 5102
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
5103 5104
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
5105
{
5106
	struct rq *rq = cpu_rq(cpu);
5107 5108 5109
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
5110
	for (tmp = sd; tmp; ) {
5111 5112 5113
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
5114

5115
		if (sd_parent_degenerate(tmp, parent)) {
5116
			tmp->parent = parent->parent;
5117 5118
			if (parent->parent)
				parent->parent->child = tmp;
5119
			destroy_sched_domain(parent, cpu);
5120 5121
		} else
			tmp = tmp->parent;
5122 5123
	}

5124
	if (sd && sd_degenerate(sd)) {
5125
		tmp = sd;
5126
		sd = sd->parent;
5127
		destroy_sched_domain(tmp, cpu);
5128 5129 5130
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
5131

5132
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
5133

G
Gregory Haskins 已提交
5134
	rq_attach_root(rq, rd);
5135
	tmp = rq->sd;
N
Nick Piggin 已提交
5136
	rcu_assign_pointer(rq->sd, sd);
5137
	destroy_sched_domains(tmp, cpu);
5138 5139

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
5140 5141 5142
}

/* cpus with isolated domains */
5143
static cpumask_var_t cpu_isolated_map;
L
Linus Torvalds 已提交
5144 5145 5146 5147

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
R
Rusty Russell 已提交
5148
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
R
Rusty Russell 已提交
5149
	cpulist_parse(str, cpu_isolated_map);
L
Linus Torvalds 已提交
5150 5151 5152
	return 1;
}

I
Ingo Molnar 已提交
5153
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
5154

5155 5156 5157 5158 5159
static const struct cpumask *cpu_cpu_mask(int cpu)
{
	return cpumask_of_node(cpu_to_node(cpu));
}

5160 5161 5162
struct sd_data {
	struct sched_domain **__percpu sd;
	struct sched_group **__percpu sg;
5163
	struct sched_group_power **__percpu sgp;
5164 5165
};

5166
struct s_data {
5167
	struct sched_domain ** __percpu sd;
5168 5169 5170
	struct root_domain	*rd;
};

5171 5172
enum s_alloc {
	sa_rootdomain,
5173
	sa_sd,
5174
	sa_sd_storage,
5175 5176 5177
	sa_none,
};

5178 5179 5180
struct sched_domain_topology_level;

typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu);
5181 5182
typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);

5183 5184
#define SDTL_OVERLAP	0x01

5185
struct sched_domain_topology_level {
5186 5187
	sched_domain_init_f init;
	sched_domain_mask_f mask;
5188
	int		    flags;
5189
	int		    numa_level;
5190
	struct sd_data      data;
5191 5192
};

P
Peter Zijlstra 已提交
5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230
/*
 * Build an iteration mask that can exclude certain CPUs from the upwards
 * domain traversal.
 *
 * Asymmetric node setups can result in situations where the domain tree is of
 * unequal depth, make sure to skip domains that already cover the entire
 * range.
 *
 * In that case build_sched_domains() will have terminated the iteration early
 * and our sibling sd spans will be empty. Domains should always include the
 * cpu they're built on, so check that.
 *
 */
static void build_group_mask(struct sched_domain *sd, struct sched_group *sg)
{
	const struct cpumask *span = sched_domain_span(sd);
	struct sd_data *sdd = sd->private;
	struct sched_domain *sibling;
	int i;

	for_each_cpu(i, span) {
		sibling = *per_cpu_ptr(sdd->sd, i);
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
			continue;

		cpumask_set_cpu(i, sched_group_mask(sg));
	}
}

/*
 * Return the canonical balance cpu for this group, this is the first cpu
 * of this group that's also in the iteration mask.
 */
int group_balance_cpu(struct sched_group *sg)
{
	return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg));
}

5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248
static int
build_overlap_sched_groups(struct sched_domain *sd, int cpu)
{
	struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg;
	const struct cpumask *span = sched_domain_span(sd);
	struct cpumask *covered = sched_domains_tmpmask;
	struct sd_data *sdd = sd->private;
	struct sched_domain *child;
	int i;

	cpumask_clear(covered);

	for_each_cpu(i, span) {
		struct cpumask *sg_span;

		if (cpumask_test_cpu(i, covered))
			continue;

P
Peter Zijlstra 已提交
5249 5250 5251 5252 5253 5254
		child = *per_cpu_ptr(sdd->sd, i);

		/* See the comment near build_group_mask(). */
		if (!cpumask_test_cpu(i, sched_domain_span(child)))
			continue;

5255
		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
5256
				GFP_KERNEL, cpu_to_node(cpu));
5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269

		if (!sg)
			goto fail;

		sg_span = sched_group_cpus(sg);
		if (child->child) {
			child = child->child;
			cpumask_copy(sg_span, sched_domain_span(child));
		} else
			cpumask_set_cpu(i, sg_span);

		cpumask_or(covered, covered, sg_span);

P
Peter Zijlstra 已提交
5270
		sg->sgp = *per_cpu_ptr(sdd->sgp, i);
P
Peter Zijlstra 已提交
5271 5272 5273
		if (atomic_inc_return(&sg->sgp->ref) == 1)
			build_group_mask(sd, sg);

5274 5275 5276 5277 5278 5279
		/*
		 * Initialize sgp->power such that even if we mess up the
		 * domains and no possible iteration will get us here, we won't
		 * die on a /0 trap.
		 */
		sg->sgp->power = SCHED_POWER_SCALE * cpumask_weight(sg_span);
5280

P
Peter Zijlstra 已提交
5281 5282 5283 5284 5285
		/*
		 * Make sure the first group of this domain contains the
		 * canonical balance cpu. Otherwise the sched_domain iteration
		 * breaks. See update_sg_lb_stats().
		 */
P
Peter Zijlstra 已提交
5286
		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
P
Peter Zijlstra 已提交
5287
		    group_balance_cpu(sg) == cpu)
5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306
			groups = sg;

		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
		last->next = first;
	}
	sd->groups = groups;

	return 0;

fail:
	free_sched_groups(first, 0);

	return -ENOMEM;
}

5307
static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
L
Linus Torvalds 已提交
5308
{
5309 5310
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	struct sched_domain *child = sd->child;
L
Linus Torvalds 已提交
5311

5312 5313
	if (child)
		cpu = cpumask_first(sched_domain_span(child));
5314

5315
	if (sg) {
5316
		*sg = *per_cpu_ptr(sdd->sg, cpu);
5317
		(*sg)->sgp = *per_cpu_ptr(sdd->sgp, cpu);
5318
		atomic_set(&(*sg)->sgp->ref, 1); /* for claim_allocations */
5319
	}
5320 5321

	return cpu;
5322 5323
}

5324
/*
5325 5326 5327
 * build_sched_groups will build a circular linked list of the groups
 * covered by the given span, and will set each group's ->cpumask correctly,
 * and ->cpu_power to 0.
5328 5329
 *
 * Assumes the sched_domain tree is fully constructed
5330
 */
5331 5332
static int
build_sched_groups(struct sched_domain *sd, int cpu)
L
Linus Torvalds 已提交
5333
{
5334 5335 5336
	struct sched_group *first = NULL, *last = NULL;
	struct sd_data *sdd = sd->private;
	const struct cpumask *span = sched_domain_span(sd);
5337
	struct cpumask *covered;
5338
	int i;
5339

5340 5341 5342
	get_group(cpu, sdd, &sd->groups);
	atomic_inc(&sd->groups->ref);

5343
	if (cpu != cpumask_first(span))
5344 5345
		return 0;

5346 5347 5348
	lockdep_assert_held(&sched_domains_mutex);
	covered = sched_domains_tmpmask;

5349
	cpumask_clear(covered);
5350

5351 5352
	for_each_cpu(i, span) {
		struct sched_group *sg;
5353
		int group, j;
5354

5355 5356
		if (cpumask_test_cpu(i, covered))
			continue;
5357

5358
		group = get_group(i, sdd, &sg);
5359
		cpumask_clear(sched_group_cpus(sg));
5360
		sg->sgp->power = 0;
P
Peter Zijlstra 已提交
5361
		cpumask_setall(sched_group_mask(sg));
5362

5363 5364 5365
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
				continue;
5366

5367 5368 5369
			cpumask_set_cpu(j, covered);
			cpumask_set_cpu(j, sched_group_cpus(sg));
		}
5370

