core.c 214.1 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
#include <linux/kasan.h>
L
Linus Torvalds 已提交
30 31 32 33
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/nmi.h>
#include <linux/init.h>
34
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
35
#include <linux/highmem.h>
36
#include <linux/mmu_context.h>
L
Linus Torvalds 已提交
37
#include <linux/interrupt.h>
38
#include <linux/capability.h>
L
Linus Torvalds 已提交
39 40
#include <linux/completion.h>
#include <linux/kernel_stat.h>
41
#include <linux/debug_locks.h>
42
#include <linux/perf_event.h>
L
Linus Torvalds 已提交
43 44 45
#include <linux/security.h>
#include <linux/notifier.h>
#include <linux/profile.h>
46
#include <linux/freezer.h>
47
#include <linux/vmalloc.h>
L
Linus Torvalds 已提交
48 49
#include <linux/blkdev.h>
#include <linux/delay.h>
50
#include <linux/pid_namespace.h>
L
Linus Torvalds 已提交
51 52 53 54 55 56 57
#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>
58
#include <linux/proc_fs.h>
L
Linus Torvalds 已提交
59
#include <linux/seq_file.h>
60
#include <linux/sysctl.h>
L
Linus Torvalds 已提交
61 62
#include <linux/syscalls.h>
#include <linux/times.h>
63
#include <linux/tsacct_kern.h>
64
#include <linux/kprobes.h>
65
#include <linux/delayacct.h>
66
#include <linux/unistd.h>
J
Jens Axboe 已提交
67
#include <linux/pagemap.h>
P
Peter Zijlstra 已提交
68
#include <linux/hrtimer.h>
R
Reynes Philippe 已提交
69
#include <linux/tick.h>
P
Peter Zijlstra 已提交
70
#include <linux/ctype.h>
71
#include <linux/ftrace.h>
72
#include <linux/slab.h>
73
#include <linux/init_task.h>
74
#include <linux/context_tracking.h>
75
#include <linux/compiler.h>
76
#include <linux/frame.h>
77
#include <linux/prefetch.h>
L
Linus Torvalds 已提交
78

79
#include <asm/switch_to.h>
80
#include <asm/tlb.h>
81
#include <asm/irq_regs.h>
82
#include <asm/mutex.h>
G
Glauber Costa 已提交
83 84 85
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
#endif
L
Linus Torvalds 已提交
86

87
#include "sched.h"
88
#include "../workqueue_internal.h"
89
#include "../smpboot.h"
90

91
#define CREATE_TRACE_POINTS
92
#include <trace/events/sched.h>
93

94 95
DEFINE_MUTEX(sched_domains_mutex);
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
96

97
static void update_rq_clock_task(struct rq *rq, s64 delta);
98

99
void update_rq_clock(struct rq *rq)
100
{
101
	s64 delta;
102

103 104 105
	lockdep_assert_held(&rq->lock);

	if (rq->clock_skip_update & RQCF_ACT_SKIP)
106
		return;
107

108
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
109 110
	if (delta < 0)
		return;
111 112
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
113 114
}

I
Ingo Molnar 已提交
115 116 117
/*
 * Debugging: various feature bits
 */
P
Peter Zijlstra 已提交
118 119 120 121

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

I
Ingo Molnar 已提交
122
const_debug unsigned int sysctl_sched_features =
123
#include "features.h"
P
Peter Zijlstra 已提交
124 125 126 127
	0;

#undef SCHED_FEAT

128 129 130 131 132 133
/*
 * 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;

134 135 136 137 138 139 140 141
/*
 * 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
142
/*
P
Peter Zijlstra 已提交
143
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
144 145
 * default: 1s
 */
P
Peter Zijlstra 已提交
146
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
147

148
__read_mostly int scheduler_running;
149

P
Peter Zijlstra 已提交
150 151 152 153 154
/*
 * 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
155

156 157 158
/* cpus with isolated domains */
cpumask_var_t cpu_isolated_map;

L
Linus Torvalds 已提交
159
/*
160
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
161
 */
A
Alexey Dobriyan 已提交
162
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
163 164
	__acquires(rq->lock)
{
165
	struct rq *rq;
L
Linus Torvalds 已提交
166 167 168

	local_irq_disable();
	rq = this_rq();
169
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
170 171 172 173

	return rq;
}

174 175 176
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
177
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
178 179 180 181 182 183 184 185 186 187
	__acquires(rq->lock)
{
	struct rq *rq;

	lockdep_assert_held(&p->pi_lock);

	for (;;) {
		rq = task_rq(p);
		raw_spin_lock(&rq->lock);
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
188
			rf->cookie = lockdep_pin_lock(&rq->lock);
189 190 191 192 193 194 195 196 197 198 199 200
			return rq;
		}
		raw_spin_unlock(&rq->lock);

		while (unlikely(task_on_rq_migrating(p)))
			cpu_relax();
	}
}

/*
 * task_rq_lock - lock p->pi_lock and lock the rq @p resides on.
 */
201
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
202 203 204 205 206 207
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
208
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
		rq = task_rq(p);
		raw_spin_lock(&rq->lock);
		/*
		 *	move_queued_task()		task_rq_lock()
		 *
		 *	ACQUIRE (rq->lock)
		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
		 *	[S] ->cpu = new_cpu		[L] task_rq()
		 *					[L] ->on_rq
		 *	RELEASE (rq->lock)
		 *
		 * If we observe the old cpu in task_rq_lock, the acquire of
		 * the old rq->lock will fully serialize against the stores.
		 *
		 * If we observe the new cpu in task_rq_lock, the acquire will
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
228
			rf->cookie = lockdep_pin_lock(&rq->lock);
229 230 231
			return rq;
		}
		raw_spin_unlock(&rq->lock);
232
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
233 234 235 236 237 238

		while (unlikely(task_on_rq_migrating(p)))
			cpu_relax();
	}
}

P
Peter Zijlstra 已提交
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259
#ifdef CONFIG_SCHED_HRTICK
/*
 * Use HR-timers to deliver accurate preemption points.
 */

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());

260
	raw_spin_lock(&rq->lock);
261
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
262
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
263
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
264 265 266 267

	return HRTIMER_NORESTART;
}

268
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
269

270
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
271 272 273
{
	struct hrtimer *timer = &rq->hrtick_timer;

274
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
275 276
}

277 278 279 280
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
281
{
282
	struct rq *rq = arg;
283

284
	raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
285
	__hrtick_restart(rq);
286
	rq->hrtick_csd_pending = 0;
287
	raw_spin_unlock(&rq->lock);
288 289
}

290 291 292 293 294
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
295
void hrtick_start(struct rq *rq, u64 delay)
296
{
297
	struct hrtimer *timer = &rq->hrtick_timer;
298 299 300 301 302 303 304 305 306
	ktime_t time;
	s64 delta;

	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense and can cause timer DoS.
	 */
	delta = max_t(s64, delay, 10000LL);
	time = ktime_add_ns(timer->base->get_time(), delta);
307

308
	hrtimer_set_expires(timer, time);
309 310

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
311
		__hrtick_restart(rq);
312
	} else if (!rq->hrtick_csd_pending) {
313
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
314 315
		rq->hrtick_csd_pending = 1;
	}
316 317
}

318 319 320 321 322 323
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
324
void hrtick_start(struct rq *rq, u64 delay)
325
{
W
Wanpeng Li 已提交
326 327 328 329 330
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
331 332
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
333 334
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
335

336
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
337
{
338 339
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
340

341 342 343 344
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
345

346 347
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
348
}
A
Andrew Morton 已提交
349
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
350 351 352 353 354 355 356
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}
A
Andrew Morton 已提交
357
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
358

359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
/*
 * cmpxchg based fetch_or, macro so it works for different integer types
 */
#define fetch_or(ptr, mask)						\
	({								\
		typeof(ptr) _ptr = (ptr);				\
		typeof(mask) _mask = (mask);				\
		typeof(*_ptr) _old, _val = *_ptr;			\
									\
		for (;;) {						\
			_old = cmpxchg(_ptr, _val, _val | _mask);	\
			if (_old == _val)				\
				break;					\
			_val = _old;					\
		}							\
	_old;								\
})

377
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
378 379 380 381 382 383 384 385 386 387
/*
 * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG,
 * this avoids any races wrt polling state changes and thereby avoids
 * spurious IPIs.
 */
static bool set_nr_and_not_polling(struct task_struct *p)
{
	struct thread_info *ti = task_thread_info(p);
	return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG);
}
388 389 390 391 392 393 394 395 396 397

/*
 * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set.
 *
 * If this returns true, then the idle task promises to call
 * sched_ttwu_pending() and reschedule soon.
 */
static bool set_nr_if_polling(struct task_struct *p)
{
	struct thread_info *ti = task_thread_info(p);
398
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
399 400 401 402 403 404 405 406 407 408 409 410 411 412

	for (;;) {
		if (!(val & _TIF_POLLING_NRFLAG))
			return false;
		if (val & _TIF_NEED_RESCHED)
			return true;
		old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED);
		if (old == val)
			break;
		val = old;
	}
	return true;
}

413 414 415 416 417 418
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
419 420 421 422 423 424 425

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
426 427
#endif

428 429 430 431 432 433 434 435 436 437
void wake_q_add(struct wake_q_head *head, struct task_struct *task)
{
	struct wake_q_node *node = &task->wake_q;

	/*
	 * Atomically grab the task, if ->wake_q is !nil already it means
	 * its already queued (either by us or someone else) and will get the
	 * wakeup due to that.
	 *
	 * This cmpxchg() implies a full barrier, which pairs with the write
438
	 * barrier implied by the wakeup in wake_up_q().
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
	 */
	if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL))
		return;

	get_task_struct(task);

	/*
	 * The head is context local, there can be no concurrency.
	 */
	*head->lastp = node;
	head->lastp = &node->next;
}

void wake_up_q(struct wake_q_head *head)
{
	struct wake_q_node *node = head->first;

	while (node != WAKE_Q_TAIL) {
		struct task_struct *task;

		task = container_of(node, struct task_struct, wake_q);
		BUG_ON(!task);
		/* task can safely be re-inserted now */
		node = node->next;
		task->wake_q.next = NULL;

		/*
		 * wake_up_process() implies a wmb() to pair with the queueing
		 * in wake_q_add() so as not to miss wakeups.
		 */
		wake_up_process(task);
		put_task_struct(task);
	}
}

I
Ingo Molnar 已提交
474
/*
475
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
476 477 478 479 480
 *
 * 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.
 */
481
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
482
{
483
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
484 485
	int cpu;

486
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
487

488
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
489 490
		return;

491
	cpu = cpu_of(rq);
492

493
	if (cpu == smp_processor_id()) {
494
		set_tsk_need_resched(curr);
495
		set_preempt_need_resched();
I
Ingo Molnar 已提交
496
		return;
497
	}
I
Ingo Molnar 已提交
498

499
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
500
		smp_send_reschedule(cpu);
501 502
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
503 504
}

505
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
506 507 508 509
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

510
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
511
		return;
512
	resched_curr(rq);
513
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
514
}
515

516
#ifdef CONFIG_SMP
517
#ifdef CONFIG_NO_HZ_COMMON
518 519 520 521 522 523 524 525
/*
 * 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).
 */
526
int get_nohz_timer_target(void)
527
{
528
	int i, cpu = smp_processor_id();
529 530
	struct sched_domain *sd;

531
	if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
532 533
		return cpu;

534
	rcu_read_lock();
535
	for_each_domain(cpu, sd) {
536
		for_each_cpu(i, sched_domain_span(sd)) {
537 538 539 540
			if (cpu == i)
				continue;

			if (!idle_cpu(i) && is_housekeeping_cpu(i)) {
541 542 543 544
				cpu = i;
				goto unlock;
			}
		}
545
	}
546 547 548

	if (!is_housekeeping_cpu(cpu))
		cpu = housekeeping_any_cpu();
549 550
unlock:
	rcu_read_unlock();
551 552
	return cpu;
}
553 554 555 556 557 558 559 560 561 562
/*
 * 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.
 */
563
static void wake_up_idle_cpu(int cpu)
564 565 566 567 568 569
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

570
	if (set_nr_and_not_polling(rq->idle))
571
		smp_send_reschedule(cpu);
572 573
	else
		trace_sched_wake_idle_without_ipi(cpu);
574 575
}

576
static bool wake_up_full_nohz_cpu(int cpu)
577
{
578 579 580 581 582 583
	/*
	 * We just need the target to call irq_exit() and re-evaluate
	 * the next tick. The nohz full kick at least implies that.
	 * If needed we can still optimize that later with an
	 * empty IRQ.
	 */
584
	if (tick_nohz_full_cpu(cpu)) {
585 586
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
587
			tick_nohz_full_kick_cpu(cpu);
588 589 590 591 592 593 594 595
		return true;
	}

	return false;
}

void wake_up_nohz_cpu(int cpu)
{
596
	if (!wake_up_full_nohz_cpu(cpu))
597 598 599
		wake_up_idle_cpu(cpu);
}

600
static inline bool got_nohz_idle_kick(void)
601
{
602
	int cpu = smp_processor_id();
603 604 605 606 607 608 609 610 611 612 613 614 615

	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;
616 617
}

618
#else /* CONFIG_NO_HZ_COMMON */
619

620
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
621
{
622
	return false;
P
Peter Zijlstra 已提交
623 624
}

625
#endif /* CONFIG_NO_HZ_COMMON */
626

627
#ifdef CONFIG_NO_HZ_FULL
628
bool sched_can_stop_tick(struct rq *rq)
629
{
630 631 632 633 634 635
	int fifo_nr_running;

	/* Deadline tasks, even if single, need the tick */
	if (rq->dl.dl_nr_running)
		return false;

636
	/*
637 638
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
639
	 */
640 641 642 643 644
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
645 646
	}

647 648 649 650 651 652 653 654 655 656 657 658 659 660
	/*
	 * If there's no RR tasks, but FIFO tasks, we can skip the tick, no
	 * forced preemption between FIFO tasks.
	 */
	fifo_nr_running = rq->rt.rt_nr_running - rq->rt.rr_nr_running;
	if (fifo_nr_running)
		return true;

	/*
	 * If there are no DL,RR/FIFO tasks, there must only be CFS tasks left;
	 * if there's more than one we need the tick for involuntary
	 * preemption.
	 */
	if (rq->nr_running > 1)
661
		return false;
662

663
	return true;
664 665
}
#endif /* CONFIG_NO_HZ_FULL */
666

667
void sched_avg_update(struct rq *rq)
668
{
669 670
	s64 period = sched_avg_period();

671
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
672 673 674 675 676 677
		/*
		 * 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));
678 679 680
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
681 682
}

683
#endif /* CONFIG_SMP */
684

685 686
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
687
/*
688 689 690 691
 * 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.
692
 */
693
int walk_tg_tree_from(struct task_group *from,
694
			     tg_visitor down, tg_visitor up, void *data)
695 696
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
697
	int ret;
698

699 700
	parent = from;

701
down:
P
Peter Zijlstra 已提交
702 703
	ret = (*down)(parent, data);
	if (ret)
704
		goto out;
705 706 707 708 709 710 711
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
712
	ret = (*up)(parent, data);
713 714
	if (ret || parent == from)
		goto out;
715 716 717 718 719

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
720
out:
P
Peter Zijlstra 已提交
721
	return ret;
722 723
}

724
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
725
{
726
	return 0;
P
Peter Zijlstra 已提交
727
}
728 729
#endif

730 731
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
732 733 734
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
735 736 737
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
738
	if (idle_policy(p->policy)) {
739
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
740
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
741 742
		return;
	}
743

744 745
	load->weight = scale_load(sched_prio_to_weight[prio]);
	load->inv_weight = sched_prio_to_wmult[prio];
746 747
}

748
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
749
{
750
	update_rq_clock(rq);
751 752
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
753
	p->sched_class->enqueue_task(rq, p, flags);
754 755
}

756
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
757
{
758
	update_rq_clock(rq);
759 760
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
761
	p->sched_class->dequeue_task(rq, p, flags);
762 763
}

764
void activate_task(struct rq *rq, struct task_struct *p, int flags)
765 766 767 768
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

769
	enqueue_task(rq, p, flags);
770 771
}

772
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
773 774 775 776
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

777
	dequeue_task(rq, p, flags);
778 779
}

780
static void update_rq_clock_task(struct rq *rq, s64 delta)
781
{
782 783 784 785 786 787 788 789
/*
 * 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
790
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811

	/*
	 * 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;
812 813
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
814
	if (static_key_false((&paravirt_steal_rq_enabled))) {
815 816 817 818 819 820 821 822 823 824 825
		steal = paravirt_steal_clock(cpu_of(rq));
		steal -= rq->prev_steal_time_rq;

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

		rq->prev_steal_time_rq += steal;
		delta -= steal;
	}
#endif

826 827
	rq->clock_task += delta;

828
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
829
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
830 831
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
832 833
}

834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863
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;
	}
}

864
/*
I
Ingo Molnar 已提交
865
 * __normal_prio - return the priority that is based on the static prio
866 867 868
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
869
	return p->static_prio;
870 871
}

872 873 874 875 876 877 878
/*
 * 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.
 */
879
static inline int normal_prio(struct task_struct *p)
880 881 882
{
	int prio;

883 884 885
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
886 887 888 889 890 891 892 893 894 895 896 897 898
		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.
 */
899
static int effective_prio(struct task_struct *p)
900 901 902 903 904 905 906 907 908 909 910 911
{
	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
912 913 914
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
915 916
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
917
 */
918
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
919 920 921 922
{
	return cpu_curr(task_cpu(p)) == p;
}

923
/*
924 925 926 927 928
 * switched_from, switched_to and prio_changed must _NOT_ drop rq->lock,
 * use the balance_callback list if you want balancing.
 *
 * this means any call to check_class_changed() must be followed by a call to
 * balance_callback().
929
 */
930 931
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
932
				       int oldprio)
933 934 935
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
936
			prev_class->switched_from(rq, p);
937

P
Peter Zijlstra 已提交
938
		p->sched_class->switched_to(rq, p);
939
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
940
		p->sched_class->prio_changed(rq, p, oldprio);
941 942
}

943
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
944 945 946 947 948 949 950 951 952 953
{
	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) {
954
				resched_curr(rq);
955 956 957 958 959 960 961 962 963
				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.
	 */
964
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
965
		rq_clock_skip_update(rq, true);
966 967
}

L
Linus Torvalds 已提交
968
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987
/*
 * This is how migration works:
 *
 * 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
 *    it and puts it into the right queue.
 * 5) stopper completes and stop_one_cpu() returns and the migration
 *    is done.
 */

/*
 * move_queued_task - move a queued task to new rq.
 *
 * Returns (locked) new rq. Old rq's lock is released.
 */
988
static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
989 990 991 992
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
993
	dequeue_task(rq, p, 0);
P
Peter Zijlstra 已提交
994 995 996 997 998 999 1000 1001
	set_task_cpu(p, new_cpu);
	raw_spin_unlock(&rq->lock);

	rq = cpu_rq(new_cpu);

	raw_spin_lock(&rq->lock);
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
1002
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
	check_preempt_curr(rq, p, 0);

	return rq;
}

struct migration_arg {
	struct task_struct *task;
	int dest_cpu;
};

/*
 * Move (not current) task off this cpu, onto dest cpu. We're doing
 * 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.
 */
1022
static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
1023 1024
{
	if (unlikely(!cpu_active(dest_cpu)))
1025
		return rq;
P
Peter Zijlstra 已提交
1026 1027 1028

	/* Affinity changed (again). */
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
1029
		return rq;
P
Peter Zijlstra 已提交
1030

1031 1032 1033
	rq = move_queued_task(rq, p, dest_cpu);

	return rq;
P
Peter Zijlstra 已提交
1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
}

/*
 * 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.
 */
static int migration_cpu_stop(void *data)
{
	struct migration_arg *arg = data;
1044 1045
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057

	/*
	 * The original target cpu might have gone down and we might
	 * be on another cpu but it doesn't matter.
	 */
	local_irq_disable();
	/*
	 * We need to explicitly wake pending tasks before running
	 * __migrate_task() such that we will not miss enforcing cpus_allowed
	 * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test.
	 */
	sched_ttwu_pending();
1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070

	raw_spin_lock(&p->pi_lock);
	raw_spin_lock(&rq->lock);
	/*
	 * If task_rq(p) != rq, it cannot be migrated here, because we're
	 * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because
	 * we're holding p->pi_lock.
	 */
	if (task_rq(p) == rq && task_on_rq_queued(p))
		rq = __migrate_task(rq, p, arg->dest_cpu);
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1071 1072 1073 1074
	local_irq_enable();
	return 0;
}

1075 1076 1077 1078 1079
/*
 * sched_class::set_cpus_allowed must do the below, but is not required to
 * actually call this function.
 */
void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask)
P
Peter Zijlstra 已提交
1080 1081 1082 1083 1084
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1085 1086
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1087 1088 1089
	struct rq *rq = task_rq(p);
	bool queued, running;

1090
	lockdep_assert_held(&p->pi_lock);
1091 1092 1093 1094 1095 1096 1097 1098 1099 1100

	queued = task_on_rq_queued(p);
	running = task_current(rq, p);

	if (queued) {
		/*
		 * Because __kthread_bind() calls this on blocked tasks without
		 * holding rq->lock.
		 */
		lockdep_assert_held(&rq->lock);
1101
		dequeue_task(rq, p, DEQUEUE_SAVE);
1102 1103 1104 1105
	}
	if (running)
		put_prev_task(rq, p);

1106
	p->sched_class->set_cpus_allowed(p, new_mask);
1107 1108 1109 1110

	if (running)
		p->sched_class->set_curr_task(rq);
	if (queued)
1111
		enqueue_task(rq, p, ENQUEUE_RESTORE);
1112 1113
}

P
Peter Zijlstra 已提交
1114 1115 1116 1117 1118 1119 1120 1121 1122
/*
 * 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
 * task must not exit() & deallocate itself prematurely. The
 * call is not atomic; no spinlocks may be held.
 */
1123 1124
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1125
{
1126
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1127
	unsigned int dest_cpu;
1128 1129
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1130 1131
	int ret = 0;

1132
	rq = task_rq_lock(p, &rf);
P
Peter Zijlstra 已提交
1133

1134 1135 1136 1137 1138 1139 1140
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1141 1142 1143 1144 1145 1146 1147 1148 1149
	/*
	 * Must re-check here, to close a race against __kthread_bind(),
	 * sched_setaffinity() is not guaranteed to observe the flag.
	 */
	if (check && (p->flags & PF_NO_SETAFFINITY)) {
		ret = -EINVAL;
		goto out;
	}

P
Peter Zijlstra 已提交
1150 1151 1152
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1153
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1154 1155 1156 1157 1158 1159
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1160 1161 1162 1163 1164 1165 1166 1167 1168 1169
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
		 * !active we want to ensure they are strict per-cpu threads.
		 */
		WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) &&
			!cpumask_intersects(new_mask, cpu_active_mask) &&
			p->nr_cpus_allowed != 1);
	}

P
Peter Zijlstra 已提交
1170 1171 1172 1173
	/* Can the task run on the task's current CPU? If so, we're done */
	if (cpumask_test_cpu(task_cpu(p), new_mask))
		goto out;

1174
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1175 1176 1177
	if (task_running(rq, p) || p->state == TASK_WAKING) {
		struct migration_arg arg = { p, dest_cpu };
		/* Need help from migration thread: drop lock and wait. */
1178
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1179 1180 1181
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1182 1183 1184 1185 1186
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1187
		lockdep_unpin_lock(&rq->lock, rf.cookie);
1188
		rq = move_queued_task(rq, p, dest_cpu);
1189
		lockdep_repin_lock(&rq->lock, rf.cookie);
1190
	}
P
Peter Zijlstra 已提交
1191
out:
1192
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1193 1194 1195

	return ret;
}
1196 1197 1198 1199 1200

int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
{
	return __set_cpus_allowed_ptr(p, new_mask, false);
}
P
Peter Zijlstra 已提交
1201 1202
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1203
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1204
{
1205 1206 1207 1208 1209
#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
1210
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1211
			!p->on_rq);
1212

