core.c 172.5 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 *  kernel/sched/core.c
L
Linus Torvalds 已提交
3
 *
I
Ingo Molnar 已提交
4
 *  Core kernel scheduler code and related syscalls
L
Linus Torvalds 已提交
5 6 7
 *
 *  Copyright (C) 1991-2002  Linus Torvalds
 */
8
#include <linux/sched.h>
9
#include <linux/sched/clock.h>
10
#include <uapi/linux/sched/types.h>
11
#include <linux/sched/loadavg.h>
12
#include <linux/sched/hotplug.h>
13
#include <linux/wait_bit.h>
L
Linus Torvalds 已提交
14
#include <linux/cpuset.h>
15
#include <linux/delayacct.h>
16
#include <linux/init_task.h>
17
#include <linux/context_tracking.h>
18
#include <linux/rcupdate_wait.h>
19
#include <linux/compat.h>
20 21 22 23 24 25

#include <linux/blkdev.h>
#include <linux/kprobes.h>
#include <linux/mmu_context.h>
#include <linux/module.h>
#include <linux/nmi.h>
26
#include <linux/prefetch.h>
27 28 29
#include <linux/profile.h>
#include <linux/security.h>
#include <linux/syscalls.h>
30
#include <linux/sched/isolation.h>
L
Linus Torvalds 已提交
31

32
#include <asm/switch_to.h>
33
#include <asm/tlb.h>
34 35 36
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
#endif
L
Linus Torvalds 已提交
37

38
#include "sched.h"
39
#include "../workqueue_internal.h"
40
#include "../smpboot.h"
41

42
#define CREATE_TRACE_POINTS
43
#include <trace/events/sched.h>
44

45
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
46

47
#if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
I
Ingo Molnar 已提交
48 49
/*
 * Debugging: various feature bits
50 51 52 53
 *
 * If SCHED_DEBUG is disabled, each compilation unit has its own copy of
 * sysctl_sched_features, defined in sched.h, to allow constants propagation
 * at compile time and compiler optimization based on features default.
I
Ingo Molnar 已提交
54
 */
P
Peter Zijlstra 已提交
55 56
#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |
I
Ingo Molnar 已提交
57
const_debug unsigned int sysctl_sched_features =
58
#include "features.h"
P
Peter Zijlstra 已提交
59 60
	0;
#undef SCHED_FEAT
61
#endif
P
Peter Zijlstra 已提交
62

63 64 65 66 67 68
/*
 * 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;

69 70 71 72 73 74 75 76
/*
 * 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
77
/*
I
Ingo Molnar 已提交
78
 * period over which we measure -rt task CPU usage in us.
P
Peter Zijlstra 已提交
79 80
 * default: 1s
 */
P
Peter Zijlstra 已提交
81
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
82

83
__read_mostly int scheduler_running;
84

P
Peter Zijlstra 已提交
85 86 87 88 89
/*
 * 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
90

91 92 93
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
94
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
95 96 97 98 99 100 101 102 103 104
	__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))) {
105
			rq_pin_lock(rq, rf);
106 107 108 109 110 111 112 113 114 115 116 117
			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.
 */
118
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
119 120 121 122 123 124
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
125
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
		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.
		 *
I
Ingo Molnar 已提交
141
		 * If we observe the new CPU in task_rq_lock, the acquire will
142 143 144
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
145
			rq_pin_lock(rq, rf);
146 147 148
			return rq;
		}
		raw_spin_unlock(&rq->lock);
149
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
150 151 152 153 154 155

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

156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223
/*
 * RQ-clock updating methods:
 */

static void update_rq_clock_task(struct rq *rq, s64 delta)
{
/*
 * 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
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;

	/*
	 * 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;
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
	if (static_key_false((&paravirt_steal_rq_enabled))) {
		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

	rq->clock_task += delta;

#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
}

void update_rq_clock(struct rq *rq)
{
	s64 delta;

	lockdep_assert_held(&rq->lock);

	if (rq->clock_update_flags & RQCF_ACT_SKIP)
		return;

#ifdef CONFIG_SCHED_DEBUG
224 225
	if (sched_feat(WARN_DOUBLE_CLOCK))
		SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
226 227
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
228

229 230 231 232 233 234 235 236
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
	if (delta < 0)
		return;
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
}


P
Peter Zijlstra 已提交
237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
#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);
255
	struct rq_flags rf;
P
Peter Zijlstra 已提交
256 257 258

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

259
	rq_lock(rq, &rf);
260
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
261
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
262
	rq_unlock(rq, &rf);
P
Peter Zijlstra 已提交
263 264 265 266

	return HRTIMER_NORESTART;
}

267
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
268

269
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
270 271 272
{
	struct hrtimer *timer = &rq->hrtick_timer;

273
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
274 275
}

276 277 278 279
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
280
{
281
	struct rq *rq = arg;
282
	struct rq_flags rf;
283

284
	rq_lock(rq, &rf);
P
Peter Zijlstra 已提交
285
	__hrtick_restart(rq);
286
	rq->hrtick_csd_pending = 0;
287
	rq_unlock(rq, &rf);
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
	 */
	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);
I
Ingo Molnar 已提交
461
		/* Task can safely be re-inserted now: */
462 463 464 465 466 467 468 469 470 471 472 473
		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
	raw_spin_lock_irqsave(&rq->lock, flags);
511 512
	if (cpu_online(cpu) || cpu == smp_processor_id())
		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
/*
I
Ingo Molnar 已提交
519 520
 * In the semi idle case, use the nearest busy CPU for migrating timers
 * from an idle CPU.  This is good for power-savings.
521 522
 *
 * We don't do similar optimization for completely idle system, as
I
Ingo Molnar 已提交
523 524
 * 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).
525
 */
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) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
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
			if (cpu == i)
				continue;

540
			if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
541 542 543 544
				cpu = i;
				goto unlock;
			}
		}
545
	}
546

547 548
	if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
		cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
549 550
unlock:
	rcu_read_unlock();
551 552
	return cpu;
}
I
Ingo Molnar 已提交
553

554 555 556 557 558 559 560 561 562 563
/*
 * 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.
 */
564
static void wake_up_idle_cpu(int cpu)
565 566 567 568 569 570
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

571
	if (set_nr_and_not_polling(rq->idle))
572
		smp_send_reschedule(cpu);
573 574
	else
		trace_sched_wake_idle_without_ipi(cpu);
575 576
}

577
static bool wake_up_full_nohz_cpu(int cpu)
578
{
579 580 581 582 583 584
	/*
	 * 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.
	 */
585 586
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
587
	if (tick_nohz_full_cpu(cpu)) {
588 589
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
590
			tick_nohz_full_kick_cpu(cpu);
591 592 593 594 595 596
		return true;
	}

	return false;
}

597 598 599 600 601
/*
 * Wake up the specified CPU.  If the CPU is going offline, it is the
 * caller's responsibility to deal with the lost wakeup, for example,
 * by hooking into the CPU_DEAD notifier like timers and hrtimers do.
 */
602 603
void wake_up_nohz_cpu(int cpu)
{
604
	if (!wake_up_full_nohz_cpu(cpu))
605 606 607
		wake_up_idle_cpu(cpu);
}

608
static inline bool got_nohz_idle_kick(void)
609
{
610
	int cpu = smp_processor_id();
611 612 613 614 615 616 617 618 619 620 621 622 623

	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;
624 625
}

626
#else /* CONFIG_NO_HZ_COMMON */
627

628
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
629
{
630
	return false;
P
Peter Zijlstra 已提交
631 632
}

633
#endif /* CONFIG_NO_HZ_COMMON */
634

635
#ifdef CONFIG_NO_HZ_FULL
636
bool sched_can_stop_tick(struct rq *rq)
637
{
638 639 640 641 642 643
	int fifo_nr_running;