5371 5372 5373 5374 5375 5376 5377
		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
	}
	last->next = first;
5378 5379

	return 0;
5380
}
5381

5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393
/*
 * Initialize sched groups cpu_power.
 *
 * cpu_power indicates the capacity of sched group, which is used while
 * distributing the load between different sched groups in a sched domain.
 * Typically cpu_power for all the groups in a sched domain will be same unless
 * there are asymmetries in the topology. If there are asymmetries, group
 * having more cpu_power will pickup more load compared to the group having
 * less cpu_power.
 */
static void init_sched_groups_power(int cpu, struct sched_domain *sd)
{
5394
	struct sched_group *sg = sd->groups;
5395

5396
	WARN_ON(!sg);
5397 5398 5399 5400 5401

	do {
		sg->group_weight = cpumask_weight(sched_group_cpus(sg));
		sg = sg->next;
	} while (sg != sd->groups);
5402

P
Peter Zijlstra 已提交
5403
	if (cpu != group_balance_cpu(sg))
5404
		return;
5405

5406
	update_group_power(sd, cpu);
5407
	atomic_set(&sg->sgp->nr_busy_cpus, sg->group_weight);
5408 5409
}

5410 5411 5412
int __weak arch_sd_sibling_asym_packing(void)
{
       return 0*SD_ASYM_PACKING;
5413 5414
}

5415 5416 5417 5418 5419
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

5420 5421 5422 5423 5424 5425
#ifdef CONFIG_SCHED_DEBUG
# define SD_INIT_NAME(sd, type)		sd->name = #type
#else
# define SD_INIT_NAME(sd, type)		do { } while (0)
#endif

5426 5427 5428 5429 5430 5431 5432 5433 5434
#define SD_INIT_FUNC(type)						\
static noinline struct sched_domain *					\
sd_init_##type(struct sched_domain_topology_level *tl, int cpu) 	\
{									\
	struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);	\
	*sd = SD_##type##_INIT;						\
	SD_INIT_NAME(sd, type);						\
	sd->private = &tl->data;					\
	return sd;							\
5435 5436 5437 5438 5439 5440 5441 5442 5443
}

SD_INIT_FUNC(CPU)
#ifdef CONFIG_SCHED_SMT
 SD_INIT_FUNC(SIBLING)
#endif
#ifdef CONFIG_SCHED_MC
 SD_INIT_FUNC(MC)
#endif
5444 5445 5446
#ifdef CONFIG_SCHED_BOOK
 SD_INIT_FUNC(BOOK)
#endif
5447

5448
static int default_relax_domain_level = -1;
5449
int sched_domain_level_max;
5450 5451 5452

static int __init setup_relax_domain_level(char *str)
{
5453 5454
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
5455

5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473
	return 1;
}
__setup("relax_domain_level=", setup_relax_domain_level);

static void set_domain_attribute(struct sched_domain *sd,
				 struct sched_domain_attr *attr)
{
	int request;

	if (!attr || attr->relax_domain_level < 0) {
		if (default_relax_domain_level < 0)
			return;
		else
			request = default_relax_domain_level;
	} else
		request = attr->relax_domain_level;
	if (request < sd->level) {
		/* turn off idle balance on this domain */
5474
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
5475 5476
	} else {
		/* turn on idle balance on this domain */
5477
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
5478 5479 5480
	}
}

5481 5482 5483
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

5484 5485 5486 5487 5488
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
5489 5490
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
5491 5492
	case sa_sd:
		free_percpu(d->sd); /* fall through */
5493
	case sa_sd_storage:
5494
		__sdt_free(cpu_map); /* fall through */
5495 5496 5497 5498
	case sa_none:
		break;
	}
}
5499

5500 5501 5502
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
5503 5504
	memset(d, 0, sizeof(*d));

5505 5506
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
5507 5508 5509
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
5510
	d->rd = alloc_rootdomain();
5511
	if (!d->rd)
5512
		return sa_sd;
5513 5514
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
5515

5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527
/*
 * NULL the sd_data elements we've used to build the sched_domain and
 * sched_group structure so that the subsequent __free_domain_allocs()
 * will not free the data we're using.
 */
static void claim_allocations(int cpu, struct sched_domain *sd)
{
	struct sd_data *sdd = sd->private;

	WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd);
	*per_cpu_ptr(sdd->sd, cpu) = NULL;

5528
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
5529
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
5530 5531

	if (atomic_read(&(*per_cpu_ptr(sdd->sgp, cpu))->ref))
5532
		*per_cpu_ptr(sdd->sgp, cpu) = NULL;
5533 5534
}

5535 5536
#ifdef CONFIG_SCHED_SMT
static const struct cpumask *cpu_smt_mask(int cpu)
5537
{
5538
	return topology_thread_cpumask(cpu);
5539
}
5540
#endif
5541

5542 5543 5544
/*
 * Topology list, bottom-up.
 */
5545
static struct sched_domain_topology_level default_topology[] = {
5546 5547
#ifdef CONFIG_SCHED_SMT
	{ sd_init_SIBLING, cpu_smt_mask, },
5548
#endif
5549
#ifdef CONFIG_SCHED_MC
5550
	{ sd_init_MC, cpu_coregroup_mask, },
5551
#endif
5552 5553 5554 5555
#ifdef CONFIG_SCHED_BOOK
	{ sd_init_BOOK, cpu_book_mask, },
#endif
	{ sd_init_CPU, cpu_cpu_mask, },
5556 5557 5558 5559 5560
	{ NULL, },
};

static struct sched_domain_topology_level *sched_domain_topology = default_topology;

5561 5562 5563
#define for_each_sd_topology(tl)			\
	for (tl = sched_domain_topology; tl->init; tl++)

5564 5565 5566 5567 5568 5569 5570 5571 5572
#ifdef CONFIG_NUMA

static int sched_domains_numa_levels;
static int *sched_domains_numa_distance;
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;

static inline int sd_local_flags(int level)
{
5573
	if (sched_domains_numa_distance[level] > RECLAIM_DISTANCE)
5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590
		return 0;

	return SD_BALANCE_EXEC | SD_BALANCE_FORK | SD_WAKE_AFFINE;
}

static struct sched_domain *
sd_numa_init(struct sched_domain_topology_level *tl, int cpu)
{
	struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);
	int level = tl->numa_level;
	int sd_weight = cpumask_weight(
			sched_domains_numa_masks[level][cpu_to_node(cpu)]);

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
5591
		.imbalance_pct		= 125,
5592 5593 5594 5595 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
		.cache_nice_tries	= 2,
		.busy_idx		= 3,
		.idle_idx		= 2,
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
					| 0*SD_BALANCE_EXEC
					| 0*SD_BALANCE_FORK
					| 0*SD_BALANCE_WAKE
					| 0*SD_WAKE_AFFINE
					| 0*SD_SHARE_CPUPOWER
					| 0*SD_SHARE_PKG_RESOURCES
					| 1*SD_SERIALIZE
					| 0*SD_PREFER_SIBLING
					| sd_local_flags(level)
					,
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
	};
	SD_INIT_NAME(sd, NUMA);
	sd->private = &tl->data;

	/*
	 * Ugly hack to pass state to sd_numa_mask()...
	 */
	sched_domains_curr_level = tl->numa_level;

	return sd;
}

static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

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
static void sched_numa_warn(const char *str)
{
	static int done = false;
	int i,j;

	if (done)
		return;

	done = true;

	printk(KERN_WARNING "ERROR: %s\n\n", str);

	for (i = 0; i < nr_node_ids; i++) {
		printk(KERN_WARNING "  ");
		for (j = 0; j < nr_node_ids; j++)
			printk(KERN_CONT "%02d ", node_distance(i,j));
		printk(KERN_CONT "\n");
	}
	printk(KERN_WARNING "\n");
}

static bool find_numa_distance(int distance)
{
	int i;

	if (distance == node_distance(0, 0))
		return true;

	for (i = 0; i < sched_domains_numa_levels; i++) {
		if (sched_domains_numa_distance[i] == distance)
			return true;
	}

	return false;
}

5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686
static void sched_init_numa(void)
{
	int next_distance, curr_distance = node_distance(0, 0);
	struct sched_domain_topology_level *tl;
	int level = 0;
	int i, j, k;

	sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL);
	if (!sched_domains_numa_distance)
		return;

	/*
	 * O(nr_nodes^2) deduplicating selection sort -- in order to find the
	 * unique distances in the node_distance() table.
	 *
	 * Assumes node_distance(0,j) includes all distances in
	 * node_distance(i,j) in order to avoid cubic time.
	 */
	next_distance = curr_distance;
	for (i = 0; i < nr_node_ids; i++) {
		for (j = 0; j < nr_node_ids; j++) {
5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710
			for (k = 0; k < nr_node_ids; k++) {
				int distance = node_distance(i, k);

				if (distance > curr_distance &&
				    (distance < next_distance ||
				     next_distance == curr_distance))
					next_distance = distance;