1213 1214 1215 1216 1217 1218 1219 1220 1221
	/*
	 * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING,
	 * because schedstat_wait_{start,end} rebase migrating task's wait_start
	 * time relying on p->on_rq.
	 */
	WARN_ON_ONCE(p->state == TASK_RUNNING &&
		     p->sched_class == &fair_sched_class &&
		     (p->on_rq && !task_on_rq_migrating(p)));

1222
#ifdef CONFIG_LOCKDEP
1223 1224 1225 1226 1227
	/*
	 * 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
1228
	 * see task_group().
1229 1230 1231 1232
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1233 1234 1235
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1236 1237
#endif

1238
	trace_sched_migrate_task(p, new_cpu);
1239

1240
	if (task_cpu(p) != new_cpu) {
1241
		if (p->sched_class->migrate_task_rq)
1242
			p->sched_class->migrate_task_rq(p);
1243
		p->se.nr_migrations++;
1244
		perf_event_task_migrate(p);
1245
	}
I
Ingo Molnar 已提交
1246 1247

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1248 1249
}

1250 1251
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1252
	if (task_on_rq_queued(p)) {
1253 1254 1255 1256 1257
		struct rq *src_rq, *dst_rq;

		src_rq = task_rq(p);
		dst_rq = cpu_rq(cpu);

1258
		p->on_rq = TASK_ON_RQ_MIGRATING;
1259 1260 1261
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1262
		p->on_rq = TASK_ON_RQ_QUEUED;
1263 1264 1265 1266 1267
		check_preempt_curr(dst_rq, p, 0);
	} else {
		/*
		 * Task isn't running anymore; make it appear like we migrated
		 * it before it went to sleep. This means on wakeup we make the
L
Leo Yan 已提交
1268
		 * previous cpu our target instead of where it really is.
1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
		 */
		p->wake_cpu = cpu;
	}
}

struct migration_swap_arg {
	struct task_struct *src_task, *dst_task;
	int src_cpu, dst_cpu;
};

static int migrate_swap_stop(void *data)
{
	struct migration_swap_arg *arg = data;
	struct rq *src_rq, *dst_rq;
	int ret = -EAGAIN;

1285 1286 1287
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1288 1289 1290
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1291 1292
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1293
	double_rq_lock(src_rq, dst_rq);
1294

1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

	if (task_cpu(arg->src_task) != arg->src_cpu)
		goto unlock;

	if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task)))
		goto unlock;

	if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task)))
		goto unlock;

	__migrate_swap_task(arg->src_task, arg->dst_cpu);
	__migrate_swap_task(arg->dst_task, arg->src_cpu);

	ret = 0;

unlock:
	double_rq_unlock(src_rq, dst_rq);
1314 1315
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337

	return ret;
}

/*
 * Cross migrate two tasks
 */
int migrate_swap(struct task_struct *cur, struct task_struct *p)
{
	struct migration_swap_arg arg;
	int ret = -EINVAL;

	arg = (struct migration_swap_arg){
		.src_task = cur,
		.src_cpu = task_cpu(cur),
		.dst_task = p,
		.dst_cpu = task_cpu(p),
	};

	if (arg.src_cpu == arg.dst_cpu)
		goto out;

1338 1339 1340 1341
	/*
	 * These three tests are all lockless; this is OK since all of them
	 * will be re-checked with proper locks held further down the line.
	 */
1342 1343 1344 1345 1346 1347 1348 1349 1350
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

	if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task)))
		goto out;

	if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task)))
		goto out;

1351
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1352 1353 1354 1355 1356 1357
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1358 1359 1360
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1361 1362 1363 1364 1365 1366 1367
 * 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
1368 1369 1370 1371 1372 1373
 * 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
1374
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1375
{
1376
	int running, queued;
1377
	struct rq_flags rf;
R
Roland McGrath 已提交
1378
	unsigned long ncsw;
1379
	struct rq *rq;
L
Linus Torvalds 已提交
1380

1381 1382 1383 1384 1385 1386 1387 1388
	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);
1389

1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400
		/*
		 * 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
1401 1402 1403
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1404
			cpu_relax();
R
Roland McGrath 已提交
1405
		}
1406

1407 1408 1409 1410 1411
		/*
		 * 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.
		 */
1412
		rq = task_rq_lock(p, &rf);
1413
		trace_sched_wait_task(p);
1414
		running = task_running(rq, p);
1415
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1416
		ncsw = 0;
1417
		if (!match_state || p->state == match_state)
1418
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1419
		task_rq_unlock(rq, p, &rf);
1420

R
Roland McGrath 已提交
1421 1422 1423 1424 1425 1426
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1427 1428 1429 1430 1431 1432 1433 1434 1435 1436
		/*
		 * 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;
		}
1437

1438 1439 1440 1441 1442
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1443
		 * So if it was still runnable (but just not actively
1444 1445 1446
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1447
		if (unlikely(queued)) {
1448 1449 1450 1451
			ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1452 1453
			continue;
		}
1454

1455 1456 1457 1458 1459 1460 1461
		/*
		 * 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
1462 1463

	return ncsw;
L
Linus Torvalds 已提交
1464 1465 1466 1467 1468 1469 1470 1471 1472
}

/***
 * 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
1473
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1474 1475 1476 1477 1478
 * 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.
 */
1479
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1480 1481 1482 1483 1484 1485 1486 1487 1488
{
	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
1489
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1490

1491
/*
1492
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
 *  - on cpu-up we allow per-cpu kthreads on the online && !active cpu,
 *    see __set_cpus_allowed_ptr(). At this point the newly online
 *    cpu isn't yet part of the sched domains, and balancing will not
 *    see it.
 *
 *  - on cpu-down we clear cpu_active() to mask the sched domains and
 *    avoid the load balancer to place new tasks on the to be removed
 *    cpu. Existing tasks will remain running there and will be taken
 *    off.
 *
 * This means that fallback selection must not select !active CPUs.
 * And can assume that any active CPU must be online. Conversely
 * select_task_rq() below may allow selection of !active CPUs in order
 * to satisfy the above rules.
1512
 */
1513 1514
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1515 1516
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1517 1518
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1519

1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534
	/*
	 * 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_active(dest_cpu))
				continue;
			if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
				return dest_cpu;
		}
1535
	}
1536

1537 1538
	for (;;) {
		/* Any allowed, online CPU? */
1539
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1540 1541 1542
			if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu))
				continue;
			if (!cpu_online(dest_cpu))
1543 1544 1545
				continue;
			goto out;
		}
1546

1547
		/* No more Mr. Nice Guy. */
1548 1549
		switch (state) {
		case cpuset:
1550 1551 1552 1553 1554 1555
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
			/* fall-through */
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574
		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()) {
1575
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1576 1577
					task_pid_nr(p), p->comm, cpu);
		}
1578 1579 1580 1581 1582
	}

	return dest_cpu;
}

1583
/*
1584
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1585
 */
1586
static inline
1587
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1588
{
1589 1590
	lockdep_assert_held(&p->pi_lock);

1591
	if (tsk_nr_cpus_allowed(p) > 1)
1592
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1593 1594
	else
		cpu = cpumask_any(tsk_cpus_allowed(p));
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605

	/*
	 * 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 ]
	 */
1606
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1607
		     !cpu_online(cpu)))
1608
		cpu = select_fallback_rq(task_cpu(p), p);
1609 1610

	return cpu;
1611
}
1612 1613 1614 1615 1616 1617

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1618 1619 1620 1621 1622 1623 1624 1625 1626

#else

static inline int __set_cpus_allowed_ptr(struct task_struct *p,
					 const struct cpumask *new_mask, bool check)
{
	return set_cpus_allowed_ptr(p, new_mask);
}

P
Peter Zijlstra 已提交
1627
#endif /* CONFIG_SMP */
1628

P
Peter Zijlstra 已提交
1629
static void
1630
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1631
{
1632
	struct rq *rq;
1633

1634 1635 1636 1637
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1638

1639 1640
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1641 1642
		schedstat_inc(rq->ttwu_local);
		schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1643 1644 1645
	} else {
		struct sched_domain *sd;

1646
		schedstat_inc(p->se.statistics.nr_wakeups_remote);
1647
		rcu_read_lock();
1648
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1649
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1650
				schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1651 1652 1653
				break;
			}
		}
1654
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1655
	}
1656 1657

	if (wake_flags & WF_MIGRATED)
1658
		schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1659 1660
#endif /* CONFIG_SMP */

1661 1662
	schedstat_inc(rq->ttwu_count);
	schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1663 1664

	if (wake_flags & WF_SYNC)
1665
		schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1666 1667
}

1668
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1669
{
T
Tejun Heo 已提交
1670
	activate_task(rq, p, en_flags);
1671
	p->on_rq = TASK_ON_RQ_QUEUED;
1672 1673 1674 1675

	/* 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
1676 1677
}

1678 1679 1680
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1681 1682
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
			   struct pin_cookie cookie)
T
Tejun Heo 已提交
1683 1684 1685
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1686 1687
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1688
#ifdef CONFIG_SMP
1689 1690
	if (p->sched_class->task_woken) {
		/*
1691 1692
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1693
		 */
1694
		lockdep_unpin_lock(&rq->lock, cookie);
T
Tejun Heo 已提交
1695
		p->sched_class->task_woken(rq, p);
1696
		lockdep_repin_lock(&rq->lock, cookie);
1697
	}
T
Tejun Heo 已提交
1698

1699
	if (rq->idle_stamp) {
1700
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1701
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1702

1703 1704 1705
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1706
			rq->avg_idle = max;
1707

T
Tejun Heo 已提交
1708 1709 1710 1711 1712
		rq->idle_stamp = 0;
	}
#endif
}

1713
static void
1714 1715
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
		 struct pin_cookie cookie)
1716
{
1717 1718
	int en_flags = ENQUEUE_WAKEUP;

1719 1720
	lockdep_assert_held(&rq->lock);

1721 1722 1723
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1724 1725

	if (wake_flags & WF_MIGRATED)
1726
		en_flags |= ENQUEUE_MIGRATED;
1727 1728
#endif

1729
	ttwu_activate(rq, p, en_flags);
1730
	ttwu_do_wakeup(rq, p, wake_flags, cookie);
1731 1732 1733 1734 1735 1736 1737 1738 1739 1740
}

/*
 * 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)
{
1741
	struct rq_flags rf;
1742 1743 1744
	struct rq *rq;
	int ret = 0;

1745
	rq = __task_rq_lock(p, &rf);
1746
	if (task_on_rq_queued(p)) {
1747 1748
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1749
		ttwu_do_wakeup(rq, p, wake_flags, rf.cookie);
1750 1751
		ret = 1;
	}
1752
	__task_rq_unlock(rq, &rf);
1753 1754 1755 1756

	return ret;
}

1757
#ifdef CONFIG_SMP
1758
void sched_ttwu_pending(void)
1759 1760
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1761
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1762
	struct pin_cookie cookie;
P
Peter Zijlstra 已提交
1763
	struct task_struct *p;
1764
	unsigned long flags;
1765

1766 1767 1768 1769
	if (!llist)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
1770
	cookie = lockdep_pin_lock(&rq->lock);
1771

P
Peter Zijlstra 已提交
1772
	while (llist) {
P
Peter Zijlstra 已提交
1773 1774
		int wake_flags = 0;

P
Peter Zijlstra 已提交
1775 1776
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
P
Peter Zijlstra 已提交
1777 1778 1779 1780 1781

		if (p->sched_remote_wakeup)
			wake_flags = WF_MIGRATED;

		ttwu_do_activate(rq, p, wake_flags, cookie);
1782 1783
	}

1784
	lockdep_unpin_lock(&rq->lock, cookie);
1785
	raw_spin_unlock_irqrestore(&rq->lock, flags);
1786 1787 1788 1789
}

void scheduler_ipi(void)
{
1790 1791 1792 1793 1794
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1795
	preempt_fold_need_resched();
1796

1797
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813
		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();
P
Peter Zijlstra 已提交
1814
	sched_ttwu_pending();
1815 1816 1817 1818

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1819
	if (unlikely(got_nohz_idle_kick())) {
1820
		this_rq()->idle_balance = 1;
1821
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1822
	}
1823
	irq_exit();
1824 1825
}

P
Peter Zijlstra 已提交
1826
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1827
{
1828 1829
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1830 1831
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1832 1833 1834 1835 1836 1837
	if (llist_add(&p->wake_entry, &cpu_rq(cpu)->wake_list)) {
		if (!set_nr_if_polling(rq->idle))
			smp_send_reschedule(cpu);
		else
			trace_sched_wake_idle_without_ipi(cpu);
	}
1838
}
1839

1840 1841 1842 1843 1844
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

1845 1846 1847 1848
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1849 1850 1851 1852 1853 1854 1855 1856 1857 1858

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
		raw_spin_lock_irqsave(&rq->lock, flags);
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
		/* Else cpu is not in idle, do nothing here */
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}
1859 1860 1861

out:
	rcu_read_unlock();
1862 1863
}

1864
bool cpus_share_cache(int this_cpu, int that_cpu)
1865 1866 1867
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1868
#endif /* CONFIG_SMP */
1869

1870
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1871 1872
{
	struct rq *rq = cpu_rq(cpu);
1873
	struct pin_cookie cookie;
1874

1875
#if defined(CONFIG_SMP)
1876
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1877
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
P
Peter Zijlstra 已提交
1878
		ttwu_queue_remote(p, cpu, wake_flags);
1879 1880 1881 1882
		return;
	}
#endif

1883
	raw_spin_lock(&rq->lock);
1884
	cookie = lockdep_pin_lock(&rq->lock);
1885
	ttwu_do_activate(rq, p, wake_flags, cookie);
1886
	lockdep_unpin_lock(&rq->lock, cookie);
1887
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
1888 1889
}

1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939
/*
 * Notes on Program-Order guarantees on SMP systems.
 *
 *  MIGRATION
 *
 * The basic program-order guarantee on SMP systems is that when a task [t]
 * migrates, all its activity on its old cpu [c0] happens-before any subsequent
 * execution on its new cpu [c1].
 *
 * For migration (of runnable tasks) this is provided by the following means:
 *
 *  A) UNLOCK of the rq(c0)->lock scheduling out task t
 *  B) migration for t is required to synchronize *both* rq(c0)->lock and
 *     rq(c1)->lock (if not at the same time, then in that order).
 *  C) LOCK of the rq(c1)->lock scheduling in task
 *
 * Transitivity guarantees that B happens after A and C after B.
 * Note: we only require RCpc transitivity.
 * Note: the cpu doing B need not be c0 or c1
 *
 * Example:
 *
 *   CPU0            CPU1            CPU2
 *
 *   LOCK rq(0)->lock
 *   sched-out X
 *   sched-in Y
 *   UNLOCK rq(0)->lock
 *
 *                                   LOCK rq(0)->lock // orders against CPU0
 *                                   dequeue X
 *                                   UNLOCK rq(0)->lock
 *
 *                                   LOCK rq(1)->lock
 *                                   enqueue X
 *                                   UNLOCK rq(1)->lock
 *
 *                   LOCK rq(1)->lock // orders against CPU2
 *                   sched-out Z
 *                   sched-in X
 *                   UNLOCK rq(1)->lock
 *
 *
 *  BLOCKING -- aka. SLEEP + WAKEUP
 *
 * For blocking we (obviously) need to provide the same guarantee as for
 * migration. However the means are completely different as there is no lock
 * chain to provide order. Instead we do:
 *
 *   1) smp_store_release(X->on_cpu, 0)
1940
 *   2) smp_cond_load_acquire(!X->on_cpu)
1941 1942 1943 1944 1945 1946 1947 1948 1949 1950
 *
 * Example:
 *
 *   CPU0 (schedule)  CPU1 (try_to_wake_up) CPU2 (schedule)
 *
 *   LOCK rq(0)->lock LOCK X->pi_lock
 *   dequeue X
 *   sched-out X
 *   smp_store_release(X->on_cpu, 0);
 *
1951
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976
 *                    X->state = WAKING
 *                    set_task_cpu(X,2)
 *
 *                    LOCK rq(2)->lock
 *                    enqueue X
 *                    X->state = RUNNING
 *                    UNLOCK rq(2)->lock
 *
 *                                          LOCK rq(2)->lock // orders against CPU1
 *                                          sched-out Z
 *                                          sched-in X
 *                                          UNLOCK rq(2)->lock
 *
 *                    UNLOCK X->pi_lock
 *   UNLOCK rq(0)->lock
 *
 *
 * However; for wakeups there is a second guarantee we must provide, namely we
 * must observe the state that lead to our wakeup. That is, not only must our
 * task observe its own prior state, it must also observe the stores prior to
 * its wakeup.
 *
 * This means that any means of doing remote wakeups must order the CPU doing
 * the wakeup against the CPU the task is going to end up running on. This,
 * however, is already required for the regular Program-Order guarantee above,
1977
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1978 1979 1980
 *
 */

T
Tejun Heo 已提交
1981
/**
L
Linus Torvalds 已提交
1982
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1983
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1984
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1985
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1986 1987 1988 1989 1990 1991 1992
 *
 * 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.
 *
1993
 * Return: %true if @p was woken up, %false if it was already running.
T
Tejun Heo 已提交
1994
 * or @state didn't match @p's state.
L
Linus Torvalds 已提交
1995
 */
1996 1997
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1998 1999
{
	unsigned long flags;
2000
	int cpu, success = 0;
P
Peter Zijlstra 已提交
2001

2002 2003 2004 2005 2006 2007 2008
	/*
	 * If we are going to wake up a thread waiting for CONDITION we
	 * need to ensure that CONDITION=1 done by the caller can not be
	 * reordered with p->state check below. This pairs with mb() in
	 * set_current_state() the waiting thread does.
	 */
	smp_mb__before_spinlock();
2009
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
2010
	if (!(p->state & state))
L
Linus Torvalds 已提交
2011 2012
		goto out;

2013 2014
	trace_sched_waking(p);

2015
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
2016 2017
	cpu = task_cpu(p);

2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039
	/*
	 * Ensure we load p->on_rq _after_ p->state, otherwise it would
	 * be possible to, falsely, observe p->on_rq == 0 and get stuck
	 * in smp_cond_load_acquire() below.
	 *
	 * sched_ttwu_pending()                 try_to_wake_up()
	 *   [S] p->on_rq = 1;                  [L] P->state
	 *       UNLOCK rq->lock  -----.
	 *                              \
	 *				 +---   RMB
	 * schedule()                   /
	 *       LOCK rq->lock    -----'
	 *       UNLOCK rq->lock
	 *
	 * [task p]
	 *   [S] p->state = UNINTERRUPTIBLE     [L] p->on_rq
	 *
	 * Pairs with the UNLOCK+LOCK on rq->lock from the
	 * last wakeup of our task and the schedule that got our task
	 * current.
	 */
	smp_rmb();
2040 2041
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2042 2043

#ifdef CONFIG_SMP
2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062
	/*
	 * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
	 * possible to, falsely, observe p->on_cpu == 0.
	 *
	 * One must be running (->on_cpu == 1) in order to remove oneself
	 * from the runqueue.
	 *
	 *  [S] ->on_cpu = 1;	[L] ->on_rq
	 *      UNLOCK rq->lock
	 *			RMB
	 *      LOCK   rq->lock
	 *  [S] ->on_rq = 0;    [L] ->on_cpu
	 *
	 * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock
	 * from the consecutive calls to schedule(); the first switching to our
	 * task, the second putting it to sleep.
	 */
	smp_rmb();

P
Peter Zijlstra 已提交
2063
	/*
2064 2065
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
2066 2067 2068 2069 2070
	 *
	 * Pairs with the smp_store_release() in finish_lock_switch().
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2071
	 */
2072
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2073

2074
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2075
	p->state = TASK_WAKING;
2076

2077
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2078 2079
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2080
		set_task_cpu(p, cpu);
2081
	}
L
Linus Torvalds 已提交
2082 2083
#endif /* CONFIG_SMP */

2084
	ttwu_queue(p, cpu, wake_flags);
2085
stat:
2086
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2087
out:
2088
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2089 2090 2091 2092

	return success;
}

T
Tejun Heo 已提交
2093 2094 2095
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2096
 * @cookie: context's cookie for pinning
T
Tejun Heo 已提交
2097
 *
2098
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2099
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2100
 * the current task.
T
Tejun Heo 已提交
2101
 */
2102
static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie)
T
Tejun Heo 已提交
2103 2104 2105
{
	struct rq *rq = task_rq(p);

2106 2107 2108 2109
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2110 2111
	lockdep_assert_held(&rq->lock);

2112
	if (!raw_spin_trylock(&p->pi_lock)) {
2113 2114 2115 2116 2117 2118
		/*
		 * This is OK, because current is on_cpu, which avoids it being
		 * picked for load-balance and preemption/IRQs are still
		 * disabled avoiding further scheduler activity on it and we've
		 * not yet picked a replacement task.
		 */
2119
		lockdep_unpin_lock(&rq->lock, cookie);
2120 2121 2122
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
2123
		lockdep_repin_lock(&rq->lock, cookie);
2124 2125
	}

T
Tejun Heo 已提交
2126
	if (!(p->state & TASK_NORMAL))
2127
		goto out;
T
Tejun Heo 已提交
2128

2129 2130
	trace_sched_waking(p);

2131
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
2132 2133
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

2134
	ttwu_do_wakeup(rq, p, 0, cookie);
2135
	ttwu_stat(p, smp_processor_id(), 0);
2136 2137
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2138 2139
}

2140 2141 2142 2143 2144
/**
 * 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
2145 2146 2147
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2148 2149 2150 2151
 *
 * 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.
 */
2152
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2153
{
2154
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2155 2156 2157
}
EXPORT_SYMBOL(wake_up_process);

2158
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2159 2160 2161 2162
{
	return try_to_wake_up(p, state, 0);
}

2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174
/*
 * This function clears the sched_dl_entity static params.
 */
void __dl_clear_params(struct task_struct *p)
{
	struct sched_dl_entity *dl_se = &p->dl;

	dl_se->dl_runtime = 0;
	dl_se->dl_deadline = 0;
	dl_se->dl_period = 0;
	dl_se->flags = 0;
	dl_se->dl_bw = 0;
2175 2176 2177

	dl_se->dl_throttled = 0;
	dl_se->dl_yielded = 0;
2178 2179
}

L
Linus Torvalds 已提交
2180 2181 2182
/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2183 2184 2185
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2186
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2187
{
P
Peter Zijlstra 已提交
2188 2189 2190
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2191 2192
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2193
	p->se.prev_sum_exec_runtime	= 0;
2194
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2195
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2196
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2197

2198 2199 2200 2201
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2202
#ifdef CONFIG_SCHEDSTATS
2203
	/* Even if schedstat is disabled, there should not be garbage */
2204
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2205
#endif
N
Nick Piggin 已提交
2206

2207
	RB_CLEAR_NODE(&p->dl.rb_node);
2208
	init_dl_task_timer(&p->dl);
2209
	__dl_clear_params(p);
2210

P
Peter Zijlstra 已提交
2211
	INIT_LIST_HEAD(&p->rt.run_list);
2212 2213 2214 2215
	p->rt.timeout		= 0;
	p->rt.time_slice	= sched_rr_timeslice;
	p->rt.on_rq		= 0;
	p->rt.on_list		= 0;
N
Nick Piggin 已提交
2216

2217 2218 2219
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2220 2221 2222

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2223
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2224 2225 2226
		p->mm->numa_scan_seq = 0;
	}

2227 2228 2229 2230 2231
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2232 2233
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2234
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2235
	p->numa_work.next = &p->numa_work;
2236
	p->numa_faults = NULL;
2237 2238
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2239 2240

	p->numa_group = NULL;
2241
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2242 2243
}

2244 2245
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2246
#ifdef CONFIG_NUMA_BALANCING
2247

2248 2249 2250
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2251
		static_branch_enable(&sched_numa_balancing);
2252
	else
2253
		static_branch_disable(&sched_numa_balancing);
2254
}
2255 2256 2257 2258 2259 2260 2261