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

644
	/*
645 646
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
647
	 */
648 649 650 651 652
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
653 654
	}

655 656 657 658 659 660 661 662 663 664 665 666 667 668
	/*
	 * 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)
669
		return false;
670

671
	return true;
672 673
}
#endif /* CONFIG_NO_HZ_FULL */
674

675
void sched_avg_update(struct rq *rq)
676
{
677 678
	s64 period = sched_avg_period();

679
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
680 681 682 683 684 685
		/*
		 * 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));
686 687 688
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
689 690
}

691
#endif /* CONFIG_SMP */
692

693 694
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
695
/*
696 697 698 699
 * 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.
700
 */
701
int walk_tg_tree_from(struct task_group *from,
702
			     tg_visitor down, tg_visitor up, void *data)
703 704
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
705
	int ret;
706

707 708
	parent = from;

709
down:
P
Peter Zijlstra 已提交
710 711
	ret = (*down)(parent, data);
	if (ret)
712
		goto out;
713 714 715 716 717 718 719
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
720
	ret = (*up)(parent, data);
721 722
	if (ret || parent == from)
		goto out;
723 724 725 726 727

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
728
out:
P
Peter Zijlstra 已提交
729
	return ret;
730 731
}

732
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
733
{
734
	return 0;
P
Peter Zijlstra 已提交
735
}
736 737
#endif

738
static void set_load_weight(struct task_struct *p, bool update_load)
739
{
N
Nikhil Rao 已提交
740 741 742
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
743 744 745
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
746
	if (idle_policy(p->policy)) {
747
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
748
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
749 750
		return;
	}
751

752 753 754 755 756 757 758 759 760 761
	/*
	 * SCHED_OTHER tasks have to update their load when changing their
	 * weight
	 */
	if (update_load && p->sched_class == &fair_sched_class) {
		reweight_task(p, prio);
	} else {
		load->weight = scale_load(sched_prio_to_weight[prio]);
		load->inv_weight = sched_prio_to_wmult[prio];
	}
762 763
}

764
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
765
{
766 767 768
	if (!(flags & ENQUEUE_NOCLOCK))
		update_rq_clock(rq);

769 770
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
771

772
	p->sched_class->enqueue_task(rq, p, flags);
773 774
}

775
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
776
{
777 778 779
	if (!(flags & DEQUEUE_NOCLOCK))
		update_rq_clock(rq);

780 781
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
782

783
	p->sched_class->dequeue_task(rq, p, flags);
784 785
}

786
void activate_task(struct rq *rq, struct task_struct *p, int flags)
787 788 789 790
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

791
	enqueue_task(rq, p, flags);
792 793
}

794
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
795 796 797 798
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

799
	dequeue_task(rq, p, flags);
800 801
}

802
/*
I
Ingo Molnar 已提交
803
 * __normal_prio - return the priority that is based on the static prio
804 805 806
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
807
	return p->static_prio;
808 809
}

810 811 812 813 814 815 816
/*
 * 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.
 */
817
static inline int normal_prio(struct task_struct *p)
818 819 820
{
	int prio;

821 822 823
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
824 825 826 827 828 829 830 831 832 833 834 835 836
		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.
 */
837
static int effective_prio(struct task_struct *p)
838 839 840 841 842 843 844 845 846 847 848 849
{
	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
850 851 852
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
853 854
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
855
 */
856
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
857 858 859 860
{
	return cpu_curr(task_cpu(p)) == p;
}

861
/*
862 863 864 865 866
 * 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().
867
 */
868 869
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
870
				       int oldprio)
871 872 873
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
874
			prev_class->switched_from(rq, p);
875

P
Peter Zijlstra 已提交
876
		p->sched_class->switched_to(rq, p);
877
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
878
		p->sched_class->prio_changed(rq, p, oldprio);
879 880
}

881
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
882 883 884 885 886 887 888 889 890 891
{
	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) {
892
				resched_curr(rq);
893 894 895 896 897 898 899 900 901
				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.
	 */
902
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
903
		rq_clock_skip_update(rq, true);
904 905
}

L
Linus Torvalds 已提交
906
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925
/*
 * 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.
 */
926 927
static struct rq *move_queued_task(struct rq *rq, struct rq_flags *rf,
				   struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
928 929 930 931
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
932
	dequeue_task(rq, p, DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
933
	set_task_cpu(p, new_cpu);
934
	rq_unlock(rq, rf);
P
Peter Zijlstra 已提交
935 936 937

	rq = cpu_rq(new_cpu);

938
	rq_lock(rq, rf);
P
Peter Zijlstra 已提交
939 940
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
941
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
942 943 944 945 946 947 948 949 950 951 952
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
I
Ingo Molnar 已提交
953
 * Move (not current) task off this CPU, onto the destination CPU. We're doing
P
Peter Zijlstra 已提交
954 955 956 957 958 959 960
 * 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.
 */
961 962
static struct rq *__migrate_task(struct rq *rq, struct rq_flags *rf,
				 struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
963
{
964 965 966 967 968 969 970
	if (p->flags & PF_KTHREAD) {
		if (unlikely(!cpu_online(dest_cpu)))
			return rq;
	} else {
		if (unlikely(!cpu_active(dest_cpu)))
			return rq;
	}
P
Peter Zijlstra 已提交
971 972

	/* Affinity changed (again). */
973
	if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
974
		return rq;
P
Peter Zijlstra 已提交
975

976
	update_rq_clock(rq);
977
	rq = move_queued_task(rq, rf, p, dest_cpu);
978 979

	return rq;
P
Peter Zijlstra 已提交
980 981 982 983 984 985 986 987 988 989
}

/*
 * 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;
990 991
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
992
	struct rq_flags rf;
P
Peter Zijlstra 已提交
993 994

	/*
I
Ingo Molnar 已提交
995 996
	 * The original target CPU might have gone down and we might
	 * be on another CPU but it doesn't matter.
P
Peter Zijlstra 已提交
997 998 999 1000 1001 1002 1003 1004
	 */
	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();
1005 1006

	raw_spin_lock(&p->pi_lock);
1007
	rq_lock(rq, &rf);
1008 1009 1010 1011 1012
	/*
	 * 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.
	 */
1013 1014
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
1015
			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
1016 1017 1018
		else
			p->wake_cpu = arg->dest_cpu;
	}
1019
	rq_unlock(rq, &rf);
1020 1021
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1022 1023 1024 1025
	local_irq_enable();
	return 0;
}

1026 1027 1028 1029 1030
/*
 * 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
1031 1032 1033 1034 1035
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1036 1037
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1038 1039 1040
	struct rq *rq = task_rq(p);
	bool queued, running;

1041
	lockdep_assert_held(&p->pi_lock);
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051

	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);
1052
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1053 1054 1055 1056
	}
	if (running)
		put_prev_task(rq, p);

1057
	p->sched_class->set_cpus_allowed(p, new_mask);
1058 1059

	if (queued)
1060
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1061
	if (running)
1062
		set_curr_task(rq, p);
1063 1064
}

P
Peter Zijlstra 已提交
1065 1066 1067 1068 1069 1070 1071 1072 1073
/*
 * 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.
 */
1074 1075
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1076
{
1077
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1078
	unsigned int dest_cpu;
1079 1080
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1081 1082
	int ret = 0;

1083
	rq = task_rq_lock(p, &rf);
1084
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
1085