				/*
				 * While not a strong assumption it would be nice to know
				 * about cases where if node A is connected to B, B is not
				 * equally connected to A.
				 */
				if (sched_debug() && node_distance(k, i) != distance)
					sched_numa_warn("Node-distance not symmetric");

				if (sched_debug() && i && !find_numa_distance(distance))
					sched_numa_warn("Node-0 not representative");
			}
			if (next_distance != curr_distance) {
				sched_domains_numa_distance[level++] = next_distance;
				sched_domains_numa_levels = level;
				curr_distance = next_distance;
			} else break;
5711
		}
5712 5713 5714 5715 5716 5717

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
5718 5719 5720 5721 5722
	}
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
5723
	 * The sched_domains_numa_distance[] array includes the actual distance
5724 5725 5726
	 * numbers.
	 */

5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737
	/*
	 * Here, we should temporarily reset sched_domains_numa_levels to 0.
	 * If it fails to allocate memory for array sched_domains_numa_masks[][],
	 * the array will contain less then 'level' members. This could be
	 * dangerous when we use it to iterate array sched_domains_numa_masks[][]
	 * in other functions.
	 *
	 * We reset it to 'level' at the end of this function.
	 */
	sched_domains_numa_levels = 0;

5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752
	sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL);
	if (!sched_domains_numa_masks)
		return;

	/*
	 * Now for each level, construct a mask per node which contains all
	 * cpus of nodes that are that many hops away from us.
	 */
	for (i = 0; i < level; i++) {
		sched_domains_numa_masks[i] =
			kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL);
		if (!sched_domains_numa_masks[i])
			return;

		for (j = 0; j < nr_node_ids; j++) {
5753
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
5754 5755 5756 5757 5758 5759
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

			for (k = 0; k < nr_node_ids; k++) {
5760
				if (node_distance(j, k) > sched_domains_numa_distance[i])
5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791
					continue;

				cpumask_or(mask, mask, cpumask_of_node(k));
			}
		}
	}

	tl = kzalloc((ARRAY_SIZE(default_topology) + level) *
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
	for (i = 0; default_topology[i].init; i++)
		tl[i] = default_topology[i];

	/*
	 * .. and append 'j' levels of NUMA goodness.
	 */
	for (j = 0; j < level; i++, j++) {
		tl[i] = (struct sched_domain_topology_level){
			.init = sd_numa_init,
			.mask = sd_numa_mask,
			.flags = SDTL_OVERLAP,
			.numa_level = j,
		};
	}

	sched_domain_topology = tl;
5792 5793

	sched_domains_numa_levels = level;
5794
}
5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841

static void sched_domains_numa_masks_set(int cpu)
{
	int i, j;
	int node = cpu_to_node(cpu);

	for (i = 0; i < sched_domains_numa_levels; i++) {
		for (j = 0; j < nr_node_ids; j++) {
			if (node_distance(j, node) <= sched_domains_numa_distance[i])
				cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
		}
	}
}

static void sched_domains_numa_masks_clear(int cpu)
{
	int i, j;
	for (i = 0; i < sched_domains_numa_levels; i++) {
		for (j = 0; j < nr_node_ids; j++)
			cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]);
	}
}

/*
 * Update sched_domains_numa_masks[level][node] array when new cpus
 * are onlined.
 */
static int sched_domains_numa_masks_update(struct notifier_block *nfb,
					   unsigned long action,
					   void *hcpu)
{
	int cpu = (long)hcpu;

	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_ONLINE:
		sched_domains_numa_masks_set(cpu);
		break;

	case CPU_DEAD:
		sched_domains_numa_masks_clear(cpu);
		break;

	default:
		return NOTIFY_DONE;
	}

	return NOTIFY_OK;
5842 5843 5844 5845 5846
}
#else
static inline void sched_init_numa(void)
{
}
5847 5848 5849 5850 5851 5852 5853

static int sched_domains_numa_masks_update(struct notifier_block *nfb,
					   unsigned long action,
					   void *hcpu)
{
	return 0;
}
5854 5855
#endif /* CONFIG_NUMA */

5856 5857 5858 5859 5860
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

5861
	for_each_sd_topology(tl) {
5862 5863 5864 5865 5866 5867 5868 5869 5870 5871
		struct sd_data *sdd = &tl->data;

		sdd->sd = alloc_percpu(struct sched_domain *);
		if (!sdd->sd)
			return -ENOMEM;

		sdd->sg = alloc_percpu(struct sched_group *);
		if (!sdd->sg)
			return -ENOMEM;

5872 5873 5874 5875
		sdd->sgp = alloc_percpu(struct sched_group_power *);
		if (!sdd->sgp)
			return -ENOMEM;

5876 5877 5878
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
			struct sched_group *sg;
5879
			struct sched_group_power *sgp;
5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892

		       	sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
					GFP_KERNEL, cpu_to_node(j));
			if (!sd)
				return -ENOMEM;

			*per_cpu_ptr(sdd->sd, j) = sd;

			sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
					GFP_KERNEL, cpu_to_node(j));
			if (!sg)
				return -ENOMEM;

5893 5894
			sg->next = sg;

5895
			*per_cpu_ptr(sdd->sg, j) = sg;
5896

P
Peter Zijlstra 已提交
5897
			sgp = kzalloc_node(sizeof(struct sched_group_power) + cpumask_size(),
5898 5899 5900 5901 5902
					GFP_KERNEL, cpu_to_node(j));
			if (!sgp)
				return -ENOMEM;

			*per_cpu_ptr(sdd->sgp, j) = sgp;
5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913
		}
	}

	return 0;
}

static void __sdt_free(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

5914
	for_each_sd_topology(tl) {
5915 5916 5917
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930
			struct sched_domain *sd;

			if (sdd->sd) {
				sd = *per_cpu_ptr(sdd->sd, j);
				if (sd && (sd->flags & SD_OVERLAP))
					free_sched_groups(sd->groups, 0);
				kfree(*per_cpu_ptr(sdd->sd, j));
			}

			if (sdd->sg)
				kfree(*per_cpu_ptr(sdd->sg, j));
			if (sdd->sgp)
				kfree(*per_cpu_ptr(sdd->sgp, j));
5931 5932
		}
		free_percpu(sdd->sd);
5933
		sdd->sd = NULL;
5934
		free_percpu(sdd->sg);
5935
		sdd->sg = NULL;
5936
		free_percpu(sdd->sgp);
5937
		sdd->sgp = NULL;
5938 5939 5940
	}
}

5941
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
5942 5943
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
5944
{
5945
	struct sched_domain *sd = tl->init(tl, cpu);
5946
	if (!sd)
5947
		return child;
5948 5949

	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
5950 5951 5952
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
5953
		child->parent = sd;
5954
		sd->child = child;
5955
	}
5956
	set_domain_attribute(sd, attr);
5957 5958 5959 5960

	return sd;
}

5961 5962 5963 5964
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
5965 5966
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
5967
{
5968
	enum s_alloc alloc_state;
5969
	struct sched_domain *sd;
5970
	struct s_data d;
5971
	int i, ret = -ENOMEM;
5972

5973 5974 5975
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
5976

5977
	/* Set up domains for cpus specified by the cpu_map. */
5978
	for_each_cpu(i, cpu_map) {
5979 5980
		struct sched_domain_topology_level *tl;

5981
		sd = NULL;
5982
		for_each_sd_topology(tl) {
5983
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
5984 5985
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
5986 5987
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
5988 5989
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
5990
		}
5991 5992 5993 5994 5995 5996
	}

	/* Build the groups for the domains */
	for_each_cpu(i, cpu_map) {
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			sd->span_weight = cpumask_weight(sched_domain_span(sd));
5997 5998 5999 6000 6001 6002 6003
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
6004
		}
6005
	}
6006

L
Linus Torvalds 已提交
6007
	/* Calculate CPU power for physical packages and nodes */
6008 6009 6010
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
6011

6012 6013
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
6014
			init_sched_groups_power(i, sd);
6015
		}
6016
	}
6017

L
Linus Torvalds 已提交
6018
	/* Attach the domains */
6019
	rcu_read_lock();
6020
	for_each_cpu(i, cpu_map) {
6021
		sd = *per_cpu_ptr(d.sd, i);
6022
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
6023
	}
6024
	rcu_read_unlock();
6025