#ifdef CONFIG_PROC_SYSCTL
int sysctl_numa_balancing(struct ctl_table *table, int write,
			 void __user *buffer, size_t *lenp, loff_t *ppos)
{
	struct ctl_table t;
	int err;
2262
	int state = static_branch_likely(&sched_numa_balancing);
2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277

	if (write && !capable(CAP_SYS_ADMIN))
		return -EPERM;

	t = *table;
	t.data = &state;
	err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
	if (err < 0)
		return err;
	if (write)
		set_numabalancing_state(state);
	return err;
}
#endif
#endif
I
Ingo Molnar 已提交
2278

2279 2280
#ifdef CONFIG_SCHEDSTATS

2281
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2282
static bool __initdata __sched_schedstats = false;
2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305

static void set_schedstats(bool enabled)
{
	if (enabled)
		static_branch_enable(&sched_schedstats);
	else
		static_branch_disable(&sched_schedstats);
}

void force_schedstat_enabled(void)
{
	if (!schedstat_enabled()) {
		pr_info("kernel profiling enabled schedstats, disable via kernel.sched_schedstats.\n");
		static_branch_enable(&sched_schedstats);
	}
}

static int __init setup_schedstats(char *str)
{
	int ret = 0;
	if (!str)
		goto out;

2306 2307 2308 2309 2310
	/*
	 * This code is called before jump labels have been set up, so we can't
	 * change the static branch directly just yet.  Instead set a temporary
	 * variable so init_schedstats() can do it later.
	 */
2311
	if (!strcmp(str, "enable")) {
2312
		__sched_schedstats = true;
2313 2314
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2315
		__sched_schedstats = false;
2316 2317 2318 2319 2320 2321 2322 2323 2324 2325
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

	return ret;
}
__setup("schedstats=", setup_schedstats);

2326 2327 2328 2329 2330
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350
#ifdef CONFIG_PROC_SYSCTL
int sysctl_schedstats(struct ctl_table *table, int write,
			 void __user *buffer, size_t *lenp, loff_t *ppos)
{
	struct ctl_table t;
	int err;
	int state = static_branch_likely(&sched_schedstats);

	if (write && !capable(CAP_SYS_ADMIN))
		return -EPERM;

	t = *table;
	t.data = &state;
	err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
	if (err < 0)
		return err;
	if (write)
		set_schedstats(state);
	return err;
}
2351 2352 2353 2354
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2355 2356 2357 2358

/*
 * fork()/clone()-time setup:
 */
2359
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2360
{
2361
	unsigned long flags;
I
Ingo Molnar 已提交
2362 2363
	int cpu = get_cpu();

2364
	__sched_fork(clone_flags, p);
2365
	/*
2366
	 * We mark the process as NEW here. This guarantees that
2367 2368 2369
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2370
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2371

2372 2373 2374 2375 2376
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2377 2378 2379 2380
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2381
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2382
			p->policy = SCHED_NORMAL;
2383
			p->static_prio = NICE_TO_PRIO(0);
2384 2385 2386 2387 2388 2389
			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);
2390

2391 2392 2393 2394 2395 2396
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2397

2398 2399 2400 2401 2402 2403
	if (dl_prio(p->prio)) {
		put_cpu();
		return -EAGAIN;
	} else if (rt_prio(p->prio)) {
		p->sched_class = &rt_sched_class;
	} else {
H
Hiroshi Shimamoto 已提交
2404
		p->sched_class = &fair_sched_class;
2405
	}
2406

2407
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2408

2409 2410 2411 2412 2413 2414 2415
	/*
	 * 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.
	 */
2416
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2417 2418 2419 2420 2421 2422 2423
	/*
	 * We're setting the cpu for the first time, we don't migrate,
	 * so use __set_task_cpu().
	 */
	__set_task_cpu(p, cpu);
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2424
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2425

2426
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2427
	if (likely(sched_info_on()))
2428
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2429
#endif
P
Peter Zijlstra 已提交
2430 2431
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2432
#endif
2433
	init_task_preempt_count(p);
2434
#ifdef CONFIG_SMP
2435
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2436
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2437
#endif
2438

N
Nick Piggin 已提交
2439
	put_cpu();
2440
	return 0;
L
Linus Torvalds 已提交
2441 2442
}

2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
		return 1ULL << 20;

	/*
	 * Doing this here saves a lot of checks in all
	 * the calling paths, and returning zero seems
	 * safe for them anyway.
	 */
	if (period == 0)
		return 0;

	return div64_u64(runtime << 20, period);
}

#ifdef CONFIG_SMP
inline struct dl_bw *dl_bw_of(int i)
{
2462 2463
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2464 2465 2466
	return &cpu_rq(i)->rd->dl_bw;
}

2467
static inline int dl_bw_cpus(int i)
2468
{
2469 2470 2471
	struct root_domain *rd = cpu_rq(i)->rd;
	int cpus = 0;

2472 2473
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2474 2475 2476 2477
	for_each_cpu_and(i, rd->span, cpu_active_mask)
		cpus++;

	return cpus;
2478 2479 2480 2481 2482 2483 2484
}
#else
inline struct dl_bw *dl_bw_of(int i)
{
	return &cpu_rq(i)->dl.dl_bw;
}

2485
static inline int dl_bw_cpus(int i)
2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497
{
	return 1;
}
#endif

/*
 * We must be sure that accepting a new task (or allowing changing the
 * parameters of an existing one) is consistent with the bandwidth
 * constraints. If yes, this function also accordingly updates the currently
 * allocated bandwidth to reflect the new situation.
 *
 * This function is called while holding p's rq->lock.
2498 2499 2500
 *
 * XXX we should delay bw change until the task's 0-lag point, see
 * __setparam_dl().
2501 2502 2503 2504 2505 2506
 */
static int dl_overflow(struct task_struct *p, int policy,
		       const struct sched_attr *attr)
{

	struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
2507
	u64 period = attr->sched_period ?: attr->sched_deadline;
2508 2509
	u64 runtime = attr->sched_runtime;
	u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2510
	int cpus, err = -1;
2511

2512 2513
	/* !deadline task may carry old deadline bandwidth */
	if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2514 2515 2516 2517 2518 2519 2520 2521
		return 0;

	/*
	 * Either if a task, enters, leave, or stays -deadline but changes
	 * its parameters, we may need to update accordingly the total
	 * allocated bandwidth of the container.
	 */
	raw_spin_lock(&dl_b->lock);
2522
	cpus = dl_bw_cpus(task_cpu(p));
2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542
	if (dl_policy(policy) && !task_has_dl_policy(p) &&
	    !__dl_overflow(dl_b, cpus, 0, new_bw)) {
		__dl_add(dl_b, new_bw);
		err = 0;
	} else if (dl_policy(policy) && task_has_dl_policy(p) &&
		   !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) {
		__dl_clear(dl_b, p->dl.dl_bw);
		__dl_add(dl_b, new_bw);
		err = 0;
	} else if (!dl_policy(policy) && task_has_dl_policy(p)) {
		__dl_clear(dl_b, p->dl.dl_bw);
		err = 0;
	}
	raw_spin_unlock(&dl_b->lock);

	return err;
}

extern void init_dl_bw(struct dl_bw *dl_b);

L
Linus Torvalds 已提交
2543 2544 2545 2546 2547 2548 2549
/*
 * 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.
 */
2550
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2551
{
2552
	struct rq_flags rf;
I
Ingo Molnar 已提交
2553
	struct rq *rq;
2554

2555
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2556
	p->state = TASK_RUNNING;
2557 2558 2559 2560 2561
#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
2562 2563 2564
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2565
	 */
2566
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2567
#endif
2568
	rq = __task_rq_lock(p, &rf);
2569
	post_init_entity_util_avg(&p->se);
2570

P
Peter Zijlstra 已提交
2571
	activate_task(rq, p, 0);
2572
	p->on_rq = TASK_ON_RQ_QUEUED;
2573
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2574
	check_preempt_curr(rq, p, WF_FORK);
2575
#ifdef CONFIG_SMP
2576 2577 2578 2579 2580
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2581
		lockdep_unpin_lock(&rq->lock, rf.cookie);
2582
		p->sched_class->task_woken(rq, p);
2583
		lockdep_repin_lock(&rq->lock, rf.cookie);
2584
	}
2585
#endif
2586
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2587 2588
}

2589 2590
#ifdef CONFIG_PREEMPT_NOTIFIERS

2591 2592
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604
void preempt_notifier_inc(void)
{
	static_key_slow_inc(&preempt_notifier_key);
}
EXPORT_SYMBOL_GPL(preempt_notifier_inc);

void preempt_notifier_dec(void)
{
	static_key_slow_dec(&preempt_notifier_key);
}
EXPORT_SYMBOL_GPL(preempt_notifier_dec);

2605
/**
2606
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2607
 * @notifier: notifier struct to register
2608 2609 2610
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2611 2612 2613
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2614 2615 2616 2617 2618 2619
	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
2620
 * @notifier: notifier struct to unregister
2621
 *
2622
 * This is *not* safe to call from within a preemption notifier.
2623 2624 2625 2626 2627 2628 2629
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2630
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2631 2632 2633
{
	struct preempt_notifier *notifier;

2634
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2635 2636 2637
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2638 2639 2640 2641 2642 2643
static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
	if (static_key_false(&preempt_notifier_key))
		__fire_sched_in_preempt_notifiers(curr);
}

2644
static void
2645 2646
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2647 2648 2649
{
	struct preempt_notifier *notifier;

2650
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2651 2652 2653
		notifier->ops->sched_out(notifier, next);
}

2654 2655 2656 2657 2658 2659 2660 2661
static __always_inline void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
	if (static_key_false(&preempt_notifier_key))
		__fire_sched_out_preempt_notifiers(curr, next);
}

2662
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2663

2664
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2665 2666 2667
{
}

2668
static inline void
2669 2670 2671 2672 2673
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2674
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2675

2676 2677 2678
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2679
 * @prev: the current task that is being switched out
2680 2681 2682 2683 2684 2685 2686 2687 2688
 * @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.
 */
2689 2690 2691
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2692
{
2693
	sched_info_switch(rq, prev, next);
2694
	perf_event_task_sched_out(prev, next);
2695
	fire_sched_out_preempt_notifiers(prev, next);
2696 2697 2698 2699
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2700 2701 2702 2703
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2704 2705 2706 2707
 * 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
2708 2709
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2710
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2711 2712
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2713 2714 2715 2716 2717
 *
 * 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 we need to recalculate this_rq
 * because prev may have moved to another CPU.
L
Linus Torvalds 已提交
2718
 */
2719
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2720 2721
	__releases(rq->lock)
{
2722
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2723
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2724
	long prev_state;
L
Linus Torvalds 已提交
2725

2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736
	/*
	 * The previous task will have left us with a preempt_count of 2
	 * because it left us after:
	 *
	 *	schedule()
	 *	  preempt_disable();			// 1
	 *	  __schedule()
	 *	    raw_spin_lock_irq(&rq->lock)	// 2
	 *
	 * Also, see FORK_PREEMPT_COUNT.
	 */
2737 2738 2739 2740
	if (WARN_ONCE(preempt_count() != 2*PREEMPT_DISABLE_OFFSET,
		      "corrupted preempt_count: %s/%d/0x%x\n",
		      current->comm, current->pid, preempt_count()))
		preempt_count_set(FORK_PREEMPT_COUNT);
2741

L
Linus Torvalds 已提交
2742 2743 2744 2745
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2746
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2747 2748
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2749 2750 2751 2752 2753
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
	 * finish_lock_switch), otherwise a concurrent wakeup can get prev
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2754
	 */
O
Oleg Nesterov 已提交
2755
	prev_state = prev->state;
2756
	vtime_task_switch(prev);
2757
	perf_event_task_sched_in(prev, current);
2758
	finish_lock_switch(rq, prev);
2759
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2760

2761
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
2762 2763
	if (mm)
		mmdrop(mm);
2764
	if (unlikely(prev_state == TASK_DEAD)) {
2765 2766 2767
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2768 2769 2770
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2771
		 */
2772
		kprobe_flush_task(prev);
L
Linus Torvalds 已提交
2773
		put_task_struct(prev);
2774
	}
2775

2776
	tick_nohz_task_switch();
2777
	return rq;
L
Linus Torvalds 已提交
2778 2779
}

2780 2781 2782
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2783
static void __balance_callback(struct rq *rq)
2784
{
2785 2786 2787
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2788

2789 2790 2791 2792 2793 2794 2795 2796
	raw_spin_lock_irqsave(&rq->lock, flags);
	head = rq->balance_callback;
	rq->balance_callback = NULL;
	while (head) {
		func = (void (*)(struct rq *))head->func;
		next = head->next;
		head->next = NULL;
		head = next;
2797

2798
		func(rq);
2799
	}
2800 2801 2802 2803 2804 2805 2806
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2807 2808 2809
}

#else
2810

2811
static inline void balance_callback(struct rq *rq)
2812
{
L
Linus Torvalds 已提交
2813 2814
}

2815 2816
#endif

L
Linus Torvalds 已提交
2817 2818 2819 2820
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2821
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2822 2823
	__releases(rq->lock)
{
2824
	struct rq *rq;
2825

2826 2827 2828 2829 2830 2831 2832 2833 2834
	/*
	 * New tasks start with FORK_PREEMPT_COUNT, see there and
	 * finish_task_switch() for details.
	 *
	 * finish_task_switch() will drop rq->lock() and lower preempt_count
	 * and the preempt_enable() will end up enabling preemption (on
	 * PREEMPT_COUNT kernels).
	 */

2835
	rq = finish_task_switch(prev);
2836
	balance_callback(rq);
2837
	preempt_enable();
2838

L
Linus Torvalds 已提交
2839
	if (current->set_child_tid)
2840
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2841 2842 2843
}

/*
2844
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2845
 */
2846
static __always_inline struct rq *
2847
context_switch(struct rq *rq, struct task_struct *prev,
2848
	       struct task_struct *next, struct pin_cookie cookie)
L
Linus Torvalds 已提交
2849
{
I
Ingo Molnar 已提交
2850
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2851

2852
	prepare_task_switch(rq, prev, next);
2853

I
Ingo Molnar 已提交
2854 2855
	mm = next->mm;
	oldmm = prev->active_mm;
2856 2857 2858 2859 2860
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2861
	arch_start_context_switch(prev);
2862

2863
	if (!mm) {
L
Linus Torvalds 已提交
2864 2865 2866 2867
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
2868
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2869

2870
	if (!prev->mm) {
L
Linus Torvalds 已提交
2871 2872 2873
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2874 2875 2876 2877 2878 2879
	/*
	 * 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:
	 */
2880
	lockdep_unpin_lock(&rq->lock, cookie);
2881
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2882 2883 2884

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2885
	barrier();
2886 2887

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2888 2889 2890
}

/*
2891
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2892 2893
 *
 * externally visible scheduler statistics: current number of runnable
2894
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2895 2896 2897 2898 2899 2900 2901 2902 2903
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2904
}
L
Linus Torvalds 已提交
2905

2906 2907
/*
 * Check if only the current task is running on the cpu.
2908 2909 2910 2911 2912 2913 2914 2915 2916 2917
 *
 * Caution: this function does not check that the caller has disabled
 * preemption, thus the result might have a time-of-check-to-time-of-use
 * race.  The caller is responsible to use it correctly, for example:
 *
 * - from a non-preemptable section (of course)
 *
 * - from a thread that is bound to a single CPU
 *
 * - in a loop with very short iterations (e.g. a polling loop)
2918 2919 2920
 */
bool single_task_running(void)
{
2921
	return raw_rq()->nr_running == 1;
2922 2923 2924
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2925
unsigned long long nr_context_switches(void)
2926
{
2927 2928
	int i;
	unsigned long long sum = 0;
2929

2930
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2931
		sum += cpu_rq(i)->nr_switches;
2932

L
Linus Torvalds 已提交
2933 2934
	return sum;
}
2935

L
Linus Torvalds 已提交
2936 2937 2938
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2939

2940
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2941
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2942

L
Linus Torvalds 已提交
2943 2944
	return sum;
}
2945

2946
unsigned long nr_iowait_cpu(int cpu)
2947
{
2948
	struct rq *this = cpu_rq(cpu);
2949 2950
	return atomic_read(&this->nr_iowait);
}
2951

2952 2953
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2954 2955 2956
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2957 2958
}

I
Ingo Molnar 已提交
2959
#ifdef CONFIG_SMP
2960

2961
/*
P
Peter Zijlstra 已提交
2962 2963
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2964
 */
P
Peter Zijlstra 已提交
2965
void sched_exec(void)
2966
{
P
Peter Zijlstra 已提交
2967
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2968
	unsigned long flags;
2969
	int dest_cpu;
2970

2971
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2972
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2973 2974
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2975

2976
	if (likely(cpu_active(dest_cpu))) {
2977
		struct migration_arg arg = { p, dest_cpu };
2978

2979 2980
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2981 2982
		return;
	}
2983
unlock:
2984
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2985
}
I
Ingo Molnar 已提交
2986

L
Linus Torvalds 已提交
2987 2988 2989
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2990
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2991 2992

EXPORT_PER_CPU_SYMBOL(kstat);
2993
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2994

2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011
/*
 * The function fair_sched_class.update_curr accesses the struct curr
 * and its field curr->exec_start; when called from task_sched_runtime(),
 * we observe a high rate of cache misses in practice.
 * Prefetching this data results in improved performance.
 */
static inline void prefetch_curr_exec_start(struct task_struct *p)
{
#ifdef CONFIG_FAIR_GROUP_SCHED
	struct sched_entity *curr = (&p->se)->cfs_rq->curr;
#else
	struct sched_entity *curr = (&task_rq(p)->cfs)->curr;
#endif
	prefetch(curr);
	prefetch(&curr->exec_start);
}

3012 3013 3014 3015 3016 3017 3018
/*
 * 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)
{
3019
	struct rq_flags rf;
3020
	struct rq *rq;
3021
	u64 ns;
3022

3023 3024 3025 3026 3027 3028 3029 3030 3031
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
	 * 64-bit doesn't need locks to atomically read a 64bit value.
	 * So we have a optimization chance when the task's delta_exec is 0.
	 * Reading ->on_cpu is racy, but this is ok.
	 *
	 * If we race with it leaving cpu, we'll take a lock. So we're correct.
	 * If we race with it entering cpu, unaccounted time is 0. This is
	 * indistinguishable from the read occurring a few cycles earlier.
3032 3033
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3034
	 */
3035
	if (!p->on_cpu || !task_on_rq_queued(p))
3036 3037 3038
		return p->se.sum_exec_runtime;
#endif

3039
	rq = task_rq_lock(p, &rf);
3040 3041 3042 3043 3044 3045
	/*
	 * Must be ->curr _and_ ->on_rq.  If dequeued, we would
	 * project cycles that may never be accounted to this
	 * thread, breaking clock_gettime().
	 */
	if (task_current(rq, p) && task_on_rq_queued(p)) {
3046
		prefetch_curr_exec_start(p);
3047 3048 3049 3050
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3051
	task_rq_unlock(rq, p, &rf);
3052 3053 3054

	return ns;
}
3055

3056 3057 3058 3059 3060 3061 3062 3063
/*
 * 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
3064
	struct task_struct *curr = rq->curr;
3065 3066

	sched_clock_tick();
I
Ingo Molnar 已提交
3067

3068
	raw_spin_lock(&rq->lock);
3069
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3070
	curr->sched_class->task_tick(rq, curr, 0);
3071
	cpu_load_update_active(rq);
3072
	calc_global_load_tick(rq);
3073
	raw_spin_unlock(&rq->lock);
3074

3075
	perf_event_task_tick();
3076

3077
#ifdef CONFIG_SMP
3078
	rq->idle_balance = idle_cpu(cpu);
3079
	trigger_load_balance(rq);
3080
#endif
3081
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
3082 3083
}

3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094
#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.
3095 3096
 *
 * Return: Maximum deferment in nanoseconds.
3097 3098 3099 3100
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
3101
	unsigned long next, now = READ_ONCE(jiffies);
3102 3103 3104 3105 3106 3107

	next = rq->last_sched_tick + HZ;

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

3108
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
3109
}
3110
#endif
L
Linus Torvalds 已提交
3111

3112 3113
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127
/*
 * If the value passed in is equal to the current preempt count
 * then we just disabled preemption. Start timing the latency.
 */
static inline void preempt_latency_start(int val)
{
	if (preempt_count() == val) {
		unsigned long ip = get_lock_parent_ip();
#ifdef CONFIG_DEBUG_PREEMPT
		current->preempt_disable_ip = ip;
#endif
		trace_preempt_off(CALLER_ADDR0, ip);
	}
}
3128

3129
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3130
{
3131
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3132 3133 3134
	/*
	 * Underflow?
	 */
3135 3136
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3137
#endif
3138
	__preempt_count_add(val);
3139
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3140 3141 3142
	/*
	 * Spinlock count overflowing soon?
	 */
3143 3144
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3145
#endif
3146
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3147
}
3148
EXPORT_SYMBOL(preempt_count_add);
3149
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3150

3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
/*
 * If the value passed in equals to the current preempt count
 * then we just enabled preemption. Stop timing the latency.
 */
static inline void preempt_latency_stop(int val)
{
	if (preempt_count() == val)
		trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
}

3161
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3162
{
3163
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3164 3165 3166
	/*
	 * Underflow?
	 */
3167
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3168
		return;
L
Linus Torvalds 已提交
3169 3170 3171
	/*
	 * Is the spinlock portion underflowing?
	 */
3172 3173 3174
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3175
#endif
3176

3177
	preempt_latency_stop(val);
3178
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3179
}
3180
EXPORT_SYMBOL(preempt_count_sub);
3181
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3182

3183 3184 3185
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3186 3187 3188
#endif

/*
I
Ingo Molnar 已提交
3189
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3190
 */
I
Ingo Molnar 已提交
3191
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3192
{
3193 3194 3195
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3196 3197 3198
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3199 3200
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3201

I
Ingo Molnar 已提交
3202
	debug_show_held_locks(prev);
3203
	print_modules();
I
Ingo Molnar 已提交
3204 3205
	if (irqs_disabled())
		print_irqtrace_events(prev);
3206 3207
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3208
		pr_err("Preemption disabled at:");
3209
		print_ip_sym(preempt_disable_ip);
3210 3211
		pr_cont("\n");
	}
3212 3213 3214
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3215
	dump_stack();
3216
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3217
}
L
Linus Torvalds 已提交
3218

I
Ingo Molnar 已提交
3219 3220 3221 3222 3223
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3224
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3225 3226
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3227
#endif
3228

3229
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3230
		__schedule_bug(prev);
3231 3232
		preempt_count_set(PREEMPT_DISABLED);
	}
3233
	rcu_sleep_check();
I
Ingo Molnar 已提交
3234

L
Linus Torvalds 已提交
3235 3236
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3237
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3238 3239 3240 3241 3242 3243
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3244
pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
I
Ingo Molnar 已提交
3245
{
3246
	const struct sched_class *class = &fair_sched_class;
I
Ingo Molnar 已提交
3247
	struct task_struct *p;
L
Linus Torvalds 已提交
3248 3249

	/*
I
Ingo Molnar 已提交
3250 3251
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
3252
	 */
3253
	if (likely(prev->sched_class == class &&
3254
		   rq->nr_running == rq->cfs.h_nr_running)) {
3255
		p = fair_sched_class.pick_next_task(rq, prev, cookie);
3256 3257 3258 3259 3260
		if (unlikely(p == RETRY_TASK))
			goto again;

		/* assumes fair_sched_class->next == idle_sched_class */
		if (unlikely(!p))
3261
			p = idle_sched_class.pick_next_task(rq, prev, cookie);
3262 3263

		return p;
L
Linus Torvalds 已提交
3264 3265
	}

3266
again:
3267
	for_each_class(class) {
3268
		p = class->pick_next_task(rq, prev, cookie);
3269 3270 3271
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3272
			return p;
3273
		}
I
Ingo Molnar 已提交
3274
	}
3275 3276