1086 1087 1088 1089 1090 1091 1092
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1093 1094 1095 1096 1097 1098 1099 1100 1101
	/*
	 * 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
1102 1103 1104
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1105
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1106 1107 1108 1109 1110 1111
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1112 1113 1114
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
I
Ingo Molnar 已提交
1115
		 * !active we want to ensure they are strict per-CPU threads.
1116 1117 1118 1119 1120 1121
		 */
		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
1122 1123 1124 1125
	/* 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;

1126
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1127 1128 1129
	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. */
1130
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1131 1132 1133
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1134 1135 1136 1137 1138
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1139
		rq = move_queued_task(rq, &rf, p, dest_cpu);
1140
	}
P
Peter Zijlstra 已提交
1141
out:
1142
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1143 1144 1145

	return ret;
}
1146 1147 1148 1149 1150

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
1151 1152
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1153
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1154
{
1155 1156 1157 1158 1159
#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
1160
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1161
			!p->on_rq);
1162

1163 1164 1165 1166 1167 1168 1169 1170 1171
	/*
	 * 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)));

1172
#ifdef CONFIG_LOCKDEP
1173 1174 1175 1176 1177
	/*
	 * 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
1178
	 * see task_group().
1179 1180 1181 1182
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1183 1184 1185
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1186 1187 1188 1189
	/*
	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
	 */
	WARN_ON_ONCE(!cpu_online(new_cpu));
1190 1191
#endif

1192
	trace_sched_migrate_task(p, new_cpu);
1193

1194
	if (task_cpu(p) != new_cpu) {
1195
		if (p->sched_class->migrate_task_rq)
1196
			p->sched_class->migrate_task_rq(p);
1197
		p->se.nr_migrations++;
1198
		perf_event_task_migrate(p);
1199
	}
I
Ingo Molnar 已提交
1200 1201

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1202 1203
}

1204 1205
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1206
	if (task_on_rq_queued(p)) {
1207
		struct rq *src_rq, *dst_rq;
1208
		struct rq_flags srf, drf;
1209 1210 1211 1212

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

1213 1214 1215
		rq_pin_lock(src_rq, &srf);
		rq_pin_lock(dst_rq, &drf);

1216
		p->on_rq = TASK_ON_RQ_MIGRATING;
1217 1218 1219
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1220
		p->on_rq = TASK_ON_RQ_QUEUED;
1221
		check_preempt_curr(dst_rq, p, 0);
1222 1223 1224 1225

		rq_unpin_lock(dst_rq, &drf);
		rq_unpin_lock(src_rq, &srf);

1226 1227 1228 1229
	} 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
I
Ingo Molnar 已提交
1230
		 * previous CPU our target instead of where it really is.
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246
		 */
		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;

1247 1248 1249
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1250 1251 1252
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1253 1254
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1255
	double_rq_lock(src_rq, dst_rq);
1256

1257 1258 1259 1260 1261 1262
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

1263
	if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1264 1265
		goto unlock;

1266
	if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1267 1268 1269 1270 1271 1272 1273 1274 1275
		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);
1276 1277
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299

	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;

1300 1301 1302 1303
	/*
	 * 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.
	 */
1304 1305 1306
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

1307
	if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
1308 1309
		goto out;

1310
	if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1311 1312
		goto out;

1313
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1314 1315 1316 1317 1318 1319
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1320 1321 1322
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1323 1324 1325 1326 1327 1328 1329
 * 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
1330 1331 1332 1333 1334 1335
 * 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
1336
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1337
{
1338
	int running, queued;
1339
	struct rq_flags rf;
R
Roland McGrath 已提交
1340
	unsigned long ncsw;
1341
	struct rq *rq;
L
Linus Torvalds 已提交
1342

1343 1344 1345 1346 1347 1348 1349 1350
	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);
1351

1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362
		/*
		 * 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
1363 1364 1365
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1366
			cpu_relax();
R
Roland McGrath 已提交
1367
		}
1368

1369 1370 1371 1372 1373
		/*
		 * 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.
		 */
1374
		rq = task_rq_lock(p, &rf);
1375
		trace_sched_wait_task(p);
1376
		running = task_running(rq, p);
1377
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1378
		ncsw = 0;
1379
		if (!match_state || p->state == match_state)
1380
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1381
		task_rq_unlock(rq, p, &rf);
1382

R
Roland McGrath 已提交
1383 1384 1385 1386 1387 1388
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1389 1390 1391 1392 1393 1394 1395 1396 1397 1398
		/*
		 * 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;
		}
1399

1400 1401 1402 1403 1404
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1405
		 * So if it was still runnable (but just not actively
1406 1407 1408
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1409
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1410
			ktime_t to = NSEC_PER_SEC / HZ;
1411 1412 1413

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1414 1415
			continue;
		}
1416

1417 1418 1419 1420 1421 1422 1423
		/*
		 * 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
1424 1425

	return ncsw;
L
Linus Torvalds 已提交
1426 1427 1428 1429 1430 1431 1432 1433 1434
}

/***
 * 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
1435
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1436 1437 1438 1439 1440
 * 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.
 */
1441
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1442 1443 1444 1445 1446 1447 1448 1449 1450
{
	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
1451
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1452

1453
/*
1454
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1455 1456 1457 1458 1459 1460 1461
 *
 * 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
I
Ingo Molnar 已提交
1462
 *    CPU isn't yet part of the sched domains, and balancing will not
1463 1464
 *    see it.
 *
I
Ingo Molnar 已提交
1465
 *  - on CPU-down we clear cpu_active() to mask the sched domains and
1466
 *    avoid the load balancer to place new tasks on the to be removed
I
Ingo Molnar 已提交
1467
 *    CPU. Existing tasks will remain running there and will be taken
1468 1469 1470 1471 1472 1473
 *    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.
1474
 */
1475 1476
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1477 1478
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1479 1480
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1481

1482
	/*
I
Ingo Molnar 已提交
1483 1484 1485
	 * 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.
1486 1487 1488 1489 1490 1491 1492 1493
	 */
	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;
1494
			if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1495 1496
				return dest_cpu;
		}
1497
	}
1498

1499 1500
	for (;;) {
		/* Any allowed, online CPU? */
1501
		for_each_cpu(dest_cpu, &p->cpus_allowed) {
1502 1503 1504
			if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu))
				continue;
			if (!cpu_online(dest_cpu))
1505 1506 1507
				continue;
			goto out;
		}
1508

1509
		/* No more Mr. Nice Guy. */
1510 1511
		switch (state) {
		case cpuset:
1512 1513 1514 1515 1516
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
I
Ingo Molnar 已提交
1517
			/* Fall-through */
1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536
		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()) {
1537
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1538 1539
					task_pid_nr(p), p->comm, cpu);
		}
1540 1541 1542 1543 1544
	}

	return dest_cpu;
}

1545
/*
1546
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1547
 */
1548
static inline
1549
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1550
{
1551 1552
	lockdep_assert_held(&p->pi_lock);

1553
	if (p->nr_cpus_allowed > 1)
1554
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1555
	else
1556
		cpu = cpumask_any(&p->cpus_allowed);
1557 1558 1559 1560

	/*
	 * 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
I
Ingo Molnar 已提交
1561
	 * CPU.
1562 1563 1564 1565 1566 1567
	 *
	 * 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 ]
	 */
1568
	if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
P
Peter Zijlstra 已提交
1569
		     !cpu_online(cpu)))
1570
		cpu = select_fallback_rq(task_cpu(p), p);
1571 1572

	return cpu;
1573
}
1574 1575 1576 1577 1578 1579

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1580

1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610
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;
	}
}

1611 1612 1613 1614 1615 1616 1617 1618
#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
1619
#endif /* CONFIG_SMP */
1620

P
Peter Zijlstra 已提交
1621
static void
1622
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1623
{
1624
	struct rq *rq;
1625