6026
	ret = 0;
6027
error:
6028
	__free_domain_allocs(&d, alloc_state, cpu_map);
6029
	return ret;
L
Linus Torvalds 已提交
6030
}
P
Paul Jackson 已提交
6031

6032
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
6033
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
6034 6035
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
6036 6037 6038

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
6039 6040
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
6041
 */
6042
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
6043

6044 6045 6046 6047 6048 6049
/*
 * arch_update_cpu_topology lets virtualized architectures update the
 * cpu core maps. It is supposed to return 1 if the topology changed
 * or 0 if it stayed the same.
 */
int __attribute__((weak)) arch_update_cpu_topology(void)
6050
{
6051
	return 0;
6052 6053
}

6054 6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078
cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
{
	int i;
	cpumask_var_t *doms;

	doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
	if (!doms)
		return NULL;
	for (i = 0; i < ndoms; i++) {
		if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
			free_sched_domains(doms, i);
			return NULL;
		}
	}
	return doms;
}

void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
{
	unsigned int i;
	for (i = 0; i < ndoms; i++)
		free_cpumask_var(doms[i]);
	kfree(doms);
}

6079
/*
I
Ingo Molnar 已提交
6080
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
6081 6082
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
6083
 */
6084
static int init_sched_domains(const struct cpumask *cpu_map)
6085
{
6086 6087
	int err;

6088
	arch_update_cpu_topology();
P
Paul Jackson 已提交
6089
	ndoms_cur = 1;
6090
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
6091
	if (!doms_cur)
6092 6093
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
6094
	err = build_sched_domains(doms_cur[0], NULL);
6095
	register_sched_domain_sysctl();
6096 6097

	return err;
6098 6099 6100 6101 6102 6103
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
6104
static void detach_destroy_domains(const struct cpumask *cpu_map)
6105 6106 6107
{
	int i;

6108
	rcu_read_lock();
6109
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
6110
		cpu_attach_domain(NULL, &def_root_domain, i);
6111
	rcu_read_unlock();
6112 6113
}

6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129
/* handle null as "default" */
static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
			struct sched_domain_attr *new, int idx_new)
{
	struct sched_domain_attr tmp;

	/* fast path */
	if (!new && !cur)
		return 1;

	tmp = SD_ATTR_INIT;
	return !memcmp(cur ? (cur + idx_cur) : &tmp,
			new ? (new + idx_new) : &tmp,
			sizeof(struct sched_domain_attr));
}

P
Paul Jackson 已提交
6130 6131
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
6132
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
6133 6134 6135
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
6136
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
6137 6138 6139
 * The masks don't intersect (don't overlap.) We should setup one
 * sched domain for each mask. CPUs not in any of the cpumasks will
 * not be load balanced. If the same cpumask appears both in the
P
Paul Jackson 已提交
6140 6141 6142
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
6143 6144 6145 6146 6147 6148
 * The passed in 'doms_new' should be allocated using
 * alloc_sched_domains.  This routine takes ownership of it and will
 * free_sched_domains it when done with it. If the caller failed the
 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
 * and partition_sched_domains() will fallback to the single partition
 * 'fallback_doms', it also forces the domains to be rebuilt.
P
Paul Jackson 已提交
6149
 *
6150
 * If doms_new == NULL it will be replaced with cpu_online_mask.
6151 6152
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
6153
 *
P
Paul Jackson 已提交
6154 6155
 * Call with hotplug lock held
 */
6156
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
6157
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
6158
{
6159
	int i, j, n;
6160
	int new_topology;
P
Paul Jackson 已提交
6161

6162
	mutex_lock(&sched_domains_mutex);
6163

6164 6165 6166
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

6167 6168 6169
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

6170
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
6171 6172 6173

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
6174
		for (j = 0; j < n && !new_topology; j++) {
6175
			if (cpumask_equal(doms_cur[i], doms_new[j])
6176
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
6177 6178 6179
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
6180
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
6181 6182 6183 6184
match1:
		;
	}

6185 6186
	if (doms_new == NULL) {
		ndoms_cur = 0;
6187
		doms_new = &fallback_doms;
6188
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
6189
		WARN_ON_ONCE(dattr_new);
6190 6191
	}

P
Paul Jackson 已提交
6192 6193
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
6194
		for (j = 0; j < ndoms_cur && !new_topology; j++) {
6195
			if (cpumask_equal(doms_new[i], doms_cur[j])
6196
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
6197 6198 6199
				goto match2;
		}
		/* no match - add a new doms_new */
6200
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
6201 6202 6203 6204 6205
match2:
		;
	}

	/* Remember the new sched domains */
6206 6207
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
6208
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
6209
	doms_cur = doms_new;
6210
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
6211
	ndoms_cur = ndoms_new;
6212 6213

	register_sched_domain_sysctl();
6214

6215
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
6216 6217
}

6218 6219
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
6220
/*
6221 6222 6223
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
6224 6225 6226
 *
 * If we come here as part of a suspend/resume, don't touch cpusets because we
 * want to restore it back to its original state upon resume anyway.
L
Linus Torvalds 已提交
6227
 */
6228 6229
static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
			     void *hcpu)
6230
{
6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252
	switch (action) {
	case CPU_ONLINE_FROZEN:
	case CPU_DOWN_FAILED_FROZEN:

		/*
		 * num_cpus_frozen tracks how many CPUs are involved in suspend
		 * resume sequence. As long as this is not the last online
		 * operation in the resume sequence, just build a single sched
		 * domain, ignoring cpusets.
		 */
		num_cpus_frozen--;
		if (likely(num_cpus_frozen)) {
			partition_sched_domains(1, NULL, NULL);
			break;
		}

		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */

6253
	case CPU_ONLINE:
6254
	case CPU_DOWN_FAILED:
6255
		cpuset_update_active_cpus(true);
6256
		break;
6257 6258 6259
	default:
		return NOTIFY_DONE;
	}
6260
	return NOTIFY_OK;
6261
}
6262

6263 6264
static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
			       void *hcpu)
6265
{
6266
	switch (action) {
6267
	case CPU_DOWN_PREPARE:
6268
		cpuset_update_active_cpus(false);
6269 6270 6271 6272 6273
		break;
	case CPU_DOWN_PREPARE_FROZEN:
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
		break;
6274 6275 6276
	default:
		return NOTIFY_DONE;
	}
6277
	return NOTIFY_OK;
6278 6279
}

L
Linus Torvalds 已提交
6280 6281
void __init sched_init_smp(void)
{
6282 6283 6284
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
6285
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
6286

6287 6288
	sched_init_numa();

6289
	get_online_cpus();
6290
	mutex_lock(&sched_domains_mutex);
6291
	init_sched_domains(cpu_active_mask);
6292 6293 6294
	cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
	if (cpumask_empty(non_isolated_cpus))
		cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
6295
	mutex_unlock(&sched_domains_mutex);
6296
	put_online_cpus();
6297

6298
	hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE);
6299 6300
	hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
	hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
6301

6302
	init_hrtick();
6303 6304

	/* Move init over to a non-isolated CPU */
6305
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
6306
		BUG();
I
Ingo Molnar 已提交
6307
	sched_init_granularity();
6308
	free_cpumask_var(non_isolated_cpus);
6309

6310
	init_sched_rt_class();
L
Linus Torvalds 已提交
6311 6312 6313 6314
}
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
6315
	sched_init_granularity();
L
Linus Torvalds 已提交
6316 6317 6318
}
#endif /* CONFIG_SMP */

6319 6320
const_debug unsigned int sysctl_timer_migration = 1;

L
Linus Torvalds 已提交
6321 6322 6323 6324 6325 6326 6327
int in_sched_functions(unsigned long addr)
{
	return in_lock_functions(addr) ||
		(addr >= (unsigned long)__sched_text_start
		&& addr < (unsigned long)__sched_text_end);
}

6328
#ifdef CONFIG_CGROUP_SCHED
6329 6330 6331 6332
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
6333
struct task_group root_task_group;
6334
LIST_HEAD(task_groups);
6335
#endif
P
Peter Zijlstra 已提交
6336

6337
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
P
Peter Zijlstra 已提交
6338

L
Linus Torvalds 已提交
6339 6340
void __init sched_init(void)
{
I
Ingo Molnar 已提交
6341
	int i, j;
6342 6343 6344 6345 6346 6347 6348
	unsigned long alloc_size = 0, ptr;