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
3277
}
L
Linus Torvalds 已提交
3278

I
Ingo Molnar 已提交
3279
/*
3280
 * __schedule() is the main scheduler function.
3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314
 *
 * 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
3315
 *
3316
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3317
 */
3318
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3319 3320
{
	struct task_struct *prev, *next;
3321
	unsigned long *switch_count;
3322
	struct pin_cookie cookie;
I
Ingo Molnar 已提交
3323
	struct rq *rq;
3324
	int cpu;
I
Ingo Molnar 已提交
3325 3326 3327 3328 3329

	cpu = smp_processor_id();
	rq = cpu_rq(cpu);
	prev = rq->curr;

3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340
	/*
	 * do_exit() calls schedule() with preemption disabled as an exception;
	 * however we must fix that up, otherwise the next task will see an
	 * inconsistent (higher) preempt count.
	 *
	 * It also avoids the below schedule_debug() test from complaining
	 * about this.
	 */
	if (unlikely(prev->state == TASK_DEAD))
		preempt_enable_no_resched_notrace();

I
Ingo Molnar 已提交
3341
	schedule_debug(prev);
L
Linus Torvalds 已提交
3342

3343
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3344
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3345

3346 3347 3348
	local_irq_disable();
	rcu_note_context_switch();

3349 3350 3351 3352 3353 3354
	/*
	 * Make sure that signal_pending_state()->signal_pending() below
	 * can't be reordered with __set_current_state(TASK_INTERRUPTIBLE)
	 * done by the caller to avoid the race with signal_wake_up().
	 */
	smp_mb__before_spinlock();
3355
	raw_spin_lock(&rq->lock);
3356
	cookie = lockdep_pin_lock(&rq->lock);
L
Linus Torvalds 已提交
3357

3358 3359
	rq->clock_skip_update <<= 1; /* promote REQ to ACT */

3360
	switch_count = &prev->nivcsw;
3361
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3362
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3363
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3364
		} else {
3365 3366 3367
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
3368
			/*
3369 3370 3371
			 * 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
3372 3373 3374 3375
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3376
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3377
				if (to_wakeup)
3378
					try_to_wake_up_local(to_wakeup, cookie);
T
Tejun Heo 已提交
3379 3380
			}
		}
I
Ingo Molnar 已提交
3381
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3382 3383
	}

3384
	if (task_on_rq_queued(prev))
3385 3386
		update_rq_clock(rq);

3387
	next = pick_next_task(rq, prev, cookie);
3388
	clear_tsk_need_resched(prev);
3389
	clear_preempt_need_resched();
3390
	rq->clock_skip_update = 0;
L
Linus Torvalds 已提交
3391 3392 3393 3394 3395 3396

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

3397
		trace_sched_switch(preempt, prev, next);
3398
		rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */
3399
	} else {
3400
		lockdep_unpin_lock(&rq->lock, cookie);
3401
		raw_spin_unlock_irq(&rq->lock);
3402
	}
L
Linus Torvalds 已提交
3403

3404
	balance_callback(rq);
L
Linus Torvalds 已提交
3405
}
3406
STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */
3407

3408 3409
static inline void sched_submit_work(struct task_struct *tsk)
{
3410
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3411 3412 3413 3414 3415 3416 3417 3418 3419
		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);
}

3420
asmlinkage __visible void __sched schedule(void)
3421
{
3422 3423 3424
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3425
	do {
3426
		preempt_disable();
3427
		__schedule(false);
3428
		sched_preempt_enable_no_resched();
3429
	} while (need_resched());
3430
}
L
Linus Torvalds 已提交
3431 3432
EXPORT_SYMBOL(schedule);

3433
#ifdef CONFIG_CONTEXT_TRACKING
3434
asmlinkage __visible void __sched schedule_user(void)
3435 3436 3437 3438 3439 3440
{
	/*
	 * 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.
3441 3442
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3443
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3444
	 * too frequently to make sense yet.
3445
	 */
3446
	enum ctx_state prev_state = exception_enter();
3447
	schedule();
3448
	exception_exit(prev_state);
3449 3450 3451
}
#endif

3452 3453 3454 3455 3456 3457 3458
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3459
	sched_preempt_enable_no_resched();
3460 3461 3462 3463
	schedule();
	preempt_disable();
}

3464
static void __sched notrace preempt_schedule_common(void)
3465 3466
{
	do {
3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479
		/*
		 * Because the function tracer can trace preempt_count_sub()
		 * and it also uses preempt_enable/disable_notrace(), if
		 * NEED_RESCHED is set, the preempt_enable_notrace() called
		 * by the function tracer will call this function again and
		 * cause infinite recursion.
		 *
		 * Preemption must be disabled here before the function
		 * tracer can trace. Break up preempt_disable() into two
		 * calls. One to disable preemption without fear of being
		 * traced. The other to still record the preemption latency,
		 * which can also be traced by the function tracer.
		 */
3480
		preempt_disable_notrace();
3481
		preempt_latency_start(1);
3482
		__schedule(true);
3483
		preempt_latency_stop(1);
3484
		preempt_enable_no_resched_notrace();
3485 3486 3487 3488 3489 3490 3491 3492

		/*
		 * Check again in case we missed a preemption opportunity
		 * between schedule and now.
		 */
	} while (need_resched());
}

L
Linus Torvalds 已提交
3493 3494
#ifdef CONFIG_PREEMPT
/*
3495
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3496
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3497 3498
 * occur there and call schedule directly.
 */
3499
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3500 3501 3502
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3503
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3504
	 */
3505
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3506 3507
		return;

3508
	preempt_schedule_common();
L
Linus Torvalds 已提交
3509
}
3510
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3511
EXPORT_SYMBOL(preempt_schedule);
3512 3513

/**
3514
 * preempt_schedule_notrace - preempt_schedule called by tracing
3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526
 *
 * The tracing infrastructure uses preempt_enable_notrace to prevent
 * recursion and tracing preempt enabling caused by the tracing
 * infrastructure itself. But as tracing can happen in areas coming
 * from userspace or just about to enter userspace, a preempt enable
 * can occur before user_exit() is called. This will cause the scheduler
 * to be called when the system is still in usermode.
 *
 * To prevent this, the preempt_enable_notrace will use this function
 * instead of preempt_schedule() to exit user context if needed before
 * calling the scheduler.
 */
3527
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3528 3529 3530 3531 3532 3533 3534
{
	enum ctx_state prev_ctx;

	if (likely(!preemptible()))
		return;

	do {
3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547
		/*
		 * Because the function tracer can trace preempt_count_sub()
		 * and it also uses preempt_enable/disable_notrace(), if
		 * NEED_RESCHED is set, the preempt_enable_notrace() called
		 * by the function tracer will call this function again and
		 * cause infinite recursion.
		 *
		 * Preemption must be disabled here before the function
		 * tracer can trace. Break up preempt_disable() into two
		 * calls. One to disable preemption without fear of being
		 * traced. The other to still record the preemption latency,
		 * which can also be traced by the function tracer.
		 */
3548
		preempt_disable_notrace();
3549
		preempt_latency_start(1);
3550 3551 3552 3553 3554 3555
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3556
		__schedule(true);
3557 3558
		exception_exit(prev_ctx);

3559
		preempt_latency_stop(1);
3560
		preempt_enable_no_resched_notrace();
3561 3562
	} while (need_resched());
}
3563
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3564

3565
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3566 3567

/*
3568
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3569 3570 3571 3572
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3573
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3574
{
3575
	enum ctx_state prev_state;
3576

3577
	/* Catch callers which need to be fixed */
3578
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3579

3580 3581
	prev_state = exception_enter();

3582
	do {
3583
		preempt_disable();
3584
		local_irq_enable();
3585
		__schedule(true);
3586
		local_irq_disable();
3587
		sched_preempt_enable_no_resched();
3588
	} while (need_resched());
3589 3590

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3591 3592
}

P
Peter Zijlstra 已提交
3593
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3594
			  void *key)
L
Linus Torvalds 已提交
3595
{
P
Peter Zijlstra 已提交
3596
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3597 3598 3599
}
EXPORT_SYMBOL(default_wake_function);

3600 3601 3602 3603 3604 3605 3606 3607 3608 3609
#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().
 *
3610 3611
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3612
 */
3613
void rt_mutex_setprio(struct task_struct *p, int prio)
3614
{
3615
	int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE;
3616
	const struct sched_class *prev_class;
3617 3618
	struct rq_flags rf;
	struct rq *rq;
3619

3620
	BUG_ON(prio > MAX_PRIO);
3621

3622
	rq = __task_rq_lock(p, &rf);
3623

3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
	/*
	 * 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;
	}

3642
	trace_sched_pi_setprio(p, prio);
3643
	oldprio = p->prio;
3644 3645 3646 3647

	if (oldprio == prio)
		queue_flag &= ~DEQUEUE_MOVE;

3648
	prev_class = p->sched_class;
3649
	queued = task_on_rq_queued(p);
3650
	running = task_current(rq, p);
3651
	if (queued)
3652
		dequeue_task(rq, p, queue_flag);
3653
	if (running)
3654
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3655

3656 3657 3658 3659 3660 3661 3662 3663 3664 3665
	/*
	 * Boosting condition are:
	 * 1. -rt task is running and holds mutex A
	 *      --> -dl task blocks on mutex A
	 *
	 * 2. -dl task is running and holds mutex A
	 *      --> -dl task blocks on mutex A and could preempt the
	 *          running task
	 */
	if (dl_prio(prio)) {
3666 3667 3668
		struct task_struct *pi_task = rt_mutex_get_top_task(p);
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3669
			p->dl.dl_boosted = 1;
3670
			queue_flag |= ENQUEUE_REPLENISH;
3671 3672
		} else
			p->dl.dl_boosted = 0;
3673
		p->sched_class = &dl_sched_class;
3674 3675 3676 3677
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3678
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3679
		p->sched_class = &rt_sched_class;
3680 3681 3682
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3683 3684
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3685
		p->sched_class = &fair_sched_class;
3686
	}
I
Ingo Molnar 已提交
3687

3688 3689
	p->prio = prio;

3690 3691
	if (running)
		p->sched_class->set_curr_task(rq);
3692
	if (queued)
3693
		enqueue_task(rq, p, queue_flag);
3694

P
Peter Zijlstra 已提交
3695
	check_class_changed(rq, p, prev_class, oldprio);
3696
out_unlock:
3697
	preempt_disable(); /* avoid rq from going away on us */
3698
	__task_rq_unlock(rq, &rf);
3699 3700 3701

	balance_callback(rq);
	preempt_enable();
3702 3703
}
#endif
3704

3705
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3706
{
3707
	int old_prio, delta, queued;
3708
	struct rq_flags rf;
3709
	struct rq *rq;
L
Linus Torvalds 已提交
3710

3711
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3712 3713 3714 3715 3716
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3717
	rq = task_rq_lock(p, &rf);
L
Linus Torvalds 已提交
3718 3719 3720 3721
	/*
	 * 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
3722
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3723
	 */
3724
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3725 3726 3727
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3728 3729
	queued = task_on_rq_queued(p);
	if (queued)
3730
		dequeue_task(rq, p, DEQUEUE_SAVE);
L
Linus Torvalds 已提交
3731 3732

	p->static_prio = NICE_TO_PRIO(nice);
3733
	set_load_weight(p);
3734 3735 3736
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3737

3738
	if (queued) {
3739
		enqueue_task(rq, p, ENQUEUE_RESTORE);
L
Linus Torvalds 已提交
3740
		/*
3741 3742
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3743
		 */
3744
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3745
			resched_curr(rq);
L
Linus Torvalds 已提交
3746 3747
	}
out_unlock:
3748
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3749 3750 3751
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3752 3753 3754 3755 3756
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3757
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3758
{
3759
	/* convert nice value [19,-20] to rlimit style value [1,40] */
3760
	int nice_rlim = nice_to_rlimit(nice);
3761

3762
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3763 3764 3765
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3766 3767 3768 3769 3770 3771 3772 3773 3774
#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.
 */
3775
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3776
{
3777
	long nice, retval;
L
Linus Torvalds 已提交
3778 3779 3780 3781 3782 3783

	/*
	 * 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.
	 */
3784
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3785
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3786

3787
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3788 3789 3790
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804
	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.
 *
3805
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3806 3807 3808
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3809
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3810 3811 3812 3813 3814 3815 3816
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
3817 3818
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3819 3820 3821
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835
	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
3836 3837 3838 3839 3840
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
3841 3842
 *
 * Return: The idle task for the cpu @cpu.
L
Linus Torvalds 已提交
3843
 */
3844
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3845 3846 3847 3848 3849 3850 3851
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3852 3853
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3854
 */
A
Alexey Dobriyan 已提交
3855
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3856
{
3857
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3858 3859
}

3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874
/*
 * This function initializes the sched_dl_entity of a newly becoming
 * SCHED_DEADLINE task.
 *
 * Only the static values are considered here, the actual runtime and the
 * absolute deadline will be properly calculated when the task is enqueued
 * for the first time with its new policy.
 */
static void
__setparam_dl(struct task_struct *p, const struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	dl_se->dl_runtime = attr->sched_runtime;
	dl_se->dl_deadline = attr->sched_deadline;
3875
	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
3876
	dl_se->flags = attr->sched_flags;
3877
	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897

	/*
	 * Changing the parameters of a task is 'tricky' and we're not doing
	 * the correct thing -- also see task_dead_dl() and switched_from_dl().
	 *
	 * What we SHOULD do is delay the bandwidth release until the 0-lag
	 * point. This would include retaining the task_struct until that time
	 * and change dl_overflow() to not immediately decrement the current
	 * amount.
	 *
	 * Instead we retain the current runtime/deadline and let the new
	 * parameters take effect after the current reservation period lapses.
	 * This is safe (albeit pessimistic) because the 0-lag point is always
	 * before the current scheduling deadline.
	 *
	 * We can still have temporary overloads because we do not delay the
	 * change in bandwidth until that time; so admission control is
	 * not on the safe side. It does however guarantee tasks will never
	 * consume more than promised.
	 */
3898 3899
}

3900 3901 3902 3903 3904 3905
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

3906 3907
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
3908
{
3909 3910
	int policy = attr->sched_policy;

3911
	if (policy == SETPARAM_POLICY)
3912 3913
		policy = p->policy;

L
Linus Torvalds 已提交
3914
	p->policy = policy;
3915

3916 3917
	if (dl_policy(policy))
		__setparam_dl(p, attr);
3918
	else if (fair_policy(policy))
3919 3920
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

3921 3922 3923 3924 3925 3926
	/*
	 * __sched_setscheduler() ensures attr->sched_priority == 0 when
	 * !rt_policy. Always setting this ensures that things like
	 * getparam()/getattr() don't report silly values for !rt tasks.
	 */
	p->rt_priority = attr->sched_priority;
3927
	p->normal_prio = normal_prio(p);
3928 3929
	set_load_weight(p);
}
3930

3931 3932
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
3933
			   const struct sched_attr *attr, bool keep_boost)
3934 3935
{
	__setscheduler_params(p, attr);
3936

3937
	/*
3938 3939
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
3940
	 */
3941 3942 3943 3944
	if (keep_boost)
		p->prio = rt_mutex_get_effective_prio(p, normal_prio(p));
	else
		p->prio = normal_prio(p);
3945

3946 3947 3948
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
3949 3950 3951
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
3952
}
3953 3954 3955 3956 3957 3958 3959 3960 3961

static void
__getparam_dl(struct task_struct *p, struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	attr->sched_priority = p->rt_priority;
	attr->sched_runtime = dl_se->dl_runtime;
	attr->sched_deadline = dl_se->dl_deadline;
3962
	attr->sched_period = dl_se->dl_period;
3963 3964 3965 3966 3967 3968
	attr->sched_flags = dl_se->flags;
}

/*
 * This function validates the new parameters of a -deadline task.
 * We ask for the deadline not being zero, and greater or equal
3969
 * than the runtime, as well as the period of being zero or
3970
 * greater than deadline. Furthermore, we have to be sure that
3971 3972 3973 3974
 * user parameters are above the internal resolution of 1us (we
 * check sched_runtime only since it is always the smaller one) and
 * below 2^63 ns (we have to check both sched_deadline and
 * sched_period, as the latter can be zero).
3975 3976 3977 3978
 */
static bool
__checkparam_dl(const struct sched_attr *attr)
{
3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004
	/* deadline != 0 */
	if (attr->sched_deadline == 0)
		return false;

	/*
	 * Since we truncate DL_SCALE bits, make sure we're at least
	 * that big.
	 */
	if (attr->sched_runtime < (1ULL << DL_SCALE))
		return false;

	/*
	 * Since we use the MSB for wrap-around and sign issues, make
	 * sure it's not set (mind that period can be equal to zero).
	 */
	if (attr->sched_deadline & (1ULL << 63) ||
	    attr->sched_period & (1ULL << 63))
		return false;

	/* runtime <= deadline <= period (if period != 0) */
	if ((attr->sched_period != 0 &&
	     attr->sched_period < attr->sched_deadline) ||
	    attr->sched_deadline < attr->sched_runtime)
		return false;

	return true;
4005 4006
}

4007 4008 4009 4010 4011 4012 4013 4014 4015 4016
/*
 * 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);
4017 4018
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4019 4020 4021 4022
	rcu_read_unlock();
	return match;
}

4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036
static bool dl_param_changed(struct task_struct *p,
		const struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	if (dl_se->dl_runtime != attr->sched_runtime ||
		dl_se->dl_deadline != attr->sched_deadline ||
		dl_se->dl_period != attr->sched_period ||
		dl_se->flags != attr->sched_flags)
		return true;

	return false;
}

4037 4038
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4039
				bool user, bool pi)
L
Linus Torvalds 已提交
4040
{
4041 4042
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4043
	int retval, oldprio, oldpolicy = -1, queued, running;
4044
	int new_effective_prio, policy = attr->sched_policy;
4045
	const struct sched_class *prev_class;
4046
	struct rq_flags rf;
4047
	int reset_on_fork;
4048
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE;
4049
	struct rq *rq;
L
Linus Torvalds 已提交
4050

4051 4052
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
4053 4054
recheck:
	/* double check policy once rq lock held */
4055 4056
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4057
		policy = oldpolicy = p->policy;
4058
	} else {
4059
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4060

4061
		if (!valid_policy(policy))
4062 4063 4064
			return -EINVAL;
	}

4065 4066 4067
	if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK))
		return -EINVAL;

L
Linus Torvalds 已提交
4068 4069
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4070 4071
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4072
	 */
4073
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4074
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4075
		return -EINVAL;
4076 4077
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4078 4079
		return -EINVAL;

4080 4081 4082
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4083
	if (user && !capable(CAP_SYS_NICE)) {
4084
		if (fair_policy(policy)) {
4085
			if (attr->sched_nice < task_nice(p) &&
4086
			    !can_nice(p, attr->sched_nice))
4087 4088 4089
				return -EPERM;
		}

4090
		if (rt_policy(policy)) {
4091 4092
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4093 4094 4095 4096 4097 4098

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

			/* can't increase priority */
4099 4100
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4101 4102
				return -EPERM;
		}
4103

4104 4105 4106 4107 4108 4109 4110 4111 4112
		 /*
		  * Can't set/change SCHED_DEADLINE policy at all for now
		  * (safest behavior); in the future we would like to allow
		  * unprivileged DL tasks to increase their relative deadline
		  * or reduce their runtime (both ways reducing utilization)
		  */
		if (dl_policy(policy))
			return -EPERM;

I
Ingo Molnar 已提交
4113
		/*
4114 4115
		 * 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
4116
		 */
4117
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4118
			if (!can_nice(p, task_nice(p)))
4119 4120
				return -EPERM;
		}
4121

4122
		/* can't change other user's priorities */
4123
		if (!check_same_owner(p))
4124
			return -EPERM;
4125 4126 4127 4128

		/* Normal users shall not reset the sched_reset_on_fork flag */
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4129
	}
L
Linus Torvalds 已提交
4130

4131
	if (user) {
4132
		retval = security_task_setscheduler(p);
4133 4134 4135 4136
		if (retval)
			return retval;
	}

4137 4138 4139
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
4140
	 *
L
Lucas De Marchi 已提交
4141
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4142 4143
	 * runqueue lock must be held.
	 */
4144
	rq = task_rq_lock(p, &rf);
4145

4146 4147 4148 4149
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
4150
		task_rq_unlock(rq, p, &rf);
4151 4152 4153
		return -EINVAL;
	}

4154
	/*
4155 4156
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4157
	 */
4158
	if (unlikely(policy == p->policy)) {
4159
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4160 4161 4162
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4163
		if (dl_policy(policy) && dl_param_changed(p, attr))
4164
			goto change;
4165

4166
		p->sched_reset_on_fork = reset_on_fork;
4167
		task_rq_unlock(rq, p, &rf);
4168 4169
		return 0;
	}
4170
change:
4171

4172
	if (user) {
4173
#ifdef CONFIG_RT_GROUP_SCHED
4174 4175 4176 4177 4178
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4179 4180
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4181
			task_rq_unlock(rq, p, &rf);
4182 4183 4184
			return -EPERM;
		}
#endif
4185 4186 4187 4188 4189 4190 4191 4192 4193
#ifdef CONFIG_SMP
		if (dl_bandwidth_enabled() && dl_policy(policy)) {
			cpumask_t *span = rq->rd->span;

			/*
			 * Don't allow tasks with an affinity mask smaller than
			 * the entire root_domain to become SCHED_DEADLINE. We
			 * will also fail if there's no bandwidth available.
			 */
4194 4195
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4196
				task_rq_unlock(rq, p, &rf);
4197 4198 4199 4200 4201
				return -EPERM;
			}
		}
#endif
	}
4202

L
Linus Torvalds 已提交
4203 4204 4205
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4206
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4207 4208
		goto recheck;
	}
4209 4210 4211 4212 4213 4214

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4215
	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
4216
		task_rq_unlock(rq, p, &rf);
4217 4218 4219
		return -EBUSY;
	}

4220 4221 4222
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4223 4224 4225 4226 4227 4228 4229 4230 4231
	if (pi) {
		/*
		 * Take priority boosted tasks into account. If the new
		 * effective priority is unchanged, we just store the new
		 * normal parameters and do not touch the scheduler class and
		 * the runqueue. This will be done when the task deboost
		 * itself.
		 */
		new_effective_prio = rt_mutex_get_effective_prio(p, newprio);
4232 4233
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4234 4235
	}

4236
	queued = task_on_rq_queued(p);
4237
	running = task_current(rq, p);
4238
	if (queued)
4239
		dequeue_task(rq, p, queue_flags);
4240
	if (running)
4241
		put_prev_task(rq, p);
4242

4243
	prev_class = p->sched_class;
4244
	__setscheduler(rq, p, attr, pi);
4245

4246 4247
	if (running)
		p->sched_class->set_curr_task(rq);
4248
	if (queued) {
4249 4250 4251 4252
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4253 4254
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4255

4256
		enqueue_task(rq, p, queue_flags);
4257
	}
4258

P
Peter Zijlstra 已提交
4259
	check_class_changed(rq, p, prev_class, oldprio);
4260
	preempt_disable(); /* avoid rq from going away on us */
4261
	task_rq_unlock(rq, p, &rf);
4262

4263 4264
	if (pi)
		rt_mutex_adjust_pi(p);
4265

4266 4267 4268 4269 4270
	/*
	 * Run balance callbacks after we've adjusted the PI chain.
	 */
	balance_callback(rq);
	preempt_enable();
4271

L
Linus Torvalds 已提交
4272 4273
	return 0;
}
4274

4275 4276 4277 4278 4279 4280 4281 4282 4283
static int _sched_setscheduler(struct task_struct *p, int policy,
			       const struct sched_param *param, bool check)
{
	struct sched_attr attr = {
		.sched_policy   = policy,
		.sched_priority = param->sched_priority,
		.sched_nice	= PRIO_TO_NICE(p->static_prio),
	};