1626 1627 1628 1629
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1630

1631 1632
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1633 1634
		schedstat_inc(rq->ttwu_local);
		schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1635 1636 1637
	} else {
		struct sched_domain *sd;

1638
		schedstat_inc(p->se.statistics.nr_wakeups_remote);
1639
		rcu_read_lock();
1640
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1641
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1642
				schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1643 1644 1645
				break;
			}
		}
1646
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1647
	}
1648 1649

	if (wake_flags & WF_MIGRATED)
1650
		schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1651 1652
#endif /* CONFIG_SMP */

1653 1654
	schedstat_inc(rq->ttwu_count);
	schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1655 1656

	if (wake_flags & WF_SYNC)
1657
		schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1658 1659
}

1660
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1661
{
T
Tejun Heo 已提交
1662
	activate_task(rq, p, en_flags);
1663
	p->on_rq = TASK_ON_RQ_QUEUED;
1664

I
Ingo Molnar 已提交
1665
	/* If a worker is waking up, notify the workqueue: */
1666 1667
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1668 1669
}

1670 1671 1672
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1673
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1674
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1675 1676 1677
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1678 1679
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1680
#ifdef CONFIG_SMP
1681 1682
	if (p->sched_class->task_woken) {
		/*
1683 1684
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1685
		 */
1686
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1687
		p->sched_class->task_woken(rq, p);
1688
		rq_repin_lock(rq, rf);
1689
	}
T
Tejun Heo 已提交
1690

1691
	if (rq->idle_stamp) {
1692
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1693
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1694

1695 1696 1697
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1698
			rq->avg_idle = max;
1699

T
Tejun Heo 已提交
1700 1701 1702 1703 1704
		rq->idle_stamp = 0;
	}
#endif
}

1705
static void
1706
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1707
		 struct rq_flags *rf)
1708
{
1709
	int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1710

1711 1712
	lockdep_assert_held(&rq->lock);

1713 1714 1715
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1716 1717

	if (wake_flags & WF_MIGRATED)
1718
		en_flags |= ENQUEUE_MIGRATED;
1719 1720
#endif

1721
	ttwu_activate(rq, p, en_flags);
1722
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1723 1724 1725 1726 1727 1728 1729 1730 1731 1732
}

/*
 * 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)
{
1733
	struct rq_flags rf;
1734 1735 1736
	struct rq *rq;
	int ret = 0;

1737
	rq = __task_rq_lock(p, &rf);
1738
	if (task_on_rq_queued(p)) {
1739 1740
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1741
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1742 1743
		ret = 1;
	}
1744
	__task_rq_unlock(rq, &rf);
1745 1746 1747 1748

	return ret;
}

1749
#ifdef CONFIG_SMP
1750
void sched_ttwu_pending(void)
1751 1752
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1753
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1754
	struct task_struct *p, *t;
1755
	struct rq_flags rf;
1756

1757 1758 1759
	if (!llist)
		return;

1760
	rq_lock_irqsave(rq, &rf);
1761
	update_rq_clock(rq);
1762

1763 1764
	llist_for_each_entry_safe(p, t, llist, wake_entry)
		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1765

1766
	rq_unlock_irqrestore(rq, &rf);
1767 1768 1769 1770
}

void scheduler_ipi(void)
{
1771 1772 1773 1774 1775
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1776
	preempt_fold_need_resched();
1777

1778
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794
		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
1795
	sched_ttwu_pending();
1796 1797 1798 1799

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1800
	if (unlikely(got_nohz_idle_kick())) {
1801
		this_rq()->idle_balance = 1;
1802
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1803
	}
1804
	irq_exit();
1805 1806
}

P
Peter Zijlstra 已提交
1807
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1808
{
1809 1810
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1811 1812
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1813 1814 1815 1816 1817 1818
	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);
	}
1819
}
1820

1821 1822 1823
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
1824
	struct rq_flags rf;
1825

1826 1827 1828 1829
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1830 1831 1832 1833

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
1834
		rq_lock_irqsave(rq, &rf);
1835 1836
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
I
Ingo Molnar 已提交
1837
		/* Else CPU is not idle, do nothing here: */
1838
		rq_unlock_irqrestore(rq, &rf);
1839
	}
1840 1841 1842

out:
	rcu_read_unlock();
1843 1844
}

1845
bool cpus_share_cache(int this_cpu, int that_cpu)
1846 1847 1848
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1849
#endif /* CONFIG_SMP */
1850

1851
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1852 1853
{
	struct rq *rq = cpu_rq(cpu);
1854
	struct rq_flags rf;
1855

1856
#if defined(CONFIG_SMP)
1857
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
I
Ingo Molnar 已提交
1858
		sched_clock_cpu(cpu); /* Sync clocks across CPUs */
P
Peter Zijlstra 已提交
1859
		ttwu_queue_remote(p, cpu, wake_flags);
1860 1861 1862 1863
		return;
	}
#endif

1864
	rq_lock(rq, &rf);
1865
	update_rq_clock(rq);
1866
	ttwu_do_activate(rq, p, wake_flags, &rf);
1867
	rq_unlock(rq, &rf);
T
Tejun Heo 已提交
1868 1869
}

1870 1871 1872 1873 1874 1875
/*
 * Notes on Program-Order guarantees on SMP systems.
 *
 *  MIGRATION
 *
 * The basic program-order guarantee on SMP systems is that when a task [t]
I
Ingo Molnar 已提交
1876 1877
 * migrates, all its activity on its old CPU [c0] happens-before any subsequent
 * execution on its new CPU [c1].
1878 1879 1880 1881 1882 1883 1884 1885 1886 1887
 *
 * 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.
I
Ingo Molnar 已提交
1888
 * Note: the CPU doing B need not be c0 or c1
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
 *
 * 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)
1920
 *   2) smp_cond_load_acquire(!X->on_cpu)
1921 1922 1923 1924 1925 1926 1927 1928 1929 1930
 *
 * 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);
 *
1931
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956
 *                    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,
1957
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1958 1959 1960
 *
 */

T
Tejun Heo 已提交
1961
/**
L
Linus Torvalds 已提交
1962
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1963
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1964
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1965
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1966
 *
1967
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1968
 *
1969 1970 1971 1972 1973 1974 1975
 * If the task was not queued/runnable, also place it back on a runqueue.
 *
 * Atomic against schedule() which would dequeue a task, also see
 * set_current_state().
 *
 * Return: %true if @p->state changes (an actual wakeup was done),
 *	   %false otherwise.
L
Linus Torvalds 已提交
1976
 */
1977 1978
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1979 1980
{
	unsigned long flags;
1981
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1982

1983 1984 1985 1986 1987 1988
	/*
	 * 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.
	 */
1989
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1990
	smp_mb__after_spinlock();
P
Peter Zijlstra 已提交
1991
	if (!(p->state & state))
L
Linus Torvalds 已提交
1992 1993
		goto out;

1994 1995
	trace_sched_waking(p);

I
Ingo Molnar 已提交
1996 1997
	/* We're going to change ->state: */
	success = 1;
L
Linus Torvalds 已提交
1998 1999
	cpu = task_cpu(p);

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
	/*
	 * 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();
2022 2023
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2024 2025

#ifdef CONFIG_SMP
2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044
	/*
	 * 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
2045
	/*
I
Ingo Molnar 已提交
2046
	 * If the owning (remote) CPU is still in the middle of schedule() with
2047
	 * this task as prev, wait until its done referencing the task.
2048
	 *
2049
	 * Pairs with the smp_store_release() in finish_task().
2050 2051 2052
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2053
	 */
2054
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2055

2056
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2057
	p->state = TASK_WAKING;
2058