#ifdef CONFIG_FAIR_GROUP_SCHED
	alloc_size += 2 * nr_cpu_ids * sizeof(void **);
#endif
#ifdef CONFIG_RT_GROUP_SCHED
	alloc_size += 2 * nr_cpu_ids * sizeof(void **);
6349
#endif
6350
#ifdef CONFIG_CPUMASK_OFFSTACK
6351
	alloc_size += num_possible_cpus() * cpumask_size();
6352 6353
#endif
	if (alloc_size) {
6354
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
6355 6356

#ifdef CONFIG_FAIR_GROUP_SCHED
6357
		root_task_group.se = (struct sched_entity **)ptr;
6358 6359
		ptr += nr_cpu_ids * sizeof(void **);

6360
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
6361
		ptr += nr_cpu_ids * sizeof(void **);
6362

6363
#endif /* CONFIG_FAIR_GROUP_SCHED */
6364
#ifdef CONFIG_RT_GROUP_SCHED
6365
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
6366 6367
		ptr += nr_cpu_ids * sizeof(void **);

6368
		root_task_group.rt_rq = (struct rt_rq **)ptr;
6369 6370
		ptr += nr_cpu_ids * sizeof(void **);

6371
#endif /* CONFIG_RT_GROUP_SCHED */
6372 6373
#ifdef CONFIG_CPUMASK_OFFSTACK
		for_each_possible_cpu(i) {
6374
			per_cpu(load_balance_mask, i) = (void *)ptr;
6375 6376 6377
			ptr += cpumask_size();
		}
#endif /* CONFIG_CPUMASK_OFFSTACK */
6378
	}
I
Ingo Molnar 已提交
6379

G
Gregory Haskins 已提交
6380 6381 6382 6383
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

6384 6385 6386 6387
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());

#ifdef CONFIG_RT_GROUP_SCHED
6388
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
6389
			global_rt_period(), global_rt_runtime());
6390
#endif /* CONFIG_RT_GROUP_SCHED */
6391

D
Dhaval Giani 已提交
6392
#ifdef CONFIG_CGROUP_SCHED
6393 6394
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
6395
	INIT_LIST_HEAD(&root_task_group.siblings);
6396
	autogroup_init(&init_task);
6397

D
Dhaval Giani 已提交
6398
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
6399

6400
	for_each_possible_cpu(i) {
6401
		struct rq *rq;
L
Linus Torvalds 已提交
6402 6403

		rq = cpu_rq(i);
6404
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
6405
		rq->nr_running = 0;
6406 6407
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
6408
		init_cfs_rq(&rq->cfs);
P
Peter Zijlstra 已提交
6409
		init_rt_rq(&rq->rt, rq);
I
Ingo Molnar 已提交
6410
#ifdef CONFIG_FAIR_GROUP_SCHED
6411
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
6412
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
D
Dhaval Giani 已提交
6413
		/*
6414
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
6415 6416 6417 6418
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
		 * gets 100% of the cpu resources in the system. This overall
		 * system cpu resource is divided among the tasks of
6419
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
6420 6421 6422
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
6423
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
6424 6425 6426
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
6427
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
6428
		 *
6429 6430
		 * We achieve this by letting root_task_group's tasks sit
		 * directly in rq->cfs (i.e root_task_group->se[] = NULL).
D
Dhaval Giani 已提交
6431
		 */
6432
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6433
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
6434 6435 6436
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6437
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6438
		INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
6439
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
6440
#endif
L
Linus Torvalds 已提交
6441

I
Ingo Molnar 已提交
6442 6443
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
6444 6445 6446

		rq->last_load_update_tick = jiffies;

L
Linus Torvalds 已提交
6447
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6448
		rq->sd = NULL;
G
Gregory Haskins 已提交
6449
		rq->rd = NULL;
6450
		rq->cpu_power = SCHED_POWER_SCALE;
6451
		rq->post_schedule = 0;
L
Linus Torvalds 已提交
6452
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6453
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6454
		rq->push_cpu = 0;
6455
		rq->cpu = i;
6456
		rq->online = 0;
6457 6458
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6459 6460 6461

		INIT_LIST_HEAD(&rq->cfs_tasks);

6462
		rq_attach_root(rq, &def_root_domain);
6463
#ifdef CONFIG_NO_HZ_COMMON
6464
		rq->nohz_flags = 0;
6465
#endif
6466 6467 6468
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
L
Linus Torvalds 已提交
6469
#endif
P
Peter Zijlstra 已提交
6470
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
6471 6472 6473
		atomic_set(&rq->nr_iowait, 0);
	}

6474
	set_load_weight(&init_task);
6475

6476 6477 6478 6479
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&init_task.preempt_notifiers);
#endif

6480
#ifdef CONFIG_RT_MUTEXES
6481
	plist_head_init(&init_task.pi_waiters);
6482 6483
#endif

L
Linus Torvalds 已提交
6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
	atomic_inc(&init_mm.mm_count);
	enter_lazy_tlb(&init_mm, current);

	/*
	 * Make us the idle thread. Technically, schedule() should not be
	 * called from this thread, however somewhere below it might be,
	 * but because we are the idle thread, we just pick up running again
	 * when this runqueue becomes "idle".
	 */
	init_idle(current, smp_processor_id());
6497 6498 6499

	calc_load_update = jiffies + LOAD_FREQ;

I
Ingo Molnar 已提交
6500 6501 6502 6503
	/*
	 * During early bootup we pretend to be a normal task:
	 */
	current->sched_class = &fair_sched_class;
6504

6505
#ifdef CONFIG_SMP
6506
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
6507 6508 6509
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
6510
	idle_thread_set_boot_cpu();
6511 6512
#endif
	init_sched_fair_class();
6513

6514
	scheduler_running = 1;
L
Linus Torvalds 已提交
6515 6516
}

6517
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6518 6519
static inline int preempt_count_equals(int preempt_offset)
{
6520
	int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth();
6521

A
Arnd Bergmann 已提交
6522
	return (nested == preempt_offset);
6523 6524
}

6525
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6526 6527 6528
{
	static unsigned long prev_jiffy;	/* ratelimiting */

6529
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
6530 6531
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) ||
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
6532 6533 6534 6535 6536
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

P
Peter Zijlstra 已提交
6537 6538 6539 6540 6541 6542 6543
	printk(KERN_ERR
		"BUG: sleeping function called from invalid context at %s:%d\n",
			file, line);
	printk(KERN_ERR
		"in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n",
			in_atomic(), irqs_disabled(),
			current->pid, current->comm);
I
Ingo Molnar 已提交
6544 6545 6546 6547 6548

	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
	dump_stack();
L
Linus Torvalds 已提交
6549 6550 6551 6552 6553
}
EXPORT_SYMBOL(__might_sleep);
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6554 6555
static void normalize_task(struct rq *rq, struct task_struct *p)
{
P
Peter Zijlstra 已提交
6556 6557
	const struct sched_class *prev_class = p->sched_class;
	int old_prio = p->prio;
6558
	int on_rq;
6559

P
Peter Zijlstra 已提交
6560
	on_rq = p->on_rq;
6561
	if (on_rq)
6562
		dequeue_task(rq, p, 0);
6563 6564
	__setscheduler(rq, p, SCHED_NORMAL, 0);
	if (on_rq) {
6565
		enqueue_task(rq, p, 0);
6566 6567
		resched_task(rq->curr);
	}
P
Peter Zijlstra 已提交
6568 6569

	check_class_changed(rq, p, prev_class, old_prio);
6570 6571
}

L
Linus Torvalds 已提交
6572 6573
void normalize_rt_tasks(void)
{
6574
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
6575
	unsigned long flags;
6576
	struct rq *rq;
L
Linus Torvalds 已提交
6577

6578
	read_lock_irqsave(&tasklist_lock, flags);
6579
	do_each_thread(g, p) {
6580 6581 6582 6583 6584 6585
		/*
		 * Only normalize user tasks:
		 */
		if (!p->mm)
			continue;

I
Ingo Molnar 已提交
6586 6587
		p->se.exec_start		= 0;
#ifdef CONFIG_SCHEDSTATS
6588 6589 6590
		p->se.statistics.wait_start	= 0;
		p->se.statistics.sleep_start	= 0;
		p->se.statistics.block_start	= 0;
I
Ingo Molnar 已提交
6591
#endif
I
Ingo Molnar 已提交
6592 6593 6594 6595 6596 6597 6598 6599

		if (!rt_task(p)) {
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
			if (TASK_NICE(p) < 0 && p->mm)
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6600
			continue;
I
Ingo Molnar 已提交
6601
		}
L
Linus Torvalds 已提交
6602