4284 4285
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4286 4287 4288 4289 4290
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4291
	return __sched_setscheduler(p, &attr, check, true);
4292
}
4293 4294 4295 4296 4297 4298
/**
 * 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.
 *
4299 4300
 * Return: 0 on success. An error code otherwise.
 *
4301 4302 4303
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4304
		       const struct sched_param *param)
4305
{
4306
	return _sched_setscheduler(p, policy, param, true);
4307
}
L
Linus Torvalds 已提交
4308 4309
EXPORT_SYMBOL_GPL(sched_setscheduler);

4310 4311
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4312
	return __sched_setscheduler(p, attr, true, true);
4313 4314 4315
}
EXPORT_SYMBOL_GPL(sched_setattr);

4316 4317 4318 4319 4320 4321 4322 4323 4324 4325
/**
 * 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.
4326 4327
 *
 * Return: 0 on success. An error code otherwise.
4328 4329
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4330
			       const struct sched_param *param)
4331
{
4332
	return _sched_setscheduler(p, policy, param, false);
4333
}
4334
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4335

I
Ingo Molnar 已提交
4336 4337
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4338 4339 4340
{
	struct sched_param lparam;
	struct task_struct *p;
4341
	int retval;
L
Linus Torvalds 已提交
4342 4343 4344 4345 4346

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4347 4348 4349

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4350
	p = find_process_by_pid(pid);
4351 4352 4353
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4354

L
Linus Torvalds 已提交
4355 4356 4357
	return retval;
}

4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
static int sched_copy_attr(struct sched_attr __user *uattr,
			   struct sched_attr *attr)
{
	u32 size;
	int ret;

	if (!access_ok(VERIFY_WRITE, uattr, SCHED_ATTR_SIZE_VER0))
		return -EFAULT;

	/*
	 * zero the full structure, so that a short copy will be nice.
	 */
	memset(attr, 0, sizeof(*attr));

	ret = get_user(size, &uattr->size);
	if (ret)
		return ret;

	if (size > PAGE_SIZE)	/* silly large */
		goto err_size;

	if (!size)		/* abi compat */
		size = SCHED_ATTR_SIZE_VER0;

	if (size < SCHED_ATTR_SIZE_VER0)
		goto err_size;

	/*
	 * If we're handed a bigger struct than we know of,
	 * ensure all the unknown bits are 0 - i.e. new
	 * user-space does not rely on any kernel feature
	 * extensions we dont know about yet.
	 */
	if (size > sizeof(*attr)) {
		unsigned char __user *addr;
		unsigned char __user *end;
		unsigned char val;

		addr = (void __user *)uattr + sizeof(*attr);
		end  = (void __user *)uattr + size;

		for (; addr < end; addr++) {
			ret = get_user(val, addr);
			if (ret)
				return ret;
			if (val)
				goto err_size;
		}
		size = sizeof(*attr);
	}

	ret = copy_from_user(attr, uattr, size);
	if (ret)
		return -EFAULT;

	/*
	 * XXX: do we want to be lenient like existing syscalls; or do we want
	 * to be strict and return an error on out-of-bounds values?
	 */
4420
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4421

4422
	return 0;
4423 4424 4425

err_size:
	put_user(sizeof(*attr), &uattr->size);
4426
	return -E2BIG;
4427 4428
}

L
Linus Torvalds 已提交
4429 4430 4431 4432 4433
/**
 * 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.
4434 4435
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4436
 */
4437 4438
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
4439
{
4440 4441 4442 4443
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4444 4445 4446 4447 4448 4449 4450
	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.
4451 4452
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4453
 */
4454
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4455
{
4456
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4457 4458
}

4459 4460 4461
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4462
 * @uattr: structure containing the extended parameters.
4463
 * @flags: for future extension.
4464
 */
4465 4466
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4467 4468 4469 4470 4471
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4472
	if (!uattr || pid < 0 || flags)
4473 4474
		return -EINVAL;

4475 4476 4477
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4478

4479
	if ((int)attr.sched_policy < 0)
4480
		return -EINVAL;
4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491

	rcu_read_lock();
	retval = -ESRCH;
	p = find_process_by_pid(pid);
	if (p != NULL)
		retval = sched_setattr(p, &attr);
	rcu_read_unlock();

	return retval;
}

L
Linus Torvalds 已提交
4492 4493 4494
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4495 4496 4497
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4498
 */
4499
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4500
{
4501
	struct task_struct *p;
4502
	int retval;
L
Linus Torvalds 已提交
4503 4504

	if (pid < 0)
4505
		return -EINVAL;
L
Linus Torvalds 已提交
4506 4507

	retval = -ESRCH;
4508
	rcu_read_lock();
L
Linus Torvalds 已提交
4509 4510 4511 4512
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4513 4514
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4515
	}
4516
	rcu_read_unlock();
L
Linus Torvalds 已提交
4517 4518 4519 4520
	return retval;
}

/**
4521
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4522 4523
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4524 4525 4526
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4527
 */
4528
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4529
{
4530
	struct sched_param lp = { .sched_priority = 0 };
4531
	struct task_struct *p;
4532
	int retval;
L
Linus Torvalds 已提交
4533 4534

	if (!param || pid < 0)
4535
		return -EINVAL;
L
Linus Torvalds 已提交
4536

4537
	rcu_read_lock();
L
Linus Torvalds 已提交
4538 4539 4540 4541 4542 4543 4544 4545 4546
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4547 4548
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4549
	rcu_read_unlock();
L
Linus Torvalds 已提交
4550 4551 4552 4553 4554 4555 4556 4557 4558

	/*
	 * 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:
4559
	rcu_read_unlock();
L
Linus Torvalds 已提交
4560 4561 4562
	return retval;
}

4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585
static int sched_read_attr(struct sched_attr __user *uattr,
			   struct sched_attr *attr,
			   unsigned int usize)
{
	int ret;

	if (!access_ok(VERIFY_WRITE, uattr, usize))
		return -EFAULT;

	/*
	 * If we're handed a smaller struct than we know of,
	 * ensure all the unknown bits are 0 - i.e. old
	 * user-space does not get uncomplete information.
	 */
	if (usize < sizeof(*attr)) {
		unsigned char *addr;
		unsigned char *end;

		addr = (void *)attr + usize;
		end  = (void *)attr + sizeof(*attr);

		for (; addr < end; addr++) {
			if (*addr)
4586
				return -EFBIG;
4587 4588 4589 4590 4591
		}

		attr->size = usize;
	}

4592
	ret = copy_to_user(uattr, attr, attr->size);
4593 4594 4595
	if (ret)
		return -EFAULT;

4596
	return 0;
4597 4598 4599
}

/**
4600
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4601
 * @pid: the pid in question.
J
Juri Lelli 已提交
4602
 * @uattr: structure containing the extended parameters.
4603
 * @size: sizeof(attr) for fwd/bwd comp.
4604
 * @flags: for future extension.
4605
 */
4606 4607
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4608 4609 4610 4611 4612 4613 4614 4615
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4616
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629
		return -EINVAL;

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

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

	attr.sched_policy = p->policy;
4630 4631
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4632 4633 4634
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4635 4636
		attr.sched_priority = p->rt_priority;
	else
4637
		attr.sched_nice = task_nice(p);
4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648

	rcu_read_unlock();

	retval = sched_read_attr(uattr, &attr, size);
	return retval;

out_unlock:
	rcu_read_unlock();
	return retval;
}

4649
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4650
{
4651
	cpumask_var_t cpus_allowed, new_mask;
4652 4653
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4654

4655
	rcu_read_lock();
L
Linus Torvalds 已提交
4656 4657 4658

	p = find_process_by_pid(pid);
	if (!p) {
4659
		rcu_read_unlock();
L
Linus Torvalds 已提交
4660 4661 4662
		return -ESRCH;
	}

4663
	/* Prevent p going away */
L
Linus Torvalds 已提交
4664
	get_task_struct(p);
4665
	rcu_read_unlock();
L
Linus Torvalds 已提交
4666

4667 4668 4669 4670
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4671 4672 4673 4674 4675 4676 4677 4678
	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
4679
	retval = -EPERM;
E
Eric W. Biederman 已提交
4680 4681 4682 4683
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4684
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4685 4686 4687
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4688

4689
	retval = security_task_setscheduler(p);
4690
	if (retval)
4691
		goto out_free_new_mask;
4692

4693 4694 4695 4696

	cpuset_cpus_allowed(p, cpus_allowed);
	cpumask_and(new_mask, in_mask, cpus_allowed);

4697 4698 4699 4700 4701 4702 4703
	/*
	 * Since bandwidth control happens on root_domain basis,
	 * if admission test is enabled, we only admit -deadline
	 * tasks allowed to run on all the CPUs in the task's
	 * root_domain.
	 */
#ifdef CONFIG_SMP
4704 4705 4706
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4707
			retval = -EBUSY;
4708
			rcu_read_unlock();
4709
			goto out_free_new_mask;
4710
		}
4711
		rcu_read_unlock();
4712 4713
	}
#endif
P
Peter Zijlstra 已提交
4714
again:
4715
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4716

P
Paul Menage 已提交
4717
	if (!retval) {
4718 4719
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4720 4721 4722 4723 4724
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4725
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4726 4727 4728
			goto again;
		}
	}
4729
out_free_new_mask:
4730 4731 4732 4733
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4734 4735 4736 4737 4738
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4739
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4740
{
4741 4742 4743 4744 4745
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4746 4747 4748 4749 4750 4751 4752 4753
	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
4754 4755
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4756
 */
4757 4758
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4759
{
4760
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4761 4762
	int retval;

4763 4764
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4765

4766 4767 4768 4769 4770
	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
4771 4772
}

4773
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4774
{
4775
	struct task_struct *p;
4776
	unsigned long flags;
L
Linus Torvalds 已提交
4777 4778
	int retval;

4779
	rcu_read_lock();
L
Linus Torvalds 已提交
4780 4781 4782 4783 4784 4785

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

4786 4787 4788 4789
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4790
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4791
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4792
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4793 4794

out_unlock:
4795
	rcu_read_unlock();
L
Linus Torvalds 已提交
4796

4797
	return retval;
L
Linus Torvalds 已提交
4798 4799 4800 4801 4802 4803 4804
}

/**
 * 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
4805
 *
4806 4807
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4808
 */
4809 4810
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4811 4812
{
	int ret;
4813
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4814

A
Anton Blanchard 已提交
4815
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4816 4817
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4818 4819
		return -EINVAL;

4820 4821
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4822

4823 4824
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4825
		size_t retlen = min_t(size_t, len, cpumask_size());
4826 4827

		if (copy_to_user(user_mask_ptr, mask, retlen))
4828 4829
			ret = -EFAULT;
		else
4830
			ret = retlen;
4831 4832
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4833

4834
	return ret;
L
Linus Torvalds 已提交
4835 4836 4837 4838 4839
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4840 4841
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4842 4843
 *
 * Return: 0.
L
Linus Torvalds 已提交
4844
 */
4845
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4846
{
4847
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
4848

4849
	schedstat_inc(rq->yld_count);
4850
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4851 4852 4853 4854 4855 4856

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
4857
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4858
	do_raw_spin_unlock(&rq->lock);
4859
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4860 4861 4862 4863 4864 4865

	schedule();

	return 0;
}

4866
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4867
{
4868
	if (should_resched(0)) {
4869
		preempt_schedule_common();
L
Linus Torvalds 已提交
4870 4871 4872 4873
		return 1;
	}
	return 0;
}
4874
EXPORT_SYMBOL(_cond_resched);
L
Linus Torvalds 已提交
4875 4876

/*
4877
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4878 4879
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4880
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4881 4882 4883
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4884
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4885
{
4886
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4887 4888
	int ret = 0;

4889 4890
	lockdep_assert_held(lock);

4891
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4892
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4893
		if (resched)
4894
			preempt_schedule_common();
N
Nick Piggin 已提交
4895 4896
		else
			cpu_relax();
J
Jan Kara 已提交
4897
		ret = 1;
L
Linus Torvalds 已提交
4898 4899
		spin_lock(lock);
	}
J
Jan Kara 已提交
4900
	return ret;
L
Linus Torvalds 已提交
4901
}
4902
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4903

4904
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4905 4906 4907
{
	BUG_ON(!in_softirq());

4908
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4909
		local_bh_enable();
4910
		preempt_schedule_common();
L
Linus Torvalds 已提交
4911 4912 4913 4914 4915
		local_bh_disable();
		return 1;
	}
	return 0;
}
4916
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4917 4918 4919 4920

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938
 * 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
4939 4940 4941 4942 4943 4944 4945 4946
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4947 4948 4949 4950
/**
 * 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
4951 4952
 * @p: target task
 * @preempt: whether task preemption is allowed or not
4953 4954 4955 4956
 *
 * 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.
 *
4957
 * Return:
4958 4959 4960
 *	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.
4961
 */
4962
int __sched yield_to(struct task_struct *p, bool preempt)
4963 4964 4965 4966
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
4967
	int yielded = 0;
4968 4969 4970 4971 4972 4973

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
4974 4975 4976 4977 4978 4979 4980 4981 4982
	/*
	 * 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;
	}

4983
	double_rq_lock(rq, p_rq);
4984
	if (task_rq(p) != p_rq) {
4985 4986 4987 4988 4989
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
4990
		goto out_unlock;
4991 4992

	if (curr->sched_class != p->sched_class)
4993
		goto out_unlock;
4994 4995

	if (task_running(p_rq, p) || p->state)
4996
		goto out_unlock;
4997 4998

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
4999
	if (yielded) {
5000
		schedstat_inc(rq->yld_count);
5001 5002 5003 5004 5005
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5006
			resched_curr(p_rq);
5007
	}
5008

5009
out_unlock:
5010
	double_rq_unlock(rq, p_rq);
5011
out_irq:
5012 5013
	local_irq_restore(flags);

5014
	if (yielded > 0)
5015 5016 5017 5018 5019 5020
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
5021
/*
I
Ingo Molnar 已提交
5022
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5023 5024 5025 5026
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5027 5028
	int old_iowait = current->in_iowait;
	struct rq *rq;
L
Linus Torvalds 已提交
5029 5030
	long ret;

5031
	current->in_iowait = 1;
5032
	blk_schedule_flush_plug(current);
5033

5034
	delayacct_blkio_start();
5035
	rq = raw_rq();
L
Linus Torvalds 已提交
5036 5037
	atomic_inc(&rq->nr_iowait);
	ret = schedule_timeout(timeout);
5038
	current->in_iowait = old_iowait;
L
Linus Torvalds 已提交
5039
	atomic_dec(&rq->nr_iowait);
5040
	delayacct_blkio_end();
5041

L
Linus Torvalds 已提交
5042 5043
	return ret;
}
5044
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5045 5046 5047 5048 5049

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5050 5051 5052
 * Return: On success, this syscall returns the maximum
 * rt_priority that can be used by a given scheduling class.
 * On failure, a negative error code is returned.
L
Linus Torvalds 已提交
5053
 */
5054
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5055 5056 5057 5058 5059 5060 5061 5062
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5063
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5064
	case SCHED_NORMAL:
5065
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5066
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5067 5068 5069 5070 5071 5072 5073 5074 5075 5076
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5077 5078 5079
 * Return: On success, this syscall returns the minimum
 * rt_priority that can be used by a given scheduling class.
 * On failure, a negative error code is returned.
L
Linus Torvalds 已提交
5080
 */
5081
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5082 5083 5084 5085 5086 5087 5088 5089
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5090
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5091
	case SCHED_NORMAL:
5092
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5093
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105
		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.
5106 5107 5108
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
L
Linus Torvalds 已提交
5109
 */
5110
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5111
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
5112
{
5113
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5114
	unsigned int time_slice;
5115 5116
	struct rq_flags rf;
	struct timespec t;
5117
	struct rq *rq;
5118
	int retval;
L
Linus Torvalds 已提交
5119 5120

	if (pid < 0)
5121
		return -EINVAL;
L
Linus Torvalds 已提交
5122 5123

	retval = -ESRCH;
5124
	rcu_read_lock();
L
Linus Torvalds 已提交
5125 5126 5127 5128 5129 5130 5131 5132
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5133
	rq = task_rq_lock(p, &rf);
5134 5135 5136
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5137
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5138

5139
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
5140
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
5141 5142
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
5143

L
Linus Torvalds 已提交
5144
out_unlock:
5145
	rcu_read_unlock();
L
Linus Torvalds 已提交
5146 5147 5148
	return retval;
}

5149
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5150

5151
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5152 5153
{
	unsigned long free = 0;
5154
	int ppid;
5155
	unsigned long state = p->state;
L
Linus Torvalds 已提交
5156

5157 5158
	if (state)
		state = __ffs(state) + 1;
5159
	printk(KERN_INFO "%-15.15s %c", p->comm,
5160
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5161
#if BITS_PER_LONG == 32
L
Linus Torvalds 已提交
5162
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5163
		printk(KERN_CONT " running  ");
L
Linus Torvalds 已提交
5164
	else
P
Peter Zijlstra 已提交
5165
		printk(KERN_CONT " %08lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5166 5167
#else
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5168
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5169
	else
P
Peter Zijlstra 已提交
5170
		printk(KERN_CONT " %016lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5171 5172
#endif
#ifdef CONFIG_DEBUG_STACK_USAGE
5173
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5174
#endif
5175
	ppid = 0;
5176
	rcu_read_lock();
5177 5178
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5179
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5180
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5181
		task_pid_nr(p), ppid,
5182
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5183

5184
	print_worker_info(KERN_INFO, p);
5185
	show_stack(p, NULL);
L
Linus Torvalds 已提交
5186 5187
}

I
Ingo Molnar 已提交
5188
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5189
{
5190
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5191

5192
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5193 5194
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5195
#else
P
Peter Zijlstra 已提交
5196 5197
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5198
#endif
5199
	rcu_read_lock();
5200
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5201 5202
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5203
		 * console might take a lot of time:
5204 5205 5206
		 * Also, reset softlockup watchdogs on all CPUs, because
		 * another CPU might be blocked waiting for us to process
		 * an IPI.
L
Linus Torvalds 已提交
5207 5208
		 */
		touch_nmi_watchdog();
5209
		touch_all_softlockup_watchdogs();
I
Ingo Molnar 已提交
5210
		if (!state_filter || (p->state & state_filter))
5211
			sched_show_task(p);
5212
	}
L
Linus Torvalds 已提交
5213

I
Ingo Molnar 已提交
5214
#ifdef CONFIG_SCHED_DEBUG
5215 5216
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5217
#endif
5218
	rcu_read_unlock();
I
Ingo Molnar 已提交
5219 5220 5221
	/*
	 * Only show locks if all tasks are dumped:
	 */
5222
	if (!state_filter)
I
Ingo Molnar 已提交
5223
		debug_show_all_locks();
L
Linus Torvalds 已提交
5224 5225
}

5226
void init_idle_bootup_task(struct task_struct *idle)
I
Ingo Molnar 已提交
5227
{
I
Ingo Molnar 已提交
5228
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
5229 5230
}

5231 5232 5233 5234 5235 5236 5237 5238
/**
 * 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.
 */
5239
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5240
{
5241
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5242 5243
	unsigned long flags;

5244 5245
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5246

5247
	__sched_fork(0, idle);
5248
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5249 5250
	idle->se.exec_start = sched_clock();

5251 5252
	kasan_unpoison_task_stack(idle);

5253 5254 5255 5256 5257 5258 5259 5260 5261
#ifdef CONFIG_SMP
	/*
	 * Its possible that init_idle() gets called multiple times on a task,
	 * in that case do_set_cpus_allowed() will not do the right thing.
	 *
	 * And since this is boot we can forgo the serialization.
	 */
	set_cpus_allowed_common(idle, cpumask_of(cpu));
#endif
5262 5263 5264 5265 5266 5267 5268 5269 5270 5271 5272
	/*
	 * 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
5273
	__set_task_cpu(idle, cpu);
5274
	rcu_read_unlock();
L
Linus Torvalds 已提交
5275 5276

	rq->curr = rq->idle = idle;
5277
	idle->on_rq = TASK_ON_RQ_QUEUED;
5278
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5279
	idle->on_cpu = 1;
5280
#endif
5281 5282
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5283 5284

	/* Set the preempt count _outside_ the spinlocks! */
5285
	init_idle_preempt_count(idle, cpu);
5286

I
Ingo Molnar 已提交
5287 5288 5289 5290
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5291
	ftrace_graph_init_idle_task(idle, cpu);
5292
	vtime_init_idle(idle, cpu);
5293
#ifdef CONFIG_SMP
5294 5295
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5296 5297
}

5298 5299 5300 5301 5302 5303 5304
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
	int ret = 1, trial_cpus;
	struct dl_bw *cur_dl_b;
	unsigned long flags;

5305 5306 5307
	if (!cpumask_weight(cur))
		return ret;

5308
	rcu_read_lock_sched();
5309 5310 5311 5312 5313 5314 5315 5316
	cur_dl_b = dl_bw_of(cpumask_any(cur));
	trial_cpus = cpumask_weight(trial);

	raw_spin_lock_irqsave(&cur_dl_b->lock, flags);
	if (cur_dl_b->bw != -1 &&
	    cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw)
		ret = 0;
	raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags);
5317
	rcu_read_unlock_sched();
5318 5319 5320 5321

	return ret;
}

5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345
int task_can_attach(struct task_struct *p,
		    const struct cpumask *cs_cpus_allowed)
{
	int ret = 0;

	/*
	 * Kthreads which disallow setaffinity shouldn't be moved
	 * to a new cpuset; we don't want to change their cpu
	 * affinity and isolating such threads by their set of
	 * allowed nodes is unnecessary.  Thus, cpusets are not
	 * applicable for such threads.  This prevents checking for
	 * success of set_cpus_allowed_ptr() on all attached tasks
	 * before cpus_allowed may be changed.
	 */
	if (p->flags & PF_NO_SETAFFINITY) {
		ret = -EINVAL;
		goto out;
	}

#ifdef CONFIG_SMP
	if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span,
					      cs_cpus_allowed)) {
		unsigned int dest_cpu = cpumask_any_and(cpu_active_mask,
							cs_cpus_allowed);
5346
		struct dl_bw *dl_b;
5347 5348 5349 5350
		bool overflow;
		int cpus;
		unsigned long flags;

5351 5352
		rcu_read_lock_sched();
		dl_b = dl_bw_of(dest_cpu);
5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367
		raw_spin_lock_irqsave(&dl_b->lock, flags);
		cpus = dl_bw_cpus(dest_cpu);
		overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw);
		if (overflow)
			ret = -EBUSY;
		else {
			/*
			 * We reserve space for this task in the destination
			 * root_domain, as we can't fail after this point.
			 * We will free resources in the source root_domain
			 * later on (see set_cpus_allowed_dl()).
			 */
			__dl_add(dl_b, p->dl.dl_bw);
		}
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
5368
		rcu_read_unlock_sched();
5369 5370 5371 5372 5373 5374 5375

	}
#endif
out:
	return ret;
}

L
Linus Torvalds 已提交
5376 5377
#ifdef CONFIG_SMP

5378 5379
static bool sched_smp_initialized __read_mostly;

5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394
#ifdef CONFIG_NUMA_BALANCING
/* Migrate current task p to target_cpu */
int migrate_task_to(struct task_struct *p, int target_cpu)
{
	struct migration_arg arg = { p, target_cpu };
	int curr_cpu = task_cpu(p);

	if (curr_cpu == target_cpu)
		return 0;

	if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p)))
		return -EINVAL;

	/* TODO: This is not properly updating schedstats */

5395
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5396 5397
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5398 5399 5400 5401 5402 5403 5404

/*
 * Requeue a task on a given node and accurately track the number of NUMA
 * tasks on the runqueues
 */
void sched_setnuma(struct task_struct *p, int nid)
{
5405
	bool queued, running;
5406 5407
	struct rq_flags rf;
	struct rq *rq;
5408

5409
	rq = task_rq_lock(p, &rf);
5410
	queued = task_on_rq_queued(p);
5411 5412
	running = task_current(rq, p);