2059
	if (p->in_iowait) {
2060
		delayacct_blkio_end(p);
2061 2062 2063
		atomic_dec(&task_rq(p)->nr_iowait);
	}

2064
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2065 2066
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2067
		set_task_cpu(p, cpu);
2068
	}
2069 2070 2071 2072

#else /* CONFIG_SMP */

	if (p->in_iowait) {
2073
		delayacct_blkio_end(p);
2074 2075 2076
		atomic_dec(&task_rq(p)->nr_iowait);
	}

L
Linus Torvalds 已提交
2077 2078
#endif /* CONFIG_SMP */

2079
	ttwu_queue(p, cpu, wake_flags);
2080
stat:
2081
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2082
out:
2083
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2084 2085 2086 2087

	return success;
}

T
Tejun Heo 已提交
2088 2089 2090
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2091
 * @rf: request-queue flags for pinning
T
Tejun Heo 已提交
2092
 *
2093
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2094
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2095
 * the current task.
T
Tejun Heo 已提交
2096
 */
2097
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2098 2099 2100
{
	struct rq *rq = task_rq(p);

2101 2102 2103 2104
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2105 2106
	lockdep_assert_held(&rq->lock);

2107
	if (!raw_spin_trylock(&p->pi_lock)) {
2108 2109 2110 2111 2112 2113
		/*
		 * 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.
		 */
2114
		rq_unlock(rq, rf);
2115
		raw_spin_lock(&p->pi_lock);
2116
		rq_relock(rq, rf);
2117 2118
	}

T
Tejun Heo 已提交
2119
	if (!(p->state & TASK_NORMAL))
2120
		goto out;
T
Tejun Heo 已提交
2121

2122 2123
	trace_sched_waking(p);

2124 2125
	if (!task_on_rq_queued(p)) {
		if (p->in_iowait) {
2126
			delayacct_blkio_end(p);
2127 2128
			atomic_dec(&rq->nr_iowait);
		}
2129
		ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2130
	}
P
Peter Zijlstra 已提交
2131

2132
	ttwu_do_wakeup(rq, p, 0, rf);
2133
	ttwu_stat(p, smp_processor_id(), 0);
2134 2135
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2136 2137
}

2138 2139 2140 2141 2142
/**
 * 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
2143 2144 2145
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2146 2147 2148 2149
 *
 * 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.
 */
2150
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2151
{
2152
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2153 2154 2155
}
EXPORT_SYMBOL(wake_up_process);

2156
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2157 2158 2159 2160 2161 2162 2163
{
	return try_to_wake_up(p, state, 0);
}

/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2164 2165 2166
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2167
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2168
{
P
Peter Zijlstra 已提交
2169 2170 2171
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2172 2173
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2174
	p->se.prev_sum_exec_runtime	= 0;
2175
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2176
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2177
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2178

2179 2180 2181 2182
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2183
#ifdef CONFIG_SCHEDSTATS
2184
	/* Even if schedstat is disabled, there should not be garbage */
2185
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2186
#endif
N
Nick Piggin 已提交
2187

2188
	RB_CLEAR_NODE(&p->dl.rb_node);
2189
	init_dl_task_timer(&p->dl);
2190
	init_dl_inactive_task_timer(&p->dl);
2191
	__dl_clear_params(p);
2192

P
Peter Zijlstra 已提交
2193
	INIT_LIST_HEAD(&p->rt.run_list);
2194 2195 2196 2197
	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
2198

2199 2200 2201
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2202 2203 2204

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2205
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2206 2207 2208
		p->mm->numa_scan_seq = 0;
	}

2209 2210 2211 2212 2213
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2214 2215
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2216
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2217
	p->numa_work.next = &p->numa_work;
2218
	p->numa_faults = NULL;
2219 2220
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2221 2222

	p->numa_group = NULL;
2223
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2224 2225
}

2226 2227
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2228
#ifdef CONFIG_NUMA_BALANCING
2229

2230 2231 2232
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2233
		static_branch_enable(&sched_numa_balancing);
2234
	else
2235
		static_branch_disable(&sched_numa_balancing);
2236
}
2237 2238 2239 2240 2241 2242 2243

#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;
2244
	int state = static_branch_likely(&sched_numa_balancing);
2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259

	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
2260

2261 2262
#ifdef CONFIG_SCHEDSTATS

2263
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2264
static bool __initdata __sched_schedstats = false;
2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287

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;

2288 2289 2290 2291 2292
	/*
	 * 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.
	 */
2293
	if (!strcmp(str, "enable")) {
2294
		__sched_schedstats = true;
2295 2296
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2297
		__sched_schedstats = false;
2298 2299 2300 2301 2302 2303 2304 2305 2306 2307
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2308 2309 2310 2311 2312
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332
#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;
}
2333 2334 2335 2336
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2337 2338 2339 2340

/*
 * fork()/clone()-time setup:
 */
2341
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2342
{
2343
	unsigned long flags;
I
Ingo Molnar 已提交
2344 2345
	int cpu = get_cpu();

2346
	__sched_fork(clone_flags, p);
2347
	/*
2348
	 * We mark the process as NEW here. This guarantees that
2349 2350 2351
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2352
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2353

2354 2355 2356 2357 2358
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2359 2360 2361 2362
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2363
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2364
			p->policy = SCHED_NORMAL;
2365
			p->static_prio = NICE_TO_PRIO(0);
2366 2367 2368 2369 2370
			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);
2371
		set_load_weight(p, false);
2372

2373 2374 2375 2376 2377 2378
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2379

2380 2381 2382 2383 2384 2385
	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
2386
		p->sched_class = &fair_sched_class;
2387
	}
2388

2389
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2390

2391 2392 2393 2394 2395 2396 2397
	/*
	 * 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.
	 */
2398
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2399
	/*
I
Ingo Molnar 已提交
2400
	 * We're setting the CPU for the first time, we don't migrate,
2401 2402 2403 2404 2405
	 * so use __set_task_cpu().
	 */
	__set_task_cpu(p, cpu);
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2406
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2407

2408
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2409
	if (likely(sched_info_on()))
2410
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2411
#endif
P
Peter Zijlstra 已提交
2412 2413
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2414
#endif
2415
	init_task_preempt_count(p);
2416
#ifdef CONFIG_SMP
2417
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2418
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2419
#endif
2420

N
Nick Piggin 已提交
2421
	put_cpu();
2422
	return 0;
L
Linus Torvalds 已提交
2423 2424
}

2425 2426 2427
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
2428
		return BW_UNIT;
2429 2430 2431 2432 2433 2434 2435 2436 2437

	/*
	 * 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;

2438
	return div64_u64(runtime << BW_SHIFT, period);
2439 2440
}

L
Linus Torvalds 已提交
2441 2442 2443 2444 2445 2446 2447
/*
 * 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.
 */
2448
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2449
{
2450
	struct rq_flags rf;
I
Ingo Molnar 已提交
2451
	struct rq *rq;
2452

2453
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2454
	p->state = TASK_RUNNING;
2455 2456 2457 2458
#ifdef CONFIG_SMP
	/*
	 * Fork balancing, do it here and not earlier because:
	 *  - cpus_allowed can change in the fork path
I
Ingo Molnar 已提交
2459
	 *  - any previously selected CPU might disappear through hotplug
2460 2461 2462
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2463
	 */
2464
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2465
#endif
2466
	rq = __task_rq_lock(p, &rf);
2467
	update_rq_clock(rq);
2468
	post_init_entity_util_avg(&p->se);
2469

2470
	activate_task(rq, p, ENQUEUE_NOCLOCK);
2471
	p->on_rq = TASK_ON_RQ_QUEUED;
2472
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2473
	check_preempt_curr(rq, p, WF_FORK);
2474
#ifdef CONFIG_SMP
2475 2476 2477 2478 2479
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2480
		rq_unpin_lock(rq, &rf);
2481
		p->sched_class->task_woken(rq, p);
2482
		rq_repin_lock(rq, &rf);
2483
	}
2484
#endif
2485
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2486 2487
}