6603
		raw_spin_lock(&p->pi_lock);
6604
		rq = __task_rq_lock(p);
L
Linus Torvalds 已提交
6605

6606
		normalize_task(rq, p);
6607

6608
		__task_rq_unlock(rq);
6609
		raw_spin_unlock(&p->pi_lock);
6610 6611
	} while_each_thread(g, p);

6612
	read_unlock_irqrestore(&tasklist_lock, flags);
L
Linus Torvalds 已提交
6613 6614 6615
}

#endif /* CONFIG_MAGIC_SYSRQ */
6616

6617
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6618
/*
6619
 * These functions are only useful for the IA64 MCA handling, or kdb.
6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633
 *
 * They can only be called when the whole system has been
 * stopped - every CPU needs to be quiescent, and no scheduling
 * activity can take place. Using them for anything else would
 * be a serious bug, and as a result, they aren't even visible
 * under any other configuration.
 */

/**
 * curr_task - return the current task for a given cpu.
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
 */
6634
struct task_struct *curr_task(int cpu)
6635 6636 6637 6638
{
	return cpu_curr(cpu);
}

6639 6640 6641
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6642 6643 6644 6645 6646 6647
/**
 * set_curr_task - set the current task for a given cpu.
 * @cpu: the processor in question.
 * @p: the task pointer to set.
 *
 * Description: This function must only be used when non-maskable interrupts
I
Ingo Molnar 已提交
6648 6649
 * are serviced on a separate stack. It allows the architecture to switch the
 * notion of the current task on a cpu in a non-blocking manner. This function
6650 6651 6652 6653 6654 6655 6656
 * must be called with all CPU's synchronized, and interrupts disabled, the
 * and caller must save the original value of the current task (see
 * curr_task() above) and restore that value before reenabling interrupts and
 * re-starting the system.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
 */
6657
void set_curr_task(int cpu, struct task_struct *p)
6658 6659 6660 6661 6662
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6663

D
Dhaval Giani 已提交
6664
#ifdef CONFIG_CGROUP_SCHED
6665 6666 6667
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6668 6669 6670 6671
static void free_sched_group(struct task_group *tg)
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6672
	autogroup_free(tg);
6673 6674 6675 6676
	kfree(tg);
}

/* allocate runqueue etc for a new task group */
6677
struct task_group *sched_create_group(struct task_group *parent)
6678 6679 6680 6681 6682 6683 6684
{
	struct task_group *tg;

	tg = kzalloc(sizeof(*tg), GFP_KERNEL);
	if (!tg)
		return ERR_PTR(-ENOMEM);

6685
	if (!alloc_fair_sched_group(tg, parent))
6686 6687
		goto err;

6688
	if (!alloc_rt_sched_group(tg, parent))
6689 6690
		goto err;

6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701
	return tg;

err:
	free_sched_group(tg);
	return ERR_PTR(-ENOMEM);
}

void sched_online_group(struct task_group *tg, struct task_group *parent)
{
	unsigned long flags;

6702
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6703
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6704 6705 6706 6707 6708

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6709
	list_add_rcu(&tg->siblings, &parent->children);
6710
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6711 6712
}

6713
/* rcu callback to free various structures associated with a task group */
P
Peter Zijlstra 已提交
6714
static void free_sched_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6715 6716
{
	/* now it should be safe to free those cfs_rqs */
P
Peter Zijlstra 已提交
6717
	free_sched_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6718 6719
}

6720
/* Destroy runqueue etc associated with a task group */
6721
void sched_destroy_group(struct task_group *tg)
6722 6723 6724 6725 6726 6727
{
	/* wait for possible concurrent references to cfs_rqs complete */
	call_rcu(&tg->rcu, free_sched_group_rcu);
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6728
{
6729
	unsigned long flags;
6730
	int i;
S
Srivatsa Vaddagiri 已提交
6731

6732 6733
	/* end participation in shares distribution */
	for_each_possible_cpu(i)
6734
		unregister_fair_sched_group(tg, i);
6735 6736

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6737
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6738
	list_del_rcu(&tg->siblings);
6739
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6740 6741
}

6742
/* change task's runqueue when it moves between groups.
I
Ingo Molnar 已提交
6743 6744 6745
 *	The caller of this function should have put the task in its new group
 *	by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to
 *	reflect its new group.
6746 6747
 */
void sched_move_task(struct task_struct *tsk)
S
Srivatsa Vaddagiri 已提交
6748
{
P
Peter Zijlstra 已提交
6749
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6750 6751 6752 6753 6754 6755
	int on_rq, running;
	unsigned long flags;
	struct rq *rq;

	rq = task_rq_lock(tsk, &flags);

6756
	running = task_current(rq, tsk);
P
Peter Zijlstra 已提交
6757
	on_rq = tsk->on_rq;
S
Srivatsa Vaddagiri 已提交
6758

6759
	if (on_rq)
S
Srivatsa Vaddagiri 已提交
6760
		dequeue_task(rq, tsk, 0);
6761 6762
	if (unlikely(running))
		tsk->sched_class->put_prev_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6763

6764
	tg = container_of(task_css_check(tsk, cpu_cgroup_subsys_id,
P
Peter Zijlstra 已提交
6765 6766 6767 6768 6769
				lockdep_is_held(&tsk->sighand->siglock)),
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6770
#ifdef CONFIG_FAIR_GROUP_SCHED
6771 6772 6773
	if (tsk->sched_class->task_move_group)
		tsk->sched_class->task_move_group(tsk, on_rq);
	else
P
Peter Zijlstra 已提交
6774
#endif
6775
		set_task_rq(tsk, task_cpu(tsk));
P
Peter Zijlstra 已提交
6776

6777 6778 6779
	if (unlikely(running))
		tsk->sched_class->set_curr_task(rq);
	if (on_rq)
6780
		enqueue_task(rq, tsk, 0);
S
Srivatsa Vaddagiri 已提交
6781

6782
	task_rq_unlock(rq, tsk, &flags);
S
Srivatsa Vaddagiri 已提交
6783
}
D
Dhaval Giani 已提交
6784
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
6785

6786
#if defined(CONFIG_RT_GROUP_SCHED) || defined(CONFIG_CFS_BANDWIDTH)
P
Peter Zijlstra 已提交
6787 6788 6789
static unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
6790
		return 1ULL << 20;
P
Peter Zijlstra 已提交
6791

P
Peter Zijlstra 已提交
6792
	return div64_u64(runtime << 20, period);
P
Peter Zijlstra 已提交
6793
}
6794 6795 6796 6797 6798 6799 6800
#endif

#ifdef CONFIG_RT_GROUP_SCHED
/*
 * Ensure that the real time constraints are schedulable.
 */
static DEFINE_MUTEX(rt_constraints_mutex);
P
Peter Zijlstra 已提交
6801

P
Peter Zijlstra 已提交
6802 6803
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
6804
{
P
Peter Zijlstra 已提交
6805
	struct task_struct *g, *p;
6806

P
Peter Zijlstra 已提交
6807
	do_each_thread(g, p) {
6808
		if (rt_task(p) && task_rq(p)->rt.tg == tg)
P
Peter Zijlstra 已提交
6809 6810
			return 1;
	} while_each_thread(g, p);
6811

P
Peter Zijlstra 已提交
6812 6813
	return 0;
}
6814

P
Peter Zijlstra 已提交
6815 6816 6817 6818 6819
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
6820

6821
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
6822 6823 6824 6825 6826
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
6827

P
Peter Zijlstra 已提交
6828 6829
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
6830

P
Peter Zijlstra 已提交
6831 6832 6833
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
6834 6835
	}

6836 6837 6838 6839 6840
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
6841

6842 6843 6844
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
6845 6846
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
6847

P
Peter Zijlstra 已提交
6848
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
6849

6850 6851 6852 6853 6854
	/*
	 * Nobody can have more than the global setting allows.
	 */
	if (total > to_ratio(global_rt_period(), global_rt_runtime()))
		return -EINVAL;
P
Peter Zijlstra 已提交
6855

6856 6857 6858
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
6859 6860 6861
	list_for_each_entry_rcu(child, &tg->children, siblings) {
		period = ktime_to_ns(child->rt_bandwidth.rt_period);
		runtime = child->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
6862

P
Peter Zijlstra 已提交
6863 6864 6865 6866
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
6867

P
Peter Zijlstra 已提交
6868
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
6869
	}
P
Peter Zijlstra 已提交
6870

P
Peter Zijlstra 已提交
6871 6872 6873 6874
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
6875 6876
}