5413
	if (queued)
5414
		dequeue_task(rq, p, DEQUEUE_SAVE);
5415
	if (running)
5416
		put_prev_task(rq, p);
5417 5418 5419 5420 5421

	p->numa_preferred_nid = nid;

	if (running)
		p->sched_class->set_curr_task(rq);
5422
	if (queued)
5423
		enqueue_task(rq, p, ENQUEUE_RESTORE);
5424
	task_rq_unlock(rq, p, &rf);
5425
}
P
Peter Zijlstra 已提交
5426
#endif /* CONFIG_NUMA_BALANCING */
5427

L
Linus Torvalds 已提交
5428
#ifdef CONFIG_HOTPLUG_CPU
5429
/*
5430 5431
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
5432
 */
5433
void idle_task_exit(void)
L
Linus Torvalds 已提交
5434
{
5435
	struct mm_struct *mm = current->active_mm;
5436

5437
	BUG_ON(cpu_online(smp_processor_id()));
5438

5439
	if (mm != &init_mm) {
5440
		switch_mm_irqs_off(mm, &init_mm, current);
5441 5442
		finish_arch_post_lock_switch();
	}
5443
	mmdrop(mm);
L
Linus Torvalds 已提交
5444 5445 5446
}

/*
5447 5448
 * 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
5449 5450 5451
 * nr_active count is stable. We need to take the teardown thread which
 * is calling this into account, so we hand in adjust = 1 to the load
 * calculation.
5452 5453
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5454
 */
5455
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5456
{
5457
	long delta = calc_load_fold_active(rq, 1);
5458 5459
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5460 5461
}

5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474 5475 5476 5477
static void put_prev_task_fake(struct rq *rq, struct task_struct *prev)
{
}

static const struct sched_class fake_sched_class = {
	.put_prev_task = put_prev_task_fake,
};

static struct task_struct fake_task = {
	/*
	 * Avoid pull_{rt,dl}_task()
	 */
	.prio = MAX_PRIO + 1,
	.sched_class = &fake_sched_class,
};

5478
/*
5479 5480 5481 5482 5483 5484
 * 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
5485
 */
5486
static void migrate_tasks(struct rq *dead_rq)
L
Linus Torvalds 已提交
5487
{
5488
	struct rq *rq = dead_rq;
5489
	struct task_struct *next, *stop = rq->stop;
5490
	struct pin_cookie cookie;
5491
	int dest_cpu;
L
Linus Torvalds 已提交
5492 5493

	/*
5494 5495 5496 5497 5498 5499 5500
	 * 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
5501
	 */
5502
	rq->stop = NULL;
5503

5504 5505 5506 5507 5508 5509 5510
	/*
	 * 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);

5511
	for (;;) {
5512 5513 5514 5515 5516
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5517
			break;
5518

5519
		/*
W
Wanpeng Li 已提交
5520
		 * pick_next_task assumes pinned rq->lock.
5521
		 */
5522 5523
		cookie = lockdep_pin_lock(&rq->lock);
		next = pick_next_task(rq, &fake_task, cookie);
5524
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
5525
		next->sched_class->put_prev_task(rq, next);
5526

W
Wanpeng Li 已提交
5527 5528 5529 5530 5531 5532 5533 5534 5535
		/*
		 * Rules for changing task_struct::cpus_allowed are holding
		 * both pi_lock and rq->lock, such that holding either
		 * stabilizes the mask.
		 *
		 * Drop rq->lock is not quite as disastrous as it usually is
		 * because !cpu_active at this point, which means load-balance
		 * will not interfere. Also, stop-machine.
		 */
5536
		lockdep_unpin_lock(&rq->lock, cookie);
W
Wanpeng Li 已提交
5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&next->pi_lock);
		raw_spin_lock(&rq->lock);

		/*
		 * Since we're inside stop-machine, _nothing_ should have
		 * changed the task, WARN if weird stuff happened, because in
		 * that case the above rq->lock drop is a fail too.
		 */
		if (WARN_ON(task_rq(next) != rq || !task_on_rq_queued(next))) {
			raw_spin_unlock(&next->pi_lock);
			continue;
		}

5551
		/* Find suitable destination for @next, with force if needed. */
5552
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5553

5554 5555 5556 5557 5558 5559
		rq = __migrate_task(rq, next, dest_cpu);
		if (rq != dead_rq) {
			raw_spin_unlock(&rq->lock);
			rq = dead_rq;
			raw_spin_lock(&rq->lock);
		}
W
Wanpeng Li 已提交
5560
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5561
	}
5562

5563
	rq->stop = stop;
5564
}
L
Linus Torvalds 已提交
5565 5566
#endif /* CONFIG_HOTPLUG_CPU */

5567 5568 5569 5570 5571
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

5572
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591
		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);
		}

5592
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5593 5594 5595 5596
		rq->online = 0;
	}
}

5597
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5598
{
5599
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5600

5601 5602 5603
	rq->age_stamp = sched_clock_cpu(cpu);
}

5604 5605
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

5606
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
5607

5608
static __read_mostly int sched_debug_enabled;
5609

5610
static int __init sched_debug_setup(char *str)
5611
{
5612
	sched_debug_enabled = 1;
5613 5614 5615

	return 0;
}
5616 5617 5618 5619 5620 5621
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
5622

5623
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
5624
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
5625
{
I
Ingo Molnar 已提交
5626
	struct sched_group *group = sd->groups;
L
Linus Torvalds 已提交
5627

5628
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
5629 5630 5631 5632

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

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
5633
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
5634
		if (sd->parent)
P
Peter Zijlstra 已提交
5635 5636
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
5637
		return -1;
N
Nick Piggin 已提交
5638 5639
	}

5640 5641
	printk(KERN_CONT "span %*pbl level %s\n",
	       cpumask_pr_args(sched_domain_span(sd)), sd->name);
I
Ingo Molnar 已提交
5642

5643
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
5644 5645
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
5646
	}
5647
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5648 5649
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
5650
	}
L
Linus Torvalds 已提交
5651

I
Ingo Molnar 已提交
5652
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
5653
	do {
I
Ingo Molnar 已提交
5654
		if (!group) {
P
Peter Zijlstra 已提交
5655 5656
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
5657 5658 5659
			break;
		}

5660
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5661 5662
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
5663 5664
			break;
		}
L
Linus Torvalds 已提交
5665

5666 5667
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5668 5669
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
5670 5671
			break;
		}
L
Linus Torvalds 已提交
5672

5673
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
5674

5675 5676
		printk(KERN_CONT " %*pbl",
		       cpumask_pr_args(sched_group_cpus(group)));
5677
		if (group->sgc->capacity != SCHED_CAPACITY_SCALE) {
5678 5679
			printk(KERN_CONT " (cpu_capacity = %d)",
				group->sgc->capacity);
5680
		}
L
Linus Torvalds 已提交
5681

I
Ingo Molnar 已提交
5682 5683
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
5684
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
5685

5686
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
5687
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
5688

5689 5690
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
5691 5692
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
5693 5694
	return 0;
}
L
Linus Torvalds 已提交
5695

I
Ingo Molnar 已提交
5696 5697 5698
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
5699

5700
	if (!sched_debug_enabled)
5701 5702
		return;

I
Ingo Molnar 已提交
5703 5704 5705 5706
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
5707

I
Ingo Molnar 已提交
5708 5709 5710
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
5711
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
5712
			break;
L
Linus Torvalds 已提交
5713 5714
		level++;
		sd = sd->parent;
5715
		if (!sd)
I
Ingo Molnar 已提交
5716 5717
			break;
	}
L
Linus Torvalds 已提交
5718
}
5719
#else /* !CONFIG_SCHED_DEBUG */
5720
# define sched_domain_debug(sd, cpu) do { } while (0)
5721 5722 5723 5724
static inline bool sched_debug(void)
{
	return false;
}
5725
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
5726

5727
static int sd_degenerate(struct sched_domain *sd)
5728
{
5729
	if (cpumask_weight(sched_domain_span(sd)) == 1)
5730 5731 5732 5733 5734 5735
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
5736
			 SD_BALANCE_EXEC |
5737
			 SD_SHARE_CPUCAPACITY |
5738
			 SD_ASYM_CPUCAPACITY |
5739 5740
			 SD_SHARE_PKG_RESOURCES |
			 SD_SHARE_POWERDOMAIN)) {
5741 5742 5743 5744 5745
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
5746
	if (sd->flags & (SD_WAKE_AFFINE))
5747 5748 5749 5750 5751
		return 0;

	return 1;
}

5752 5753
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
5754 5755 5756 5757 5758 5759
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

5760
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
5761 5762 5763 5764 5765 5766 5767
		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 |
5768
				SD_BALANCE_EXEC |
5769
				SD_ASYM_CPUCAPACITY |
5770
				SD_SHARE_CPUCAPACITY |
5771
				SD_SHARE_PKG_RESOURCES |
5772 5773
				SD_PREFER_SIBLING |
				SD_SHARE_POWERDOMAIN);
5774 5775
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
5776 5777 5778 5779 5780 5781 5782
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

5783
static void free_rootdomain(struct rcu_head *rcu)
5784
{
5785
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
5786

5787
	cpupri_cleanup(&rd->cpupri);
5788
	cpudl_cleanup(&rd->cpudl);
5789
	free_cpumask_var(rd->dlo_mask);
5790 5791 5792 5793 5794 5795
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
5796 5797
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
5798
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
5799 5800
	unsigned long flags;

5801
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5802 5803

	if (rq->rd) {
I
Ingo Molnar 已提交
5804
		old_rd = rq->rd;
G
Gregory Haskins 已提交
5805

5806
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
5807
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5808

5809
		cpumask_clear_cpu(rq->cpu, old_rd->span);
5810

I
Ingo Molnar 已提交
5811
		/*
5812
		 * If we dont want to free the old_rd yet then
I
Ingo Molnar 已提交
5813 5814 5815 5816 5817
		 * 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
5818 5819 5820 5821 5822
	}

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

5823
	cpumask_set_cpu(rq->cpu, rd->span);
5824
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
5825
		set_rq_online(rq);
G
Gregory Haskins 已提交
5826

5827
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
5828 5829

	if (old_rd)
5830
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
5831 5832
}

5833
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
5834 5835 5836
{
	memset(rd, 0, sizeof(*rd));

5837
	if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
5838
		goto out;
5839
	if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
5840
		goto free_span;
5841
	if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
5842
		goto free_online;
5843
	if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
5844
		goto free_dlo_mask;
5845

5846
	init_dl_bw(&rd->dl_bw);
5847 5848
	if (cpudl_init(&rd->cpudl) != 0)
		goto free_dlo_mask;
5849

5850
	if (cpupri_init(&rd->cpupri) != 0)
5851
		goto free_rto_mask;
5852
	return 0;
5853

5854 5855
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
5856 5857
free_dlo_mask:
	free_cpumask_var(rd->dlo_mask);
5858 5859 5860 5861
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
5862
out:
5863
	return -ENOMEM;
G
Gregory Haskins 已提交
5864 5865
}

5866 5867 5868 5869 5870 5871
/*
 * 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
5872 5873
static void init_defrootdomain(void)
{
5874
	init_rootdomain(&def_root_domain);
5875

G
Gregory Haskins 已提交
5876 5877 5878
	atomic_set(&def_root_domain.refcount, 1);
}

5879
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5880 5881 5882 5883 5884 5885 5886
{
	struct root_domain *rd;

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

5887
	if (init_rootdomain(rd) != 0) {
5888 5889 5890
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5891 5892 5893 5894

	return rd;
}

5895
static void free_sched_groups(struct sched_group *sg, int free_sgc)
5896 5897 5898 5899 5900 5901 5902 5903 5904 5905
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

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

5906 5907
		if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
			kfree(sg->sgc);
5908 5909 5910 5911 5912 5913

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

5914 5915 5916
static void free_sched_domain(struct rcu_head *rcu)
{
	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
5917 5918 5919 5920 5921 5922 5923 5924

	/*
	 * 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)) {
5925
		kfree(sd->groups->sgc);
5926
		kfree(sd->groups);
5927
	}
5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941
	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);
}

5942 5943 5944 5945 5946 5947 5948
/*
 * 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
5949
 * two cpus are in the same cache domain, see cpus_share_cache().
5950 5951
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
5952
DEFINE_PER_CPU(int, sd_llc_size);
5953
DEFINE_PER_CPU(int, sd_llc_id);
5954
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
5955 5956
DEFINE_PER_CPU(struct sched_domain *, sd_busy);
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
5957 5958 5959 5960

static void update_top_cache_domain(int cpu)
{
	struct sched_domain *sd;
5961
	struct sched_domain *busy_sd = NULL;
5962
	int id = cpu;
5963
	int size = 1;
5964 5965

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
5966
	if (sd) {
5967
		id = cpumask_first(sched_domain_span(sd));
5968
		size = cpumask_weight(sched_domain_span(sd));
5969
		busy_sd = sd->parent; /* sd_busy */
5970
	}
5971
	rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd);
5972 5973

	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
5974
	per_cpu(sd_llc_size, cpu) = size;
5975
	per_cpu(sd_llc_id, cpu) = id;
5976 5977 5978

	sd = lowest_flag_domain(cpu, SD_NUMA);
	rcu_assign_pointer(per_cpu(sd_numa, cpu), sd);
5979 5980 5981

	sd = highest_flag_domain(cpu, SD_ASYM_PACKING);
	rcu_assign_pointer(per_cpu(sd_asym, cpu), sd);
5982 5983
}

L
Linus Torvalds 已提交
5984
/*
I
Ingo Molnar 已提交
5985
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
5986 5987
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
5988 5989
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
5990
{
5991
	struct rq *rq = cpu_rq(cpu);
5992 5993 5994
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
5995
	for (tmp = sd; tmp; ) {
5996 5997 5998
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
5999

6000
		if (sd_parent_degenerate(tmp, parent)) {
6001
			tmp->parent = parent->parent;
6002 6003
			if (parent->parent)
				parent->parent->child = tmp;
6004 6005 6006 6007 6008 6009 6010
			/*
			 * Transfer SD_PREFER_SIBLING down in case of a
			 * degenerate parent; the spans match for this
			 * so the property transfers.
			 */
			if (parent->flags & SD_PREFER_SIBLING)
				tmp->flags |= SD_PREFER_SIBLING;
6011
			destroy_sched_domain(parent, cpu);
6012 6013
		} else
			tmp = tmp->parent;
6014 6015
	}

6016
	if (sd && sd_degenerate(sd)) {
6017
		tmp = sd;
6018
		sd = sd->parent;
6019
		destroy_sched_domain(tmp, cpu);
6020 6021 6022
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
6023

6024
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
6025

G
Gregory Haskins 已提交
6026
	rq_attach_root(rq, rd);
6027
	tmp = rq->sd;
N
Nick Piggin 已提交
6028
	rcu_assign_pointer(rq->sd, sd);
6029
	destroy_sched_domains(tmp, cpu);
6030 6031

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
6032 6033 6034 6035 6036
}

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
6037 6038
	int ret;

R
Rusty Russell 已提交
6039
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
6040 6041 6042 6043 6044
	ret = cpulist_parse(str, cpu_isolated_map);
	if (ret) {
		pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids);
		return 0;
	}
L
Linus Torvalds 已提交
6045 6046
	return 1;
}
I
Ingo Molnar 已提交
6047
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
6048

6049
struct s_data {
6050
	struct sched_domain ** __percpu sd;
6051 6052 6053
	struct root_domain	*rd;
};

6054 6055
enum s_alloc {
	sa_rootdomain,
6056
	sa_sd,
6057
	sa_sd_storage,
6058 6059 6060
	sa_none,
};

P
Peter Zijlstra 已提交
6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095 6096 6097 6098
/*
 * 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));
}

6099 6100 6101 6102 6103 6104 6105
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;
6106
	struct sched_domain *sibling;
6107 6108 6109 6110 6111 6112 6113 6114 6115 6116
	int i;

	cpumask_clear(covered);

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

		if (cpumask_test_cpu(i, covered))
			continue;

6117
		sibling = *per_cpu_ptr(sdd->sd, i);
P
Peter Zijlstra 已提交
6118 6119

		/* See the comment near build_group_mask(). */
6120
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
P
Peter Zijlstra 已提交
6121 6122
			continue;

6123
		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
6124
				GFP_KERNEL, cpu_to_node(cpu));
6125 6126 6127 6128 6129

		if (!sg)
			goto fail;

		sg_span = sched_group_cpus(sg);
6130 6131 6132
		if (sibling->child)
			cpumask_copy(sg_span, sched_domain_span(sibling->child));
		else
6133 6134 6135 6136
			cpumask_set_cpu(i, sg_span);

		cpumask_or(covered, covered, sg_span);

6137 6138
		sg->sgc = *per_cpu_ptr(sdd->sgc, i);
		if (atomic_inc_return(&sg->sgc->ref) == 1)
P
Peter Zijlstra 已提交
6139 6140
			build_group_mask(sd, sg);

6141
		/*
6142
		 * Initialize sgc->capacity such that even if we mess up the
6143 6144 6145
		 * domains and no possible iteration will get us here, we won't
		 * die on a /0 trap.
		 */
6146
		sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span);
6147

P
Peter Zijlstra 已提交
6148 6149 6150 6151 6152
		/*
		 * 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
6153
		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
P
Peter Zijlstra 已提交
6154
		    group_balance_cpu(sg) == cpu)
6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172 6173
			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;
}

6174
static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
L
Linus Torvalds 已提交
6175
{
6176 6177
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	struct sched_domain *child = sd->child;
L
Linus Torvalds 已提交
6178

6179 6180
	if (child)
		cpu = cpumask_first(sched_domain_span(child));
6181

6182
	if (sg) {
6183
		*sg = *per_cpu_ptr(sdd->sg, cpu);
6184 6185
		(*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu);
		atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */
6186
	}
6187 6188

	return cpu;
6189 6190
}

6191
/*
6192 6193
 * 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,
6194
 * and ->cpu_capacity to 0.
6195 6196
 *
 * Assumes the sched_domain tree is fully constructed
6197
 */
6198 6199
static int
build_sched_groups(struct sched_domain *sd, int cpu)
L
Linus Torvalds 已提交
6200
{
6201 6202 6203
	struct sched_group *first = NULL, *last = NULL;
	struct sd_data *sdd = sd->private;
	const struct cpumask *span = sched_domain_span(sd);
6204
	struct cpumask *covered;
6205
	int i;
6206

6207 6208 6209
	get_group(cpu, sdd, &sd->groups);
	atomic_inc(&sd->groups->ref);

6210
	if (cpu != cpumask_first(span))
6211 6212
		return 0;

6213 6214 6215
	lockdep_assert_held(&sched_domains_mutex);
	covered = sched_domains_tmpmask;

6216
	cpumask_clear(covered);
6217

6218 6219
	for_each_cpu(i, span) {
		struct sched_group *sg;
6220
		int group, j;
6221

6222 6223
		if (cpumask_test_cpu(i, covered))
			continue;
6224

6225
		group = get_group(i, sdd, &sg);
P
Peter Zijlstra 已提交
6226
		cpumask_setall(sched_group_mask(sg));
6227

6228 6229 6230
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
				continue;
6231

6232 6233 6234
			cpumask_set_cpu(j, covered);
			cpumask_set_cpu(j, sched_group_cpus(sg));
		}
6235

6236 6237 6238 6239 6240 6241 6242
		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
	}
	last->next = first;
6243 6244

	return 0;
6245
}
6246

6247
/*
6248
 * Initialize sched groups cpu_capacity.
6249
 *
6250
 * cpu_capacity indicates the capacity of sched group, which is used while
6251
 * distributing the load between different sched groups in a sched domain.
6252 6253 6254 6255
 * Typically cpu_capacity 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_capacity will pickup more load compared to the
 * group having less cpu_capacity.
6256
 */
6257
static void init_sched_groups_capacity(int cpu, struct sched_domain *sd)
6258
{
6259
	struct sched_group *sg = sd->groups;
6260

6261
	WARN_ON(!sg);
6262 6263 6264 6265 6266

	do {
		sg->group_weight = cpumask_weight(sched_group_cpus(sg));
		sg = sg->next;
	} while (sg != sd->groups);
6267

P
Peter Zijlstra 已提交
6268
	if (cpu != group_balance_cpu(sg))
6269
		return;
6270

6271 6272
	update_group_capacity(sd, cpu);
	atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight);
6273 6274
}

6275 6276 6277 6278 6279
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

6280
static int default_relax_domain_level = -1;
6281
int sched_domain_level_max;
6282 6283 6284

static int __init setup_relax_domain_level(char *str)
{
6285 6286
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
6287

6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305
	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 */
6306
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6307 6308
	} else {
		/* turn on idle balance on this domain */
6309
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6310 6311 6312
	}
}

6313 6314 6315
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

6316 6317 6318 6319 6320
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
6321 6322
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
6323 6324
	case sa_sd:
		free_percpu(d->sd); /* fall through */
6325
	case sa_sd_storage:
6326
		__sdt_free(cpu_map); /* fall through */
6327 6328 6329 6330
	case sa_none:
		break;
	}
}
6331

6332 6333 6334
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
6335 6336
	memset(d, 0, sizeof(*d));

6337 6338
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
6339 6340 6341
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
6342
	d->rd = alloc_rootdomain();
6343
	if (!d->rd)
6344
		return sa_sd;
6345 6346
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
6347

6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359
/*
 * 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;

6360
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
6361
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
6362

6363 6364
	if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
		*per_cpu_ptr(sdd->sgc, cpu) = NULL;
6365 6366
}

6367 6368
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
6369
enum numa_topology_type sched_numa_topology_type;
6370
static int *sched_domains_numa_distance;
6371
int sched_max_numa_distance;
6372 6373
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
6374
#endif
6375

6376 6377 6378
/*
 * SD_flags allowed in topology descriptions.
 *
6379 6380 6381
 * These flags are purely descriptive of the topology and do not prescribe
 * behaviour. Behaviour is artificial and mapped in the below sd_init()
 * function:
6382
 *
6383 6384 6385 6386
 *   SD_SHARE_CPUCAPACITY   - describes SMT topologies
 *   SD_SHARE_PKG_RESOURCES - describes shared caches
 *   SD_NUMA                - describes NUMA topologies
 *   SD_SHARE_POWERDOMAIN   - describes shared power domain
6387
 *   SD_ASYM_CPUCAPACITY    - describes mixed capacity topologies
6388 6389 6390 6391 6392
 *
 * Odd one out, which beside describing the topology has a quirk also
 * prescribes the desired behaviour that goes along with it:
 *
 *   SD_ASYM_PACKING        - describes SMT quirks
6393 6394
 */
#define TOPOLOGY_SD_FLAGS		\
6395
	(SD_SHARE_CPUCAPACITY |		\
6396 6397
	 SD_SHARE_PKG_RESOURCES |	\
	 SD_NUMA |			\
6398
	 SD_ASYM_PACKING |		\
6399
	 SD_ASYM_CPUCAPACITY |		\
6400
	 SD_SHARE_POWERDOMAIN)
6401 6402

static struct sched_domain *
6403 6404
sd_init(struct sched_domain_topology_level *tl,
	struct sched_domain *child, int cpu)
6405 6406
{
	struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);
6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422
	int sd_weight, sd_flags = 0;

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

	sd_weight = cpumask_weight(tl->mask(cpu));

	if (tl->sd_flags)
		sd_flags = (*tl->sd_flags)();
	if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS,
			"wrong sd_flags in topology description\n"))
		sd_flags &= ~TOPOLOGY_SD_FLAGS;
6423 6424 6425 6426 6427

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
6428
		.imbalance_pct		= 125,
6429 6430 6431 6432

		.cache_nice_tries	= 0,
		.busy_idx		= 0,
		.idle_idx		= 0,
6433 6434 6435 6436 6437 6438
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
6439 6440
					| 1*SD_BALANCE_EXEC
					| 1*SD_BALANCE_FORK
6441
					| 0*SD_BALANCE_WAKE
6442
					| 1*SD_WAKE_AFFINE
6443
					| 0*SD_SHARE_CPUCAPACITY
6444
					| 0*SD_SHARE_PKG_RESOURCES
6445
					| 0*SD_SERIALIZE
6446
					| 0*SD_PREFER_SIBLING
6447 6448
					| 0*SD_NUMA
					| sd_flags
6449
					,
6450