2488 2489
#ifdef CONFIG_PREEMPT_NOTIFIERS

2490 2491
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503
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);

2504
/**
2505
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2506
 * @notifier: notifier struct to register
2507 2508 2509
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2510 2511 2512
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2513 2514 2515 2516 2517 2518
	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
2519
 * @notifier: notifier struct to unregister
2520
 *
2521
 * This is *not* safe to call from within a preemption notifier.
2522 2523 2524 2525 2526 2527 2528
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2529
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2530 2531 2532
{
	struct preempt_notifier *notifier;

2533
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2534 2535 2536
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2537 2538 2539 2540 2541 2542
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);
}

2543
static void
2544 2545
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2546 2547 2548
{
	struct preempt_notifier *notifier;

2549
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2550 2551 2552
		notifier->ops->sched_out(notifier, next);
}

2553 2554 2555 2556 2557 2558 2559 2560
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);
}

2561
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2562

2563
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2564 2565 2566
{
}

2567
static inline void
2568 2569 2570 2571 2572
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2573
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2574

2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618
static inline void prepare_task(struct task_struct *next)
{
#ifdef CONFIG_SMP
	/*
	 * Claim the task as running, we do this before switching to it
	 * such that any running task will have this set.
	 */
	next->on_cpu = 1;
#endif
}

static inline void finish_task(struct task_struct *prev)
{
#ifdef CONFIG_SMP
	/*
	 * After ->on_cpu is cleared, the task can be moved to a different CPU.
	 * We must ensure this doesn't happen until the switch is completely
	 * finished.
	 *
	 * In particular, the load of prev->state in finish_task_switch() must
	 * happen before this.
	 *
	 * Pairs with the smp_cond_load_acquire() in try_to_wake_up().
	 */
	smp_store_release(&prev->on_cpu, 0);
#endif
}

static inline void finish_lock_switch(struct rq *rq)
{
#ifdef CONFIG_DEBUG_SPINLOCK
	/* this is a valid case when another task releases the spinlock */
	rq->lock.owner = current;
#endif
	/*
	 * If we are tracking spinlock dependencies then we have to
	 * fix up the runqueue lock - which gets 'carried over' from
	 * prev into current:
	 */
	spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_);

	raw_spin_unlock_irq(&rq->lock);
}

2619 2620 2621
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2622
 * @prev: the current task that is being switched out
2623 2624 2625 2626 2627 2628 2629 2630 2631
 * @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.
 */
2632 2633 2634
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2635
{
2636
	sched_info_switch(rq, prev, next);
2637
	perf_event_task_sched_out(prev, next);
2638
	fire_sched_out_preempt_notifiers(prev, next);
2639
	prepare_task(next);
2640 2641 2642
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2643 2644 2645 2646
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2647 2648 2649 2650
 * 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
2651 2652
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2653
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2654 2655
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2656 2657 2658 2659 2660
 *
 * 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
2661
 */
2662
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2663 2664
	__releases(rq->lock)
{
2665
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2666
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2667
	long prev_state;
L
Linus Torvalds 已提交
2668

2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679
	/*
	 * 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.
	 */
2680 2681 2682 2683
	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);
2684

L
Linus Torvalds 已提交
2685 2686 2687 2688
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2689
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2690 2691
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2692 2693
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
2694
	 * finish_task), otherwise a concurrent wakeup can get prev
2695 2696
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2697
	 */
O
Oleg Nesterov 已提交
2698
	prev_state = prev->state;
2699
	vtime_task_switch(prev);
2700
	perf_event_task_sched_in(prev, current);
2701 2702
	finish_task(prev);
	finish_lock_switch(rq);
2703
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2704

2705
	fire_sched_in_preempt_notifiers(current);
2706
	/*
2707 2708 2709 2710 2711 2712 2713 2714 2715 2716
	 * When switching through a kernel thread, the loop in
	 * membarrier_{private,global}_expedited() may have observed that
	 * kernel thread and not issued an IPI. It is therefore possible to
	 * schedule between user->kernel->user threads without passing though
	 * switch_mm(). Membarrier requires a barrier after storing to
	 * rq->curr, before returning to userspace, so provide them here:
	 *
	 * - a full memory barrier for {PRIVATE,GLOBAL}_EXPEDITED, implicitly
	 *   provided by mmdrop(),
	 * - a sync_core for SYNC_CORE.
2717
	 */
2718 2719
	if (mm) {
		membarrier_mm_sync_core_before_usermode(mm);
L
Linus Torvalds 已提交
2720
		mmdrop(mm);
2721
	}
2722
	if (unlikely(prev_state == TASK_DEAD)) {
2723 2724 2725
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2726 2727 2728
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2729
		 */
2730
		kprobe_flush_task(prev);
2731 2732 2733 2734

		/* Task is done with its stack. */
		put_task_stack(prev);

L
Linus Torvalds 已提交
2735
		put_task_struct(prev);
2736
	}
2737

2738
	tick_nohz_task_switch();
2739
	return rq;
L
Linus Torvalds 已提交
2740 2741
}

2742 2743 2744
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2745
static void __balance_callback(struct rq *rq)
2746
{
2747 2748 2749
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2750

2751 2752 2753 2754 2755 2756 2757 2758
	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;
2759

2760
		func(rq);
2761
	}
2762 2763 2764 2765 2766 2767 2768
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2769 2770 2771
}

#else
2772

2773
static inline void balance_callback(struct rq *rq)
2774
{
L
Linus Torvalds 已提交
2775 2776
}

2777 2778
#endif

L
Linus Torvalds 已提交
2779 2780 2781 2782
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2783
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2784 2785
	__releases(rq->lock)
{
2786
	struct rq *rq;
2787

2788 2789 2790 2791 2792 2793 2794 2795 2796
	/*
	 * 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).
	 */

2797
	rq = finish_task_switch(prev);
2798
	balance_callback(rq);
2799
	preempt_enable();
2800

L
Linus Torvalds 已提交
2801
	if (current->set_child_tid)
2802
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2803 2804 2805
}

/*
2806
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2807
 */
2808
static __always_inline struct rq *
2809
context_switch(struct rq *rq, struct task_struct *prev,
2810
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2811
{
I
Ingo Molnar 已提交
2812
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2813

2814
	prepare_task_switch(rq, prev, next);
2815

I
Ingo Molnar 已提交
2816 2817
	mm = next->mm;
	oldmm = prev->active_mm;
2818 2819 2820 2821 2822
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2823
	arch_start_context_switch(prev);
2824

2825 2826 2827 2828 2829 2830 2831
	/*
	 * If mm is non-NULL, we pass through switch_mm(). If mm is
	 * NULL, we will pass through mmdrop() in finish_task_switch().
	 * Both of these contain the full memory barrier required by
	 * membarrier after storing to rq->curr, before returning to
	 * user-space.
	 */
2832
	if (!mm) {
L
Linus Torvalds 已提交
2833
		next->active_mm = oldmm;
V
Vegard Nossum 已提交
2834
		mmgrab(oldmm);
L
Linus Torvalds 已提交
2835 2836
		enter_lazy_tlb(oldmm, next);
	} else
2837
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2838

2839
	if (!prev->mm) {
L
Linus Torvalds 已提交
2840 2841 2842
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2843

2844
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2845

2846 2847 2848 2849 2850 2851
	/*
	 * 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:
	 */
2852
	rq_unpin_lock(rq, rf);
2853
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2854 2855 2856