P
Peter Zijlstra 已提交
6877
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
6878
{
6879 6880
	int ret;

P
Peter Zijlstra 已提交
6881 6882 6883 6884 6885 6886
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

6887 6888 6889 6890 6891
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6892 6893
}

6894
static int tg_set_rt_bandwidth(struct task_group *tg,
6895
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
6896
{
P
Peter Zijlstra 已提交
6897
	int i, err = 0;
P
Peter Zijlstra 已提交
6898 6899

	mutex_lock(&rt_constraints_mutex);
6900
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
6901 6902
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
6903
		goto unlock;
P
Peter Zijlstra 已提交
6904

6905
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
6906 6907
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
6908 6909 6910 6911

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

6912
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
6913
		rt_rq->rt_runtime = rt_runtime;
6914
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
6915
	}
6916
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
6917
unlock:
6918
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
6919 6920 6921
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
6922 6923
}

6924
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
6925 6926 6927 6928 6929 6930 6931 6932
{
	u64 rt_runtime, rt_period;

	rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
	if (rt_runtime_us < 0)
		rt_runtime = RUNTIME_INF;

6933
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
6934 6935
}

6936
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
6937 6938 6939
{
	u64 rt_runtime_us;

6940
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
6941 6942
		return -1;

6943
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
6944 6945 6946
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
6947

6948
static int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
6949 6950 6951 6952 6953 6954
{
	u64 rt_runtime, rt_period;

	rt_period = (u64)rt_period_us * NSEC_PER_USEC;
	rt_runtime = tg->rt_bandwidth.rt_runtime;

6955 6956 6957
	if (rt_period == 0)
		return -EINVAL;

6958
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
6959 6960
}

6961
static long sched_group_rt_period(struct task_group *tg)
6962 6963 6964 6965 6966 6967 6968 6969 6970 6971
{
	u64 rt_period_us;

	rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
	do_div(rt_period_us, NSEC_PER_USEC);
	return rt_period_us;
}

static int sched_rt_global_constraints(void)
{
6972
	u64 runtime, period;
6973 6974
	int ret = 0;

6975 6976 6977
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

6978 6979 6980 6981 6982 6983 6984 6985
	runtime = global_rt_runtime();
	period = global_rt_period();

	/*
	 * Sanity check on the sysctl variables.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
6986

6987
	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
6988
	read_lock(&tasklist_lock);
6989
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
6990
	read_unlock(&tasklist_lock);
6991 6992 6993 6994
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
6995

6996
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
6997 6998 6999 7000 7001 7002 7003 7004
{
	/* Don't accept realtime tasks when there is no way for them to run */
	if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
		return 0;

	return 1;
}

7005
#else /* !CONFIG_RT_GROUP_SCHED */
7006 7007
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
7008 7009 7010
	unsigned long flags;
	int i;

7011 7012 7013
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

7014 7015 7016 7017 7018 7019 7020
	/*
	 * There's always some RT tasks in the root group
	 * -- migration, kstopmachine etc..
	 */
	if (sysctl_sched_rt_runtime == 0)
		return -EBUSY;

7021
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
7022 7023 7024
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

7025
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7026
		rt_rq->rt_runtime = global_rt_runtime();
7027
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7028
	}
7029
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
7030

7031 7032
	return 0;
}
7033
#endif /* CONFIG_RT_GROUP_SCHED */
7034

7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053
int sched_rr_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos)
{
	int ret;
	static DEFINE_MUTEX(mutex);

	mutex_lock(&mutex);
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
	if (!ret && write) {
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
	}
	mutex_unlock(&mutex);
	return ret;
}

7054
int sched_rt_handler(struct ctl_table *table, int write,
7055
		void __user *buffer, size_t *lenp,
7056 7057 7058 7059 7060 7061 7062 7063 7064 7065
		loff_t *ppos)
{
	int ret;
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

7066
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082

	if (!ret && write) {
		ret = sched_rt_global_constraints();
		if (ret) {
			sysctl_sched_rt_period = old_period;
			sysctl_sched_rt_runtime = old_runtime;
		} else {
			def_rt_bandwidth.rt_runtime = global_rt_runtime();
			def_rt_bandwidth.rt_period =
				ns_to_ktime(global_rt_period());
		}
	}
	mutex_unlock(&mutex);

	return ret;
}
7083

7084
#ifdef CONFIG_CGROUP_SCHED
7085

7086 7087 7088 7089 7090
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
{
	return css ? container_of(css, struct task_group, css) : NULL;
}

7091
/* return corresponding task_group object of a cgroup */
7092
static inline struct task_group *cgroup_tg(struct cgroup *cgrp)
7093
{
7094
	return css_tg(cgroup_css(cgrp, cpu_cgroup_subsys_id));
7095 7096
}

7097
static struct cgroup_subsys_state *cpu_cgroup_css_alloc(struct cgroup *cgrp)
7098
{
7099
	struct task_group *tg, *parent;
7100

7101
	if (!cgrp->parent) {
7102
		/* This is early initialization for the top cgroup */
7103
		return &root_task_group.css;
7104 7105
	}

7106 7107
	parent = cgroup_tg(cgrp->parent);
	tg = sched_create_group(parent);
7108 7109 7110 7111 7112 7113
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

7114 7115 7116
static int cpu_cgroup_css_online(struct cgroup *cgrp)
{
	struct task_group *tg = cgroup_tg(cgrp);
T
Tejun Heo 已提交
7117
	struct task_group *parent = css_tg(css_parent(&tg->css));
7118

T
Tejun Heo 已提交
7119 7120
	if (parent)
		sched_online_group(tg, parent);
7121 7122 7123
	return 0;
}

7124
static void cpu_cgroup_css_free(struct cgroup *cgrp)
7125
{
7126
	struct task_group *tg = cgroup_tg(cgrp);
7127 7128 7129 7130

	sched_destroy_group(tg);
}

7131 7132 7133 7134 7135 7136 7137
static void cpu_cgroup_css_offline(struct cgroup *cgrp)
{
	struct task_group *tg = cgroup_tg(cgrp);

	sched_offline_group(tg);
}

7138
static int cpu_cgroup_can_attach(struct cgroup *cgrp,
7139
				 struct cgroup_taskset *tset)
7140
{
7141 7142 7143
	struct task_struct *task;

	cgroup_taskset_for_each(task, cgrp, tset) {
7144
#ifdef CONFIG_RT_GROUP_SCHED
7145 7146
		if (!sched_rt_can_attach(cgroup_tg(cgrp), task))
			return -EINVAL;
7147
#else
7148 7149 7150
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
7151
#endif
7152
	}
7153 7154
	return 0;
}
7155

7156
static void cpu_cgroup_attach(struct cgroup *cgrp,
7157
			      struct cgroup_taskset *tset)
7158
{
7159 7160 7161 7162
	struct task_struct *task;

	cgroup_taskset_for_each(task, cgrp, tset)
		sched_move_task(task);
7163 7164
}

7165
static void
7166 7167
cpu_cgroup_exit(struct cgroup *cgrp, struct cgroup *old_cgrp,
		struct task_struct *task)
7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179
{
	/*
	 * cgroup_exit() is called in the copy_process() failure path.
	 * Ignore this case since the task hasn't ran yet, this avoids
	 * trying to poke a half freed task state from generic code.
	 */
	if (!(task->flags & PF_EXITING))
		return;

	sched_move_task(task);
}

7180
#ifdef CONFIG_FAIR_GROUP_SCHED
7181
static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype,
7182
				u64 shareval)
7183
{
7184
	return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval));
7185 7186
}

7187
static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft)
7188
{
7189
	struct task_group *tg = cgroup_tg(cgrp);
7190

7191
	return (u64) scale_load_down(tg->shares);
7192
}
7193 7194

#ifdef CONFIG_CFS_BANDWIDTH
7195 7196
static DEFINE_MUTEX(cfs_constraints_mutex);

7197 7198 7199
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

7200 7201
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

7202 7203
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
7204
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
7205
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225

	if (tg == &root_task_group)
		return -EINVAL;

	/*
	 * Ensure we have at some amount of bandwidth every period.  This is
	 * to prevent reaching a state of large arrears when throttled via
	 * entity_tick() resulting in prolonged exit starvation.
	 */
	if (quota < min_cfs_quota_period || period < min_cfs_quota_period)
		return -EINVAL;

	/*
	 * Likewise, bound things on the otherside by preventing insane quota
	 * periods.  This also allows us to normalize in computing quota
	 * feasibility.
	 */
	if (period > max_cfs_quota_period)
		return -EINVAL;