6451 6452
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
6453
		.smt_gain		= 0,
6454 6455
		.max_newidle_lb_cost	= 0,
		.next_decay_max_lb_cost	= jiffies,
6456
		.child			= child,
6457 6458 6459
#ifdef CONFIG_SCHED_DEBUG
		.name			= tl->name,
#endif
6460 6461 6462
	};

	/*
6463
	 * Convert topological properties into behaviour.
6464
	 */
6465

6466 6467 6468 6469 6470 6471 6472
	if (sd->flags & SD_ASYM_CPUCAPACITY) {
		struct sched_domain *t = sd;

		for_each_lower_domain(t)
			t->flags |= SD_BALANCE_WAKE;
	}

6473
	if (sd->flags & SD_SHARE_CPUCAPACITY) {
6474
		sd->flags |= SD_PREFER_SIBLING;
6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504
		sd->imbalance_pct = 110;
		sd->smt_gain = 1178; /* ~15% */

	} else if (sd->flags & SD_SHARE_PKG_RESOURCES) {
		sd->imbalance_pct = 117;
		sd->cache_nice_tries = 1;
		sd->busy_idx = 2;

#ifdef CONFIG_NUMA
	} else if (sd->flags & SD_NUMA) {
		sd->cache_nice_tries = 2;
		sd->busy_idx = 3;
		sd->idle_idx = 2;

		sd->flags |= SD_SERIALIZE;
		if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) {
			sd->flags &= ~(SD_BALANCE_EXEC |
				       SD_BALANCE_FORK |
				       SD_WAKE_AFFINE);
		}

#endif
	} else {
		sd->flags |= SD_PREFER_SIBLING;
		sd->cache_nice_tries = 1;
		sd->busy_idx = 2;
		sd->idle_idx = 1;
	}

	sd->private = &tl->data;
6505 6506 6507 6508

	return sd;
}

6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522
/*
 * Topology list, bottom-up.
 */
static struct sched_domain_topology_level default_topology[] = {
#ifdef CONFIG_SCHED_SMT
	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
#endif
#ifdef CONFIG_SCHED_MC
	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
#endif
	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
	{ NULL, },
};

6523 6524
static struct sched_domain_topology_level *sched_domain_topology =
	default_topology;
6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535

#define for_each_sd_topology(tl)			\
	for (tl = sched_domain_topology; tl->mask; tl++)

void set_sched_topology(struct sched_domain_topology_level *tl)
{
	sched_domain_topology = tl;
}

#ifdef CONFIG_NUMA

6536 6537 6538 6539 6540
static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561
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");
}

6562
bool find_numa_distance(int distance)
6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576
{
	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;
}

6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601
/*
 * A system can have three types of NUMA topology:
 * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system
 * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes
 * NUMA_BACKPLANE: nodes can reach other nodes through a backplane
 *
 * The difference between a glueless mesh topology and a backplane
 * topology lies in whether communication between not directly
 * connected nodes goes through intermediary nodes (where programs
 * could run), or through backplane controllers. This affects
 * placement of programs.
 *
 * The type of topology can be discerned with the following tests:
 * - If the maximum distance between any nodes is 1 hop, the system
 *   is directly connected.
 * - If for two nodes A and B, located N > 1 hops away from each other,
 *   there is an intermediary node C, which is < N hops away from both
 *   nodes A and B, the system is a glueless mesh.
 */
static void init_numa_topology_type(void)
{
	int a, b, c, n;

	n = sched_max_numa_distance;

6602
	if (sched_domains_numa_levels <= 1) {
6603
		sched_numa_topology_type = NUMA_DIRECT;
6604 6605
		return;
	}
6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628

	for_each_online_node(a) {
		for_each_online_node(b) {
			/* Find two nodes furthest removed from each other. */
			if (node_distance(a, b) < n)
				continue;

			/* Is there an intermediary node between a and b? */
			for_each_online_node(c) {
				if (node_distance(a, c) < n &&
				    node_distance(b, c) < n) {
					sched_numa_topology_type =
							NUMA_GLUELESS_MESH;
					return;
				}
			}

			sched_numa_topology_type = NUMA_BACKPLANE;
			return;
		}
	}
}

6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649
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++) {
6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673
			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;
6674
		}
6675 6676 6677 6678 6679 6680

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
6681
	}
6682 6683 6684 6685

	if (!level)
		return;

6686 6687 6688 6689
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
6690
	 * The sched_domains_numa_distance[] array includes the actual distance
6691 6692 6693
	 * numbers.
	 */

6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704
	/*
	 * 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;

6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719
	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++) {
6720
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
6721 6722 6723 6724 6725
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

6726
			for_each_node(k) {
6727
				if (node_distance(j, k) > sched_domains_numa_distance[i])
6728 6729 6730 6731 6732 6733 6734
					continue;

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

6735 6736 6737
	/* Compute default topology size */
	for (i = 0; sched_domain_topology[i].mask; i++);

6738
	tl = kzalloc((i + level + 1) *
6739 6740 6741 6742 6743 6744 6745
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
6746 6747
	for (i = 0; sched_domain_topology[i].mask; i++)
		tl[i] = sched_domain_topology[i];
6748 6749 6750 6751 6752 6753 6754

	/*
	 * .. and append 'j' levels of NUMA goodness.
	 */
	for (j = 0; j < level; i++, j++) {
		tl[i] = (struct sched_domain_topology_level){
			.mask = sd_numa_mask,
6755
			.sd_flags = cpu_numa_flags,
6756 6757
			.flags = SDTL_OVERLAP,
			.numa_level = j,
6758
			SD_INIT_NAME(NUMA)
6759 6760 6761 6762
		};
	}

	sched_domain_topology = tl;
6763 6764

	sched_domains_numa_levels = level;
6765
	sched_max_numa_distance = sched_domains_numa_distance[level - 1];
6766 6767

	init_numa_topology_type();
6768
}
6769

6770
static void sched_domains_numa_masks_set(unsigned int cpu)
6771 6772
{
	int node = cpu_to_node(cpu);
6773
	int i, j;
6774 6775 6776 6777 6778 6779 6780 6781 6782

	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]);
		}
	}
}

6783
static void sched_domains_numa_masks_clear(unsigned int cpu)
6784 6785
{
	int i, j;
6786

6787 6788 6789 6790 6791 6792
	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]);
	}
}

6793
#else
6794 6795 6796
static inline void sched_init_numa(void) { }
static void sched_domains_numa_masks_set(unsigned int cpu) { }
static void sched_domains_numa_masks_clear(unsigned int cpu) { }
6797 6798
#endif /* CONFIG_NUMA */

6799 6800 6801 6802 6803
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6804
	for_each_sd_topology(tl) {
6805 6806 6807 6808 6809 6810 6811 6812 6813 6814
		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;

6815 6816
		sdd->sgc = alloc_percpu(struct sched_group_capacity *);
		if (!sdd->sgc)
6817 6818
			return -ENOMEM;

6819 6820 6821
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
			struct sched_group *sg;
6822
			struct sched_group_capacity *sgc;
6823

P
Peter Zijlstra 已提交
6824
			sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835
					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;

6836 6837
			sg->next = sg;

6838
			*per_cpu_ptr(sdd->sg, j) = sg;
6839

6840
			sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
6841
					GFP_KERNEL, cpu_to_node(j));
6842
			if (!sgc)
6843 6844
				return -ENOMEM;

6845
			*per_cpu_ptr(sdd->sgc, j) = sgc;
6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856
		}
	}

	return 0;
}

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

6857
	for_each_sd_topology(tl) {
6858 6859 6860
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871
			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));
6872 6873
			if (sdd->sgc)
				kfree(*per_cpu_ptr(sdd->sgc, j));
6874 6875
		}
		free_percpu(sdd->sd);
6876
		sdd->sd = NULL;
6877
		free_percpu(sdd->sg);
6878
		sdd->sg = NULL;
6879 6880
		free_percpu(sdd->sgc);
		sdd->sgc = NULL;
6881 6882 6883
	}
}

6884
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6885 6886
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
6887
{
6888
	struct sched_domain *sd = sd_init(tl, child, cpu);
6889 6890

	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
6891 6892 6893
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
6894
		child->parent = sd;
P
Peter Zijlstra 已提交
6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908

		if (!cpumask_subset(sched_domain_span(child),
				    sched_domain_span(sd))) {
			pr_err("BUG: arch topology borken\n");
#ifdef CONFIG_SCHED_DEBUG
			pr_err("     the %s domain not a subset of the %s domain\n",
					child->name, sd->name);
#endif
			/* Fixup, ensure @sd has at least @child cpus. */
			cpumask_or(sched_domain_span(sd),
				   sched_domain_span(sd),
				   sched_domain_span(child));
		}

6909
	}
6910
	set_domain_attribute(sd, attr);
6911 6912 6913 6914

	return sd;
}

6915 6916 6917 6918
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
6919 6920
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
6921
{
6922
	enum s_alloc alloc_state;
6923
	struct sched_domain *sd;
6924
	struct s_data d;
6925
	struct rq *rq = NULL;
6926
	int i, ret = -ENOMEM;
6927

6928 6929 6930
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
6931

6932
	/* Set up domains for cpus specified by the cpu_map. */
6933
	for_each_cpu(i, cpu_map) {
6934 6935
		struct sched_domain_topology_level *tl;

6936
		sd = NULL;
6937
		for_each_sd_topology(tl) {
6938
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
6939 6940
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
6941 6942
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
6943 6944
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
6945
		}
6946 6947 6948 6949 6950 6951
	}

	/* 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));
6952 6953 6954 6955 6956 6957 6958
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
6959
		}
6960
	}
6961

6962
	/* Calculate CPU capacity for physical packages and nodes */
6963 6964 6965
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
6966

6967 6968
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
6969
			init_sched_groups_capacity(i, sd);
6970
		}
6971
	}
6972

L
Linus Torvalds 已提交
6973
	/* Attach the domains */
6974
	rcu_read_lock();
6975
	for_each_cpu(i, cpu_map) {
6976
		rq = cpu_rq(i);
6977
		sd = *per_cpu_ptr(d.sd, i);
6978 6979 6980 6981 6982

		/* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */
		if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity))
			WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig);

6983
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
6984
	}
6985
	rcu_read_unlock();
6986

6987 6988 6989 6990 6991
	if (rq) {
		pr_info("span: %*pbl (max cpu_capacity = %lu)\n",
			cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity);
	}

6992
	ret = 0;
6993
error:
6994
	__free_domain_allocs(&d, alloc_state, cpu_map);
6995
	return ret;
L
Linus Torvalds 已提交
6996
}
P
Paul Jackson 已提交
6997

6998
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
6999
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
7000 7001
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
7002 7003 7004

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
7005 7006
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
7007
 */
7008
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
7009

7010 7011 7012 7013 7014
/*
 * 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.
 */
7015
int __weak arch_update_cpu_topology(void)
7016
{
7017
	return 0;
7018 7019
}

7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044
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);
}

7045
/*
I
Ingo Molnar 已提交
7046
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
7047 7048
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
7049
 */
7050
static int init_sched_domains(const struct cpumask *cpu_map)
7051
{
7052 7053
	int err;

7054
	arch_update_cpu_topology();
P
Paul Jackson 已提交
7055
	ndoms_cur = 1;
7056
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
7057
	if (!doms_cur)
7058 7059
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
7060
	err = build_sched_domains(doms_cur[0], NULL);
7061
	register_sched_domain_sysctl();
7062 7063

	return err;
7064 7065 7066 7067 7068 7069
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
7070
static void detach_destroy_domains(const struct cpumask *cpu_map)
7071 7072 7073
{
	int i;

7074
	rcu_read_lock();
7075
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
7076
		cpu_attach_domain(NULL, &def_root_domain, i);
7077
	rcu_read_unlock();
7078 7079
}

7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095
/* 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
7096 7097
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
7098
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
7099 7100 7101
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
7102
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
7103 7104 7105
 * 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
7106 7107 7108
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
7109 7110 7111 7112 7113 7114
 * 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
7115
 *
7116
 * If doms_new == NULL it will be replaced with cpu_online_mask.
7117 7118
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
7119
 *
P
Paul Jackson 已提交
7120 7121
 * Call with hotplug lock held
 */
7122
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
7123
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
7124
{
7125
	int i, j, n;
7126
	int new_topology;
P
Paul Jackson 已提交
7127

7128
	mutex_lock(&sched_domains_mutex);
7129

7130 7131 7132
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

7133 7134 7135
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

7136
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
7137 7138 7139

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
7140
		for (j = 0; j < n && !new_topology; j++) {
7141
			if (cpumask_equal(doms_cur[i], doms_new[j])
7142
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
7143 7144 7145
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
7146
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
7147 7148 7149 7150
match1:
		;
	}

7151
	n = ndoms_cur;
7152
	if (doms_new == NULL) {
7153
		n = 0;
7154
		doms_new = &fallback_doms;
7155
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
7156
		WARN_ON_ONCE(dattr_new);
7157 7158
	}

P
Paul Jackson 已提交
7159 7160
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
7161
		for (j = 0; j < n && !new_topology; j++) {
7162
			if (cpumask_equal(doms_new[i], doms_cur[j])
7163
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
7164 7165 7166
				goto match2;
		}
		/* no match - add a new doms_new */
7167
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
7168 7169 7170 7171 7172
match2:
		;
	}

	/* Remember the new sched domains */
7173 7174
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
7175
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
7176
	doms_cur = doms_new;
7177
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
7178
	ndoms_cur = ndoms_new;
7179 7180

	register_sched_domain_sysctl();
7181

7182
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
7183 7184
}

7185 7186
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
7187
/*
7188 7189 7190
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
7191 7192 7193
 *
 * 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
7194
 */
7195
static void cpuset_cpu_active(void)
7196
{
7197
	if (cpuhp_tasks_frozen) {
7198 7199 7200 7201 7202 7203 7204 7205 7206
		/*
		 * 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);
7207
			return;
7208 7209 7210 7211 7212 7213
		}
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
7214
	}
7215
	cpuset_update_active_cpus(true);
7216
}
7217

7218
static int cpuset_cpu_inactive(unsigned int cpu)
7219
{
7220 7221
	unsigned long flags;
	struct dl_bw *dl_b;
7222 7223
	bool overflow;
	int cpus;
7224

7225
	if (!cpuhp_tasks_frozen) {
7226 7227
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
7228

7229 7230 7231 7232
		raw_spin_lock_irqsave(&dl_b->lock, flags);
		cpus = dl_bw_cpus(cpu);
		overflow = __dl_overflow(dl_b, cpus, 0, 0);
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
7233

7234
		rcu_read_unlock_sched();
7235

7236
		if (overflow)
7237
			return -EBUSY;
7238
		cpuset_update_active_cpus(false);
7239
	} else {
7240 7241
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
7242
	}
7243
	return 0;
7244 7245
}

7246
int sched_cpu_activate(unsigned int cpu)
7247
{
7248 7249 7250
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

7251
	set_cpu_active(cpu, true);
7252

7253
	if (sched_smp_initialized) {
7254
		sched_domains_numa_masks_set(cpu);
7255
		cpuset_cpu_active();
7256
	}
7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272 7273 7274 7275

	/*
	 * Put the rq online, if not already. This happens:
	 *
	 * 1) In the early boot process, because we build the real domains
	 *    after all cpus have been brought up.
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
	raw_spin_lock_irqsave(&rq->lock, flags);
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
	raw_spin_unlock_irqrestore(&rq->lock, flags);

	update_max_interval();

7276
	return 0;
7277 7278
}

7279
int sched_cpu_deactivate(unsigned int cpu)
7280 7281 7282
{
	int ret;

7283
	set_cpu_active(cpu, false);
7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297
	/*
	 * We've cleared cpu_active_mask, wait for all preempt-disabled and RCU
	 * users of this state to go away such that all new such users will
	 * observe it.
	 *
	 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
	 * not imply sync_sched(), so wait for both.
	 *
	 * Do sync before park smpboot threads to take care the rcu boost case.
	 */
	if (IS_ENABLED(CONFIG_PREEMPT))
		synchronize_rcu_mult(call_rcu, call_rcu_sched);
	else
		synchronize_rcu();
7298 7299 7300 7301 7302 7303 7304 7305

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
7306
	}
7307 7308
	sched_domains_numa_masks_clear(cpu);
	return 0;
7309 7310
}

7311 7312 7313 7314 7315 7316 7317 7318
static void sched_rq_cpu_starting(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);

	rq->calc_load_update = calc_load_update;
	update_max_interval();
}

7319 7320 7321
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
7322
	sched_rq_cpu_starting(cpu);
7323
	return 0;
7324 7325
}

7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
	raw_spin_lock_irqsave(&rq->lock, flags);
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
	migrate_tasks(rq);
	BUG_ON(rq->nr_running != 1);
	raw_spin_unlock_irqrestore(&rq->lock, flags);
	calc_load_migrate(rq);
	update_max_interval();
7344
	nohz_balance_exit_idle(cpu);
7345
	hrtick_clear(rq);
7346 7347 7348 7349
	return 0;
}
#endif

L
Linus Torvalds 已提交
7350 7351
void __init sched_init_smp(void)
{
7352 7353 7354
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7355
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
7356

7357 7358
	sched_init_numa();

7359 7360 7361 7362 7363
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
	 * cpu masks are stable and all blatant races in the below code cannot
	 * happen.
	 */
7364
	mutex_lock(&sched_domains_mutex);
7365
	init_sched_domains(cpu_active_mask);
7366 7367 7368
	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);
7369
	mutex_unlock(&sched_domains_mutex);
7370

7371
	/* Move init over to a non-isolated CPU */
7372
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
7373
		BUG();
I
Ingo Molnar 已提交
7374
	sched_init_granularity();
7375
	free_cpumask_var(non_isolated_cpus);
7376

7377
	init_sched_rt_class();
7378
	init_sched_dl_class();
7379
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
7380
}
7381 7382 7383

static int __init migration_init(void)
{
7384
	sched_rq_cpu_starting(smp_processor_id());
7385
	return 0;
L
Linus Torvalds 已提交
7386
}
7387 7388
early_initcall(migration_init);

L
Linus Torvalds 已提交
7389 7390 7391
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
7392
	sched_init_granularity();
L
Linus Torvalds 已提交
7393 7394 7395 7396 7397 7398 7399 7400 7401 7402
}
#endif /* CONFIG_SMP */

int in_sched_functions(unsigned long addr)
{
	return in_lock_functions(addr) ||
		(addr >= (unsigned long)__sched_text_start
		&& addr < (unsigned long)__sched_text_end);
}

7403
#ifdef CONFIG_CGROUP_SCHED
7404 7405 7406 7407
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
7408
struct task_group root_task_group;
7409
LIST_HEAD(task_groups);
7410 7411 7412

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
7413
#endif
P
Peter Zijlstra 已提交
7414

7415
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
P
Peter Zijlstra 已提交
7416

L
Linus Torvalds 已提交
7417 7418
void __init sched_init(void)
{
I
Ingo Molnar 已提交
7419
	int i, j;
7420 7421 7422 7423 7424 7425 7426 7427 7428
	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 **);
#endif
	if (alloc_size) {
7429
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
7430 7431

#ifdef CONFIG_FAIR_GROUP_SCHED
7432
		root_task_group.se = (struct sched_entity **)ptr;
7433 7434
		ptr += nr_cpu_ids * sizeof(void **);

7435
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
7436
		ptr += nr_cpu_ids * sizeof(void **);
7437

7438
#endif /* CONFIG_FAIR_GROUP_SCHED */
7439
#ifdef CONFIG_RT_GROUP_SCHED
7440
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
7441 7442
		ptr += nr_cpu_ids * sizeof(void **);

7443
		root_task_group.rt_rq = (struct rt_rq **)ptr;
7444 7445
		ptr += nr_cpu_ids * sizeof(void **);

7446
#endif /* CONFIG_RT_GROUP_SCHED */
7447
	}
7448
#ifdef CONFIG_CPUMASK_OFFSTACK
7449 7450 7451
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7452
	}
7453
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
7454

7455 7456 7457
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth,
7458
			global_rt_period(), global_rt_runtime());
7459

G
Gregory Haskins 已提交
7460 7461 7462 7463
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

7464
#ifdef CONFIG_RT_GROUP_SCHED
7465
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
7466
			global_rt_period(), global_rt_runtime());
7467
#endif /* CONFIG_RT_GROUP_SCHED */
7468

D
Dhaval Giani 已提交
7469
#ifdef CONFIG_CGROUP_SCHED
7470 7471
	task_group_cache = KMEM_CACHE(task_group, 0);

7472 7473
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
7474
	INIT_LIST_HEAD(&root_task_group.siblings);
7475
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
7476
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
7477

7478
	for_each_possible_cpu(i) {
7479
		struct rq *rq;
L
Linus Torvalds 已提交
7480 7481

		rq = cpu_rq(i);
7482
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
7483
		rq->nr_running = 0;
7484 7485
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
7486
		init_cfs_rq(&rq->cfs);
7487 7488
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
7489
#ifdef CONFIG_FAIR_GROUP_SCHED
7490
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
7491
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
D
Dhaval Giani 已提交
7492
		/*
7493
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
7494 7495 7496 7497
		 *
		 * 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
7498
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
7499 7500 7501
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
7502
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
7503 7504 7505
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
7506
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
7507
		 *
7508 7509
		 * 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
7510
		 */
7511
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
7512
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
7513 7514 7515
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
7516
#ifdef CONFIG_RT_GROUP_SCHED
7517
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
7518
#endif
L
Linus Torvalds 已提交
7519

I
Ingo Molnar 已提交
7520 7521
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
7522

L
Linus Torvalds 已提交
7523
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
7524
		rq->sd = NULL;
G
Gregory Haskins 已提交
7525
		rq->rd = NULL;
7526
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
7527
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
7528
		rq->active_balance = 0;
I
Ingo Molnar 已提交
7529
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
7530
		rq->push_cpu = 0;
7531
		rq->cpu = i;
7532
		rq->online = 0;
7533 7534
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
7535
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
7536 7537 7538

		INIT_LIST_HEAD(&rq->cfs_tasks);

7539
		rq_attach_root(rq, &def_root_domain);
7540
#ifdef CONFIG_NO_HZ_COMMON
7541
		rq->last_load_update_tick = jiffies;
7542
		rq->nohz_flags = 0;
7543
#endif
7544 7545 7546
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
7547
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
7548
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
7549 7550 7551
		atomic_set(&rq->nr_iowait, 0);
	}

7552
	set_load_weight(&init_task);
7553

L
Linus Torvalds 已提交
7554 7555 7556 7557 7558 7559 7560 7561 7562 7563 7564 7565 7566
	/*
	 * 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());
7567 7568 7569

	calc_load_update = jiffies + LOAD_FREQ;

7570
#ifdef CONFIG_SMP
7571
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
7572 7573 7574
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
7575
	idle_thread_set_boot_cpu();
7576
	set_cpu_rq_start_time(smp_processor_id());
7577 7578
#endif
	init_sched_fair_class();
7579

7580 7581
	init_schedstats();

7582
	scheduler_running = 1;
L
Linus Torvalds 已提交
7583 7584
}

7585
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
7586 7587
static inline int preempt_count_equals(int preempt_offset)
{
7588
	int nested = preempt_count() + rcu_preempt_depth();
7589

A
Arnd Bergmann 已提交
7590
	return (nested == preempt_offset);
7591 7592
}

7593
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7594
{
P
Peter Zijlstra 已提交
7595 7596 7597 7598 7599
	/*
	 * Blocking primitives will set (and therefore destroy) current->state,
	 * since we will exit with TASK_RUNNING make sure we enter with it,
	 * otherwise we will destroy state.
	 */
7600
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
7601 7602 7603 7604
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
7605
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
7606