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2857
	barrier();
2858 2859

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2860 2861 2862
}

/*
2863
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2864 2865
 *
 * externally visible scheduler statistics: current number of runnable
2866
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2867 2868 2869 2870 2871 2872 2873 2874 2875
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2876
}
L
Linus Torvalds 已提交
2877

2878
/*
I
Ingo Molnar 已提交
2879
 * Check if only the current task is running on the CPU.
2880 2881 2882 2883 2884 2885 2886 2887 2888 2889
 *
 * 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)
2890 2891 2892
 */
bool single_task_running(void)
{
2893
	return raw_rq()->nr_running == 1;
2894 2895 2896
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2897
unsigned long long nr_context_switches(void)
2898
{
2899 2900
	int i;
	unsigned long long sum = 0;
2901

2902
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2903
		sum += cpu_rq(i)->nr_switches;
2904

L
Linus Torvalds 已提交
2905 2906
	return sum;
}
2907

2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937
/*
 * IO-wait accounting, and how its mostly bollocks (on SMP).
 *
 * The idea behind IO-wait account is to account the idle time that we could
 * have spend running if it were not for IO. That is, if we were to improve the
 * storage performance, we'd have a proportional reduction in IO-wait time.
 *
 * This all works nicely on UP, where, when a task blocks on IO, we account
 * idle time as IO-wait, because if the storage were faster, it could've been
 * running and we'd not be idle.
 *
 * This has been extended to SMP, by doing the same for each CPU. This however
 * is broken.
 *
 * Imagine for instance the case where two tasks block on one CPU, only the one
 * CPU will have IO-wait accounted, while the other has regular idle. Even
 * though, if the storage were faster, both could've ran at the same time,
 * utilising both CPUs.
 *
 * This means, that when looking globally, the current IO-wait accounting on
 * SMP is a lower bound, by reason of under accounting.
 *
 * Worse, since the numbers are provided per CPU, they are sometimes
 * interpreted per CPU, and that is nonsensical. A blocked task isn't strictly
 * associated with any one particular CPU, it can wake to another CPU than it
 * blocked on. This means the per CPU IO-wait number is meaningless.
 *
 * Task CPU affinities can make all that even more 'interesting'.
 */

L
Linus Torvalds 已提交
2938 2939 2940
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2941

2942
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2943
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2944

L
Linus Torvalds 已提交
2945 2946
	return sum;
}
2947

2948 2949 2950 2951 2952 2953 2954
/*
 * Consumers of these two interfaces, like for example the cpufreq menu
 * governor are using nonsensical data. Boosting frequency for a CPU that has
 * IO-wait which might not even end up running the task when it does become
 * runnable.
 */

2955
unsigned long nr_iowait_cpu(int cpu)
2956
{
2957
	struct rq *this = cpu_rq(cpu);
2958 2959
	return atomic_read(&this->nr_iowait);
}
2960

2961 2962
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2963 2964 2965
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2966 2967
}

I
Ingo Molnar 已提交
2968
#ifdef CONFIG_SMP
2969

2970
/*
P
Peter Zijlstra 已提交
2971 2972
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2973
 */
P
Peter Zijlstra 已提交
2974
void sched_exec(void)
2975
{
P
Peter Zijlstra 已提交
2976
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2977
	unsigned long flags;
2978
	int dest_cpu;
2979

2980
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2981
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2982 2983
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2984

2985
	if (likely(cpu_active(dest_cpu))) {
2986
		struct migration_arg arg = { p, dest_cpu };
2987

2988 2989
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2990 2991
		return;
	}
2992
unlock:
2993
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2994
}
I
Ingo Molnar 已提交
2995

L
Linus Torvalds 已提交
2996 2997 2998
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2999
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
3000 3001

EXPORT_PER_CPU_SYMBOL(kstat);
3002
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
3003

3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020
/*
 * 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);
}

3021 3022 3023 3024 3025 3026 3027
/*
 * 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)
{
3028
	struct rq_flags rf;
3029
	struct rq *rq;
3030
	u64 ns;
3031

3032 3033 3034 3035 3036 3037
#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.
	 *
I
Ingo Molnar 已提交
3038 3039
	 * 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
3040
	 * indistinguishable from the read occurring a few cycles earlier.
3041 3042
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3043
	 */
3044
	if (!p->on_cpu || !task_on_rq_queued(p))
3045 3046 3047
		return p->se.sum_exec_runtime;
#endif

3048
	rq = task_rq_lock(p, &rf);
3049 3050 3051 3052 3053 3054
	/*
	 * 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)) {
3055
		prefetch_curr_exec_start(p);
3056 3057 3058 3059
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3060
	task_rq_unlock(rq, p, &rf);
3061 3062 3063

	return ns;
}
3064

3065 3066 3067 3068 3069 3070 3071 3072
/*
 * 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
3073
	struct task_struct *curr = rq->curr;
3074
	struct rq_flags rf;
3075 3076

	sched_clock_tick();
I
Ingo Molnar 已提交
3077

3078 3079
	rq_lock(rq, &rf);

3080
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3081
	curr->sched_class->task_tick(rq, curr, 0);
3082
	cpu_load_update_active(rq);
3083
	calc_global_load_tick(rq);
3084 3085

	rq_unlock(rq, &rf);
3086

3087
	perf_event_task_tick();
3088

3089
#ifdef CONFIG_SMP
3090
	rq->idle_balance = idle_cpu(cpu);
3091
	trigger_load_balance(rq);
3092
#endif
3093
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
3094 3095
}

3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106
#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.
3107 3108
 *
 * Return: Maximum deferment in nanoseconds.
3109 3110 3111 3112
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
3113
	unsigned long next, now = READ_ONCE(jiffies);
3114 3115 3116 3117 3118 3119

	next = rq->last_sched_tick + HZ;

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

3120
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
3121
}
3122
#endif
L
Linus Torvalds 已提交
3123

3124 3125
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139
/*
 * 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);
	}
}
3140

3141
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3142
{
3143
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3144 3145 3146
	/*
	 * Underflow?
	 */
3147 3148
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3149
#endif
3150
	__preempt_count_add(val);
3151
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3152 3153 3154
	/*
	 * Spinlock count overflowing soon?
	 */
3155 3156
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3157
#endif
3158
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3159
}
3160
EXPORT_SYMBOL(preempt_count_add);
3161
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3162

3163 3164 3165 3166 3167 3168 3169 3170 3171 3172
/*
 * 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());
}

3173
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3174
{
3175
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3176 3177 3178
	/*
	 * Underflow?
	 */
3179
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3180
		return;
L
Linus Torvalds 已提交
3181 3182 3183
	/*
	 * Is the spinlock portion underflowing?
	 */
3184 3185 3186
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3187
#endif
3188

3189
	preempt_latency_stop(val);
3190
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3191
}
3192
EXPORT_SYMBOL(preempt_count_sub);
3193
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3194

3195 3196 3197
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3198 3199
#endif

3200 3201 3202 3203 3204 3205 3206 3207 3208
static inline unsigned long get_preempt_disable_ip(struct task_struct *p)
{
#ifdef CONFIG_DEBUG_PREEMPT
	return p->preempt_disable_ip;
#else
	return 0;
#endif
}

L
Linus Torvalds 已提交
3209
/*
I
Ingo Molnar 已提交
3210
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3211
 */
I
Ingo Molnar 已提交
3212
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3213
{
3214 3215 3216
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3217 3218 3219
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3220 3221
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3222