7226 7227 7228 7229 7230
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

7231
	runtime_enabled = quota != RUNTIME_INF;
7232 7233
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
	account_cfs_bandwidth_used(runtime_enabled, runtime_was_enabled);
7234 7235 7236
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
7237

P
Paul Turner 已提交
7238
	__refill_cfs_bandwidth_runtime(cfs_b);
7239 7240 7241 7242 7243 7244
	/* restart the period timer (if active) to handle new period expiry */
	if (runtime_enabled && cfs_b->timer_active) {
		/* force a reprogram */
		cfs_b->timer_active = 0;
		__start_cfs_bandwidth(cfs_b);
	}
7245 7246 7247 7248
	raw_spin_unlock_irq(&cfs_b->lock);

	for_each_possible_cpu(i) {
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
7249
		struct rq *rq = cfs_rq->rq;
7250 7251

		raw_spin_lock_irq(&rq->lock);
7252
		cfs_rq->runtime_enabled = runtime_enabled;
7253
		cfs_rq->runtime_remaining = 0;
7254

7255
		if (cfs_rq->throttled)
7256
			unthrottle_cfs_rq(cfs_rq);
7257 7258
		raw_spin_unlock_irq(&rq->lock);
	}
7259 7260
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
7261

7262
	return ret;
7263 7264 7265 7266 7267 7268
}

int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
{
	u64 quota, period;

7269
	period = ktime_to_ns(tg->cfs_bandwidth.period);
7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280 7281
	if (cfs_quota_us < 0)
		quota = RUNTIME_INF;
	else
		quota = (u64)cfs_quota_us * NSEC_PER_USEC;

	return tg_set_cfs_bandwidth(tg, period, quota);
}

long tg_get_cfs_quota(struct task_group *tg)
{
	u64 quota_us;

7282
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
7283 7284
		return -1;

7285
	quota_us = tg->cfs_bandwidth.quota;
7286 7287 7288 7289 7290 7291 7292 7293 7294 7295
	do_div(quota_us, NSEC_PER_USEC);

	return quota_us;
}

int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
{
	u64 quota, period;

	period = (u64)cfs_period_us * NSEC_PER_USEC;
7296
	quota = tg->cfs_bandwidth.quota;
7297 7298 7299 7300 7301 7302 7303 7304

	return tg_set_cfs_bandwidth(tg, period, quota);
}

long tg_get_cfs_period(struct task_group *tg)
{
	u64 cfs_period_us;

7305
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320 7321 7322 7323 7324 7325 7326 7327 7328 7329 7330 7331 7332
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

static s64 cpu_cfs_quota_read_s64(struct cgroup *cgrp, struct cftype *cft)
{
	return tg_get_cfs_quota(cgroup_tg(cgrp));
}

static int cpu_cfs_quota_write_s64(struct cgroup *cgrp, struct cftype *cftype,
				s64 cfs_quota_us)
{
	return tg_set_cfs_quota(cgroup_tg(cgrp), cfs_quota_us);
}

static u64 cpu_cfs_period_read_u64(struct cgroup *cgrp, struct cftype *cft)
{
	return tg_get_cfs_period(cgroup_tg(cgrp));
}

static int cpu_cfs_period_write_u64(struct cgroup *cgrp, struct cftype *cftype,
				u64 cfs_period_us)
{
	return tg_set_cfs_period(cgroup_tg(cgrp), cfs_period_us);
}

7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364
struct cfs_schedulable_data {
	struct task_group *tg;
	u64 period, quota;
};

/*
 * normalize group quota/period to be quota/max_period
 * note: units are usecs
 */
static u64 normalize_cfs_quota(struct task_group *tg,
			       struct cfs_schedulable_data *d)
{
	u64 quota, period;

	if (tg == d->tg) {
		period = d->period;
		quota = d->quota;
	} else {
		period = tg_get_cfs_period(tg);
		quota = tg_get_cfs_quota(tg);
	}

	/* note: these should typically be equivalent */
	if (quota == RUNTIME_INF || quota == -1)
		return RUNTIME_INF;

	return to_ratio(period, quota);
}

static int tg_cfs_schedulable_down(struct task_group *tg, void *data)
{
	struct cfs_schedulable_data *d = data;
7365
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7366 7367 7368 7369 7370
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
7371
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384 7385 7386 7387 7388 7389 7390 7391

		quota = normalize_cfs_quota(tg, d);
		parent_quota = parent_b->hierarchal_quota;

		/*
		 * ensure max(child_quota) <= parent_quota, inherit when no
		 * limit is set
		 */
		if (quota == RUNTIME_INF)
			quota = parent_quota;
		else if (parent_quota != RUNTIME_INF && quota > parent_quota)
			return -EINVAL;
	}
	cfs_b->hierarchal_quota = quota;

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
7392
	int ret;
7393 7394 7395 7396 7397 7398 7399 7400 7401 7402 7403
	struct cfs_schedulable_data data = {
		.tg = tg,
		.period = period,
		.quota = quota,
	};

	if (quota != RUNTIME_INF) {
		do_div(data.period, NSEC_PER_USEC);
		do_div(data.quota, NSEC_PER_USEC);
	}

7404 7405 7406 7407 7408
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
7409
}
7410 7411 7412 7413 7414

static int cpu_stats_show(struct cgroup *cgrp, struct cftype *cft,
		struct cgroup_map_cb *cb)
{
	struct task_group *tg = cgroup_tg(cgrp);
7415
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
7416 7417 7418 7419 7420 7421 7422

	cb->fill(cb, "nr_periods", cfs_b->nr_periods);
	cb->fill(cb, "nr_throttled", cfs_b->nr_throttled);
	cb->fill(cb, "throttled_time", cfs_b->throttled_time);

	return 0;
}
7423
#endif /* CONFIG_CFS_BANDWIDTH */
7424
#endif /* CONFIG_FAIR_GROUP_SCHED */
7425

7426
#ifdef CONFIG_RT_GROUP_SCHED
M
Mirco Tischler 已提交
7427
static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft,
7428
				s64 val)
P
Peter Zijlstra 已提交
7429
{
7430
	return sched_group_set_rt_runtime(cgroup_tg(cgrp), val);
P
Peter Zijlstra 已提交
7431 7432
}

7433
static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft)
P
Peter Zijlstra 已提交
7434
{
7435
	return sched_group_rt_runtime(cgroup_tg(cgrp));
P
Peter Zijlstra 已提交
7436
}
7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447

static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
		u64 rt_period_us)
{
	return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
}

static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
{
	return sched_group_rt_period(cgroup_tg(cgrp));
}
7448
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
7449

7450
static struct cftype cpu_files[] = {
7451
#ifdef CONFIG_FAIR_GROUP_SCHED
7452 7453
	{
		.name = "shares",
7454 7455
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
7456
	},
7457
#endif
7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468
#ifdef CONFIG_CFS_BANDWIDTH
	{
		.name = "cfs_quota_us",
		.read_s64 = cpu_cfs_quota_read_s64,
		.write_s64 = cpu_cfs_quota_write_s64,
	},
	{
		.name = "cfs_period_us",
		.read_u64 = cpu_cfs_period_read_u64,
		.write_u64 = cpu_cfs_period_write_u64,
	},
7469 7470 7471 7472
	{
		.name = "stat",
		.read_map = cpu_stats_show,
	},
7473
#endif
7474
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
7475
	{
P
Peter Zijlstra 已提交
7476
		.name = "rt_runtime_us",
7477 7478
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
7479
	},
7480 7481
	{
		.name = "rt_period_us",
7482 7483
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
7484
	},
7485
#endif
7486
	{ }	/* terminate */
7487 7488 7489
};

struct cgroup_subsys cpu_cgroup_subsys = {
I
Ingo Molnar 已提交
7490
	.name		= "cpu",
7491 7492
	.css_alloc	= cpu_cgroup_css_alloc,
	.css_free	= cpu_cgroup_css_free,
7493 7494
	.css_online	= cpu_cgroup_css_online,
	.css_offline	= cpu_cgroup_css_offline,
7495 7496
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
7497
	.exit		= cpu_cgroup_exit,
I
Ingo Molnar 已提交
7498
	.subsys_id	= cpu_cgroup_subsys_id,
7499
	.base_cftypes	= cpu_files,
7500 7501 7502
	.early_init	= 1,
};

7503
#endif	/* CONFIG_CGROUP_SCHED */
7504

7505 7506 7507 7508 7509
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}