7607 7608 7609 7610 7611
	___might_sleep(file, line, preempt_offset);
}
EXPORT_SYMBOL(__might_sleep);

void ___might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7612 7613
{
	static unsigned long prev_jiffy;	/* ratelimiting */
7614
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
7615

7616
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
7617 7618
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
7619
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
7620 7621 7622 7623 7624
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

7625 7626 7627
	/* Save this before calling printk(), since that will clobber it */
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
7628 7629 7630 7631 7632 7633 7634
	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
7635

7636 7637 7638
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
7639 7640 7641
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
7642 7643
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
7644
		pr_err("Preemption disabled at:");
7645
		print_ip_sym(preempt_disable_ip);
7646 7647
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
7648
	dump_stack();
7649
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
7650
}
7651
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
7652 7653 7654
#endif

#ifdef CONFIG_MAGIC_SYSRQ
7655
void normalize_rt_tasks(void)
7656
{
7657
	struct task_struct *g, *p;
7658 7659 7660
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
7661

7662
	read_lock(&tasklist_lock);
7663
	for_each_process_thread(g, p) {
7664 7665 7666
		/*
		 * Only normalize user tasks:
		 */
7667
		if (p->flags & PF_KTHREAD)
7668 7669
			continue;

7670 7671 7672 7673
		p->se.exec_start = 0;
		schedstat_set(p->se.statistics.wait_start,  0);
		schedstat_set(p->se.statistics.sleep_start, 0);
		schedstat_set(p->se.statistics.block_start, 0);
I
Ingo Molnar 已提交
7674

7675
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
7676 7677 7678 7679
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
7680
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
7681
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
7682
			continue;
I
Ingo Molnar 已提交
7683
		}
L
Linus Torvalds 已提交
7684

7685
		__sched_setscheduler(p, &attr, false, false);
7686
	}
7687
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
7688 7689 7690
}

#endif /* CONFIG_MAGIC_SYSRQ */
7691

7692
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
7693
/*
7694
 * These functions are only useful for the IA64 MCA handling, or kdb.
7695 7696 7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707
 *
 * 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!
7708 7709
 *
 * Return: The current task for @cpu.
7710
 */
7711
struct task_struct *curr_task(int cpu)
7712 7713 7714 7715
{
	return cpu_curr(cpu);
}

7716 7717 7718
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
7719 7720 7721 7722 7723 7724
/**
 * 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
7725 7726
 * 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
7727 7728 7729 7730 7731 7732 7733
 * 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!
 */
7734
void set_curr_task(int cpu, struct task_struct *p)
7735 7736 7737 7738 7739
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
7740

D
Dhaval Giani 已提交
7741
#ifdef CONFIG_CGROUP_SCHED
7742 7743 7744
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

7745
static void sched_free_group(struct task_group *tg)
7746 7747 7748
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
7749
	autogroup_free(tg);
7750
	kmem_cache_free(task_group_cache, tg);
7751 7752 7753
}

/* allocate runqueue etc for a new task group */
7754
struct task_group *sched_create_group(struct task_group *parent)
7755 7756 7757
{
	struct task_group *tg;

7758
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
7759 7760 7761
	if (!tg)
		return ERR_PTR(-ENOMEM);

7762
	if (!alloc_fair_sched_group(tg, parent))
7763 7764
		goto err;

7765
	if (!alloc_rt_sched_group(tg, parent))
7766 7767
		goto err;

7768 7769 7770
	return tg;

err:
7771
	sched_free_group(tg);
7772 7773 7774 7775 7776 7777 7778
	return ERR_PTR(-ENOMEM);
}

void sched_online_group(struct task_group *tg, struct task_group *parent)
{
	unsigned long flags;

7779
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7780
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
7781 7782 7783 7784 7785

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
7786
	list_add_rcu(&tg->siblings, &parent->children);
7787
	spin_unlock_irqrestore(&task_group_lock, flags);
7788 7789

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
7790 7791
}

7792
/* rcu callback to free various structures associated with a task group */
7793
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
7794 7795
{
	/* now it should be safe to free those cfs_rqs */
7796
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
7797 7798
}

7799
void sched_destroy_group(struct task_group *tg)
7800 7801
{
	/* wait for possible concurrent references to cfs_rqs complete */
7802
	call_rcu(&tg->rcu, sched_free_group_rcu);
7803 7804 7805
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
7806
{
7807
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
7808

7809
	/* end participation in shares distribution */
7810
	unregister_fair_sched_group(tg);
7811 7812

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7813
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
7814
	list_del_rcu(&tg->siblings);
7815
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7816 7817
}

7818
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
7819
{
P
Peter Zijlstra 已提交
7820
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
7821

7822 7823 7824 7825 7826 7827
	/*
	 * All callers are synchronized by task_rq_lock(); we do not use RCU
	 * which is pointless here. Thus, we pass "true" to task_css_check()
	 * to prevent lockdep warnings.
	 */
	tg = container_of(task_css_check(tsk, cpu_cgrp_id, true),
P
Peter Zijlstra 已提交
7828 7829 7830 7831
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
7832
#ifdef CONFIG_FAIR_GROUP_SCHED
7833 7834
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
7835
	else
P
Peter Zijlstra 已提交
7836
#endif
7837
		set_task_rq(tsk, task_cpu(tsk));
7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850 7851 7852 7853 7854 7855 7856 7857 7858 7859 7860 7861 7862 7863
}

/*
 * Change task's runqueue when it moves between groups.
 *
 * 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.
 */
void sched_move_task(struct task_struct *tsk)
{
	int queued, running;
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);

	running = task_current(rq, tsk);
	queued = task_on_rq_queued(tsk);

	if (queued)
		dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE);
	if (unlikely(running))
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
7864

7865 7866
	if (unlikely(running))
		tsk->sched_class->set_curr_task(rq);
7867
	if (queued)
7868
		enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE);
S
Srivatsa Vaddagiri 已提交
7869

7870
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
7871
}
D
Dhaval Giani 已提交
7872
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
7873

7874 7875 7876 7877 7878
#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
7879

P
Peter Zijlstra 已提交
7880 7881
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
7882
{
P
Peter Zijlstra 已提交
7883
	struct task_struct *g, *p;
7884

7885 7886 7887 7888 7889 7890
	/*
	 * Autogroups do not have RT tasks; see autogroup_create().
	 */
	if (task_group_is_autogroup(tg))
		return 0;

7891
	for_each_process_thread(g, p) {
7892
		if (rt_task(p) && task_group(p) == tg)
P
Peter Zijlstra 已提交
7893
			return 1;
7894
	}
7895

P
Peter Zijlstra 已提交
7896 7897
	return 0;
}
7898

P
Peter Zijlstra 已提交
7899 7900 7901 7902 7903
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
7904

7905
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
7906 7907 7908 7909 7910
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
7911

P
Peter Zijlstra 已提交
7912 7913
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
7914

P
Peter Zijlstra 已提交
7915 7916 7917
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
7918 7919
	}

7920 7921 7922 7923 7924
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
7925

7926 7927 7928
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
7929 7930
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
7931

P
Peter Zijlstra 已提交
7932
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7933

7934 7935 7936 7937 7938
	/*
	 * 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
7939

7940 7941 7942
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
7943 7944 7945
	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
7946

P
Peter Zijlstra 已提交
7947 7948 7949 7950
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
7951

P
Peter Zijlstra 已提交
7952
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7953
	}
P
Peter Zijlstra 已提交
7954

P
Peter Zijlstra 已提交
7955 7956 7957 7958
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
7959 7960
}

P
Peter Zijlstra 已提交
7961
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
7962
{
7963 7964
	int ret;

P
Peter Zijlstra 已提交
7965 7966 7967 7968 7969 7970
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

7971 7972 7973 7974 7975
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
7976 7977
}

7978
static int tg_set_rt_bandwidth(struct task_group *tg,
7979
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
7980
{
P
Peter Zijlstra 已提交
7981
	int i, err = 0;
P
Peter Zijlstra 已提交
7982

7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993
	/*
	 * Disallowing the root group RT runtime is BAD, it would disallow the
	 * kernel creating (and or operating) RT threads.
	 */
	if (tg == &root_task_group && rt_runtime == 0)
		return -EINVAL;

	/* No period doesn't make any sense. */
	if (rt_period == 0)
		return -EINVAL;

P
Peter Zijlstra 已提交
7994
	mutex_lock(&rt_constraints_mutex);
7995
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
7996 7997
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
7998
		goto unlock;
P
Peter Zijlstra 已提交
7999

8000
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8001 8002
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
8003 8004 8005 8006

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

8007
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8008
		rt_rq->rt_runtime = rt_runtime;
8009
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8010
	}
8011
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
8012
unlock:
8013
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
8014 8015 8016
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
8017 8018
}

8019
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8020 8021 8022 8023 8024 8025 8026 8027
{
	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;

8028
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8029 8030
}

8031
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
8032 8033 8034
{
	u64 rt_runtime_us;

8035
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
8036 8037
		return -1;

8038
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
8039 8040 8041
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
8042

8043
static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
8044 8045 8046
{
	u64 rt_runtime, rt_period;

8047
	rt_period = rt_period_us * NSEC_PER_USEC;
8048 8049
	rt_runtime = tg->rt_bandwidth.rt_runtime;

8050
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8051 8052
}

8053
static long sched_group_rt_period(struct task_group *tg)
8054 8055 8056 8057 8058 8059 8060
{
	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;
}
8061
#endif /* CONFIG_RT_GROUP_SCHED */
8062

8063
#ifdef CONFIG_RT_GROUP_SCHED
8064 8065 8066 8067 8068
static int sched_rt_global_constraints(void)
{
	int ret = 0;

	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
8069
	read_lock(&tasklist_lock);
8070
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
8071
	read_unlock(&tasklist_lock);
8072 8073 8074 8075
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
8076

8077
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8078 8079 8080 8081 8082 8083 8084 8085
{
	/* 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;
}

8086
#else /* !CONFIG_RT_GROUP_SCHED */
8087 8088
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
8089
	unsigned long flags;
8090
	int i;
8091

8092
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8093 8094 8095
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

8096
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8097
		rt_rq->rt_runtime = global_rt_runtime();
8098
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8099
	}
8100
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8101

8102
	return 0;
8103
}
8104
#endif /* CONFIG_RT_GROUP_SCHED */
8105

8106
static int sched_dl_global_validate(void)
8107
{
8108 8109
	u64 runtime = global_rt_runtime();
	u64 period = global_rt_period();
8110
	u64 new_bw = to_ratio(period, runtime);
8111
	struct dl_bw *dl_b;
8112
	int cpu, ret = 0;
8113
	unsigned long flags;
8114 8115 8116 8117 8118 8119 8120 8121 8122 8123

	/*
	 * Here we want to check the bandwidth not being set to some
	 * value smaller than the currently allocated bandwidth in
	 * any of the root_domains.
	 *
	 * FIXME: Cycling on all the CPUs is overdoing, but simpler than
	 * cycling on root_domains... Discussion on different/better
	 * solutions is welcome!
	 */
8124
	for_each_possible_cpu(cpu) {
8125 8126
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8127

8128
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8129 8130
		if (new_bw < dl_b->total_bw)
			ret = -EBUSY;
8131
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8132

8133 8134
		rcu_read_unlock_sched();

8135 8136
		if (ret)
			break;
8137 8138
	}

8139
	return ret;
8140 8141
}

8142
static void sched_dl_do_global(void)
8143
{
8144
	u64 new_bw = -1;
8145
	struct dl_bw *dl_b;
8146
	int cpu;
8147
	unsigned long flags;
8148

8149 8150 8151 8152 8153 8154 8155 8156 8157 8158
	def_dl_bandwidth.dl_period = global_rt_period();
	def_dl_bandwidth.dl_runtime = global_rt_runtime();

	if (global_rt_runtime() != RUNTIME_INF)
		new_bw = to_ratio(global_rt_period(), global_rt_runtime());

	/*
	 * FIXME: As above...
	 */
	for_each_possible_cpu(cpu) {
8159 8160
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8161

8162
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8163
		dl_b->bw = new_bw;
8164
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8165 8166

		rcu_read_unlock_sched();
8167
	}
8168 8169 8170 8171 8172 8173 8174
}

static int sched_rt_global_validate(void)
{
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

8175 8176
	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
8177 8178 8179 8180 8181 8182 8183 8184 8185
		return -EINVAL;

	return 0;
}

static void sched_rt_do_global(void)
{
	def_rt_bandwidth.rt_runtime = global_rt_runtime();
	def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
8186 8187
}

8188
int sched_rt_handler(struct ctl_table *table, int write,
8189
		void __user *buffer, size_t *lenp,
8190 8191 8192 8193
		loff_t *ppos)
{
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);
8194
	int ret;
8195 8196 8197 8198 8199

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

8200
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8201 8202

	if (!ret && write) {
8203 8204 8205 8206
		ret = sched_rt_global_validate();
		if (ret)
			goto undo;

8207
		ret = sched_dl_global_validate();
8208 8209 8210
		if (ret)
			goto undo;

8211
		ret = sched_rt_global_constraints();
8212 8213 8214 8215 8216 8217 8218 8219 8220 8221
		if (ret)
			goto undo;

		sched_rt_do_global();
		sched_dl_do_global();
	}
	if (0) {
undo:
		sysctl_sched_rt_period = old_period;
		sysctl_sched_rt_runtime = old_runtime;
8222 8223 8224 8225 8226
	}
	mutex_unlock(&mutex);

	return ret;
}
8227

8228
int sched_rr_handler(struct ctl_table *table, int write,
8229 8230 8231 8232 8233 8234 8235 8236
		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);
8237 8238
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
8239
	if (!ret && write) {
8240 8241
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
8242 8243 8244 8245 8246
	}
	mutex_unlock(&mutex);
	return ret;
}

8247
#ifdef CONFIG_CGROUP_SCHED
8248

8249
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
8250
{
8251
	return css ? container_of(css, struct task_group, css) : NULL;
8252 8253
}

8254 8255
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
8256
{
8257 8258
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
8259

8260
	if (!parent) {
8261
		/* This is early initialization for the top cgroup */
8262
		return &root_task_group.css;
8263 8264
	}

8265
	tg = sched_create_group(parent);
8266 8267 8268
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

8269 8270
	sched_online_group(tg, parent);

8271 8272 8273
	return &tg->css;
}

8274
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
8275
{
8276
	struct task_group *tg = css_tg(css);
8277

8278
	sched_offline_group(tg);
8279 8280
}

8281
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
8282
{
8283
	struct task_group *tg = css_tg(css);
8284

8285 8286 8287 8288
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
8289 8290
}

8291 8292 8293 8294
/*
 * This is called before wake_up_new_task(), therefore we really only
 * have to set its group bits, all the other stuff does not apply.
 */
8295
static void cpu_cgroup_fork(struct task_struct *task)
8296
{
8297 8298 8299 8300 8301 8302 8303 8304
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
8305 8306
}

8307
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
8308
{
8309
	struct task_struct *task;
8310
	struct cgroup_subsys_state *css;
8311
	int ret = 0;
8312

8313
	cgroup_taskset_for_each(task, css, tset) {
8314
#ifdef CONFIG_RT_GROUP_SCHED
8315
		if (!sched_rt_can_attach(css_tg(css), task))
8316
			return -EINVAL;
8317
#else
8318 8319 8320
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
8321
#endif
8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336 8337
		/*
		 * Serialize against wake_up_new_task() such that if its
		 * running, we're sure to observe its full state.
		 */
		raw_spin_lock_irq(&task->pi_lock);
		/*
		 * Avoid calling sched_move_task() before wake_up_new_task()
		 * has happened. This would lead to problems with PELT, due to
		 * move wanting to detach+attach while we're not attached yet.
		 */
		if (task->state == TASK_NEW)
			ret = -EINVAL;
		raw_spin_unlock_irq(&task->pi_lock);

		if (ret)
			break;
8338
	}
8339
	return ret;
8340
}
8341

8342
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
8343
{
8344
	struct task_struct *task;
8345
	struct cgroup_subsys_state *css;
8346

8347
	cgroup_taskset_for_each(task, css, tset)
8348
		sched_move_task(task);
8349 8350
}

8351
#ifdef CONFIG_FAIR_GROUP_SCHED
8352 8353
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
8354
{
8355
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
8356 8357
}

8358 8359
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
8360
{
8361
	struct task_group *tg = css_tg(css);
8362

8363
	return (u64) scale_load_down(tg->shares);
8364
}
8365 8366

#ifdef CONFIG_CFS_BANDWIDTH
8367 8368
static DEFINE_MUTEX(cfs_constraints_mutex);

8369 8370 8371
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

8372 8373
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

8374 8375
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
8376
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
8377
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397

	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;

8398 8399 8400 8401 8402
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
8403 8404 8405 8406 8407
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

8408
	runtime_enabled = quota != RUNTIME_INF;
8409
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
8410 8411 8412 8413 8414 8415
	/*
	 * If we need to toggle cfs_bandwidth_used, off->on must occur
	 * before making related changes, and on->off must occur afterwards
	 */
	if (runtime_enabled && !runtime_was_enabled)
		cfs_bandwidth_usage_inc();
8416 8417 8418
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
8419

P
Paul Turner 已提交
8420
	__refill_cfs_bandwidth_runtime(cfs_b);
8421
	/* restart the period timer (if active) to handle new period expiry */
P
Peter Zijlstra 已提交
8422 8423
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
8424 8425
	raw_spin_unlock_irq(&cfs_b->lock);

8426
	for_each_online_cpu(i) {
8427
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
8428
		struct rq *rq = cfs_rq->rq;
8429 8430

		raw_spin_lock_irq(&rq->lock);
8431
		cfs_rq->runtime_enabled = runtime_enabled;
8432
		cfs_rq->runtime_remaining = 0;
8433

8434
		if (cfs_rq->throttled)
8435
			unthrottle_cfs_rq(cfs_rq);
8436 8437
		raw_spin_unlock_irq(&rq->lock);
	}
8438 8439
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
8440 8441
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
8442
	put_online_cpus();
8443

8444
	return ret;
8445 8446 8447 8448 8449 8450
}

int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
{
	u64 quota, period;

8451
	period = ktime_to_ns(tg->cfs_bandwidth.period);
8452 8453 8454 8455 8456 8457 8458 8459 8460 8461 8462 8463
	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;

8464
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
8465 8466
		return -1;

8467
	quota_us = tg->cfs_bandwidth.quota;
8468 8469 8470 8471 8472 8473 8474 8475 8476 8477
	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;
8478
	quota = tg->cfs_bandwidth.quota;
8479 8480 8481 8482 8483 8484 8485 8486

	return tg_set_cfs_bandwidth(tg, period, quota);
}

long tg_get_cfs_period(struct task_group *tg)
{
	u64 cfs_period_us;

8487
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
8488 8489 8490 8491 8492
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

8493 8494
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
8495
{
8496
	return tg_get_cfs_quota(css_tg(css));
8497 8498
}

8499 8500
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
8501
{
8502
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
8503 8504
}

8505 8506
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8507
{
8508
	return tg_get_cfs_period(css_tg(css));
8509 8510
}

8511 8512
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
8513
{
8514
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
8515 8516
}

8517 8518 8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538 8539 8540 8541 8542 8543 8544 8545 8546 8547 8548
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;
8549
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8550 8551 8552 8553 8554
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
8555
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
8556 8557

		quota = normalize_cfs_quota(tg, d);
8558
		parent_quota = parent_b->hierarchical_quota;
8559 8560 8561 8562 8563 8564 8565 8566 8567 8568

		/*
		 * 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;
	}
8569
	cfs_b->hierarchical_quota = quota;
8570 8571 8572 8573 8574 8575

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
8576
	int ret;
8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587
	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);
	}

8588 8589 8590 8591 8592
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8593
}
8594

8595
static int cpu_stats_show(struct seq_file *sf, void *v)
8596
{
8597
	struct task_group *tg = css_tg(seq_css(sf));
8598
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8599

8600 8601 8602
	seq_printf(sf, "nr_periods %d\n", cfs_b->nr_periods);
	seq_printf(sf, "nr_throttled %d\n", cfs_b->nr_throttled);
	seq_printf(sf, "throttled_time %llu\n", cfs_b->throttled_time);
8603 8604 8605

	return 0;
}
8606
#endif /* CONFIG_CFS_BANDWIDTH */
8607
#endif /* CONFIG_FAIR_GROUP_SCHED */
8608

8609
#ifdef CONFIG_RT_GROUP_SCHED
8610 8611
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
8612
{
8613
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
8614 8615
}

8616 8617
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
8618
{
8619
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
8620
}
8621

8622 8623
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
8624
{
8625
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
8626 8627
}

8628 8629
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8630
{
8631
	return sched_group_rt_period(css_tg(css));
8632
}
8633
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
8634

8635
static struct cftype cpu_files[] = {
8636
#ifdef CONFIG_FAIR_GROUP_SCHED
8637 8638
	{
		.name = "shares",
8639 8640
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
8641
	},
8642
#endif
8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653
#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,
	},
8654 8655
	{
		.name = "stat",
8656
		.seq_show = cpu_stats_show,
8657
	},
8658
#endif
8659
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
8660
	{
P
Peter Zijlstra 已提交
8661
		.name = "rt_runtime_us",
8662 8663
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
8664
	},
8665 8666
	{
		.name = "rt_period_us",
8667 8668
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
8669
	},
8670
#endif
8671
	{ }	/* terminate */
8672 8673
};

8674
struct cgroup_subsys cpu_cgrp_subsys = {
8675
	.css_alloc	= cpu_cgroup_css_alloc,
8676
	.css_released	= cpu_cgroup_css_released,
8677
	.css_free	= cpu_cgroup_css_free,
8678
	.fork		= cpu_cgroup_fork,
8679 8680
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
8681
	.legacy_cftypes	= cpu_files,
8682
	.early_init	= true,
8683 8684
};

8685
#endif	/* CONFIG_CGROUP_SCHED */
8686

8687 8688 8689 8690 8691
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
8692 8693 8694 8695 8696 8697 8698 8699 8700 8701 8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715 8716 8717 8718 8719 8720 8721 8722 8723 8724 8725 8726 8727 8728 8729 8730 8731 8732

/*
 * Nice levels are multiplicative, with a gentle 10% change for every
 * nice level changed. I.e. when a CPU-bound task goes from nice 0 to
 * nice 1, it will get ~10% less CPU time than another CPU-bound task
 * that remained on nice 0.
 *
 * The "10% effect" is relative and cumulative: from _any_ nice level,
 * if you go up 1 level, it's -10% CPU usage, if you go down 1 level
 * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25.
 * If a task goes up by ~10% and another task goes down by ~10% then
 * the relative distance between them is ~25%.)
 */
const int sched_prio_to_weight[40] = {
 /* -20 */     88761,     71755,     56483,     46273,     36291,
 /* -15 */     29154,     23254,     18705,     14949,     11916,
 /* -10 */      9548,      7620,      6100,      4904,      3906,
 /*  -5 */      3121,      2501,      1991,      1586,      1277,
 /*   0 */      1024,       820,       655,       526,       423,
 /*   5 */       335,       272,       215,       172,       137,
 /*  10 */       110,        87,        70,        56,        45,
 /*  15 */        36,        29,        23,        18,        15,
};

/*
 * Inverse (2^32/x) values of the sched_prio_to_weight[] array, precalculated.
 *
 * In cases where the weight does not change often, we can use the
 * precalculated inverse to speed up arithmetics by turning divisions
 * into multiplications:
 */
const u32 sched_prio_to_wmult[40] = {
 /* -20 */     48388,     59856,     76040,     92818,    118348,
 /* -15 */    147320,    184698,    229616,    287308,    360437,
 /* -10 */    449829,    563644,    704093,    875809,   1099582,
 /*  -5 */   1376151,   1717300,   2157191,   2708050,   3363326,
 /*   0 */   4194304,   5237765,   6557202,   8165337,  10153587,
 /*   5 */  12820798,  15790321,  19976592,  24970740,  31350126,
 /*  10 */  39045157,  49367440,  61356676,  76695844,  95443717,
 /*  15 */ 119304647, 148102320, 186737708, 238609294, 286331153,
};