I
Ingo Molnar 已提交
3223
	debug_show_held_locks(prev);
3224
	print_modules();
I
Ingo Molnar 已提交
3225 3226
	if (irqs_disabled())
		print_irqtrace_events(prev);
3227 3228
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3229
		pr_err("Preemption disabled at:");
3230
		print_ip_sym(preempt_disable_ip);
3231 3232
		pr_cont("\n");
	}
3233 3234 3235
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3236
	dump_stack();
3237
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3238
}
L
Linus Torvalds 已提交
3239

I
Ingo Molnar 已提交
3240 3241 3242 3243 3244
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3245
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3246 3247
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3248
#endif
3249

3250
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3251
		__schedule_bug(prev);
3252 3253
		preempt_count_set(PREEMPT_DISABLED);
	}
3254
	rcu_sleep_check();
I
Ingo Molnar 已提交
3255

L
Linus Torvalds 已提交
3256 3257
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3258
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3259 3260 3261 3262 3263 3264
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3265
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3266
{
3267
	const struct sched_class *class;
I
Ingo Molnar 已提交
3268
	struct task_struct *p;
L
Linus Torvalds 已提交
3269 3270

	/*
3271 3272 3273 3274
	 * Optimization: we know that if all tasks are in the fair class we can
	 * call that function directly, but only if the @prev task wasn't of a
	 * higher scheduling class, because otherwise those loose the
	 * opportunity to pull in more work from other CPUs.
L
Linus Torvalds 已提交
3275
	 */
3276 3277 3278 3279
	if (likely((prev->sched_class == &idle_sched_class ||
		    prev->sched_class == &fair_sched_class) &&
		   rq->nr_running == rq->cfs.h_nr_running)) {

3280
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3281 3282 3283
		if (unlikely(p == RETRY_TASK))
			goto again;

I
Ingo Molnar 已提交
3284
		/* Assumes fair_sched_class->next == idle_sched_class */
3285
		if (unlikely(!p))
3286
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3287 3288

		return p;
L
Linus Torvalds 已提交
3289 3290
	}

3291
again:
3292
	for_each_class(class) {
3293
		p = class->pick_next_task(rq, prev, rf);
3294 3295 3296
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3297
			return p;
3298
		}
I
Ingo Molnar 已提交
3299
	}
3300

I
Ingo Molnar 已提交
3301 3302
	/* The idle class should always have a runnable task: */
	BUG();
I
Ingo Molnar 已提交
3303
}
L
Linus Torvalds 已提交
3304

I
Ingo Molnar 已提交
3305
/*
3306
 * __schedule() is the main scheduler function.
3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340
 *
 * 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
3341
 *
3342
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3343
 */
3344
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3345 3346
{
	struct task_struct *prev, *next;
3347
	unsigned long *switch_count;
3348
	struct rq_flags rf;
I
Ingo Molnar 已提交
3349
	struct rq *rq;
3350
	int cpu;
I
Ingo Molnar 已提交
3351 3352 3353 3354 3355 3356

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3357

3358
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3359
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3360

3361
	local_irq_disable();
3362
	rcu_note_context_switch(preempt);
3363

3364 3365 3366 3367
	/*
	 * 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().
3368 3369 3370
	 *
	 * The membarrier system call requires a full memory barrier
	 * after coming from user-space, before storing to rq->curr.
3371
	 */
3372
	rq_lock(rq, &rf);
3373
	smp_mb__after_spinlock();
L
Linus Torvalds 已提交
3374

I
Ingo Molnar 已提交
3375 3376
	/* Promote REQ to ACT */
	rq->clock_update_flags <<= 1;
3377
	update_rq_clock(rq);
3378

3379
	switch_count = &prev->nivcsw;
3380
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3381
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3382
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3383
		} else {
3384
			deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3385 3386
			prev->on_rq = 0;

3387 3388 3389 3390 3391
			if (prev->in_iowait) {
				atomic_inc(&rq->nr_iowait);
				delayacct_blkio_start();
			}

T
Tejun Heo 已提交
3392
			/*
3393 3394 3395
			 * 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
3396 3397 3398 3399
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3400
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3401
				if (to_wakeup)
3402
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3403 3404
			}
		}
I
Ingo Molnar 已提交
3405
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3406 3407
	}

3408
	next = pick_next_task(rq, prev, &rf);
3409
	clear_tsk_need_resched(prev);
3410
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3411 3412 3413 3414

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
3415 3416 3417
		/*
		 * The membarrier system call requires each architecture
		 * to have a full memory barrier after updating
3418 3419 3420 3421 3422 3423 3424 3425 3426 3427
		 * rq->curr, before returning to user-space.
		 *
		 * Here are the schemes providing that barrier on the
		 * various architectures:
		 * - mm ? switch_mm() : mmdrop() for x86, s390, sparc, PowerPC.
		 *   switch_mm() rely on membarrier_arch_switch_mm() on PowerPC.
		 * - finish_lock_switch() for weakly-ordered
		 *   architectures where spin_unlock is a full barrier,
		 * - switch_to() for arm64 (weakly-ordered, spin_unlock
		 *   is a RELEASE barrier),
3428
		 */
L
Linus Torvalds 已提交
3429 3430
		++*switch_count;

3431
		trace_sched_switch(preempt, prev, next);
I
Ingo Molnar 已提交
3432 3433 3434

		/* Also unlocks the rq: */
		rq = context_switch(rq, prev, next, &rf);
3435
	} else {
3436
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3437
		rq_unlock_irq(rq, &rf);
3438
	}
L
Linus Torvalds 已提交
3439

3440
	balance_callback(rq);
L
Linus Torvalds 已提交
3441
}
3442

3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456
void __noreturn do_task_dead(void)
{
	/*
	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
	 * when the following two conditions become true.
	 *   - There is race condition of mmap_sem (It is acquired by
	 *     exit_mm()), and
	 *   - SMI occurs before setting TASK_RUNINNG.
	 *     (or hypervisor of virtual machine switches to other guest)
	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
	 *
	 * To avoid it, we have to wait for releasing tsk->pi_lock which
	 * is held by try_to_wake_up()
	 */
3457 3458
	raw_spin_lock_irq(&current->pi_lock);
	raw_spin_unlock_irq(&current->pi_lock);
3459

I
Ingo Molnar 已提交
3460
	/* Causes final put_task_struct in finish_task_switch(): */
3461
	__set_current_state(TASK_DEAD);
I
Ingo Molnar 已提交
3462 3463 3464 3465

	/* Tell freezer to ignore us: */
	current->flags |= PF_NOFREEZE;

3466 3467
	__schedule(false);
	BUG();
I
Ingo Molnar 已提交
3468 3469

	/* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3470
	for (;;)
I
Ingo Molnar 已提交
3471
		cpu_relax();
3472 3473
}

3474 3475
static inline void sched_submit_work(struct task_struct *tsk)
{
3476
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3477 3478 3479 3480 3481 3482 3483 3484 3485
		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);
}

3486
asmlinkage __visible void __sched schedule(void)
3487
{
3488 3489 3490
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3491
	do {
3492
		preempt_disable();
3493
		__schedule(false);
3494
		sched_preempt_enable_no_resched();
3495
	} while (need_resched());
3496
}
L
Linus Torvalds 已提交
3497 3498
EXPORT_SYMBOL(schedule);

3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523
/*
 * synchronize_rcu_tasks() makes sure that no task is stuck in preempted
 * state (have scheduled out non-voluntarily) by making sure that all
 * tasks have either left the run queue or have gone into user space.
 * As idle tasks do not do either, they must not ever be preempted
 * (schedule out non-voluntarily).
 *
 * schedule_idle() is similar to schedule_preempt_disable() except that it
 * never enables preemption because it does not call sched_submit_work().
 */
void __sched schedule_idle(void)
{
	/*
	 * As this skips calling sched_submit_work(), which the idle task does
	 * regardless because that function is a nop when the task is in a
	 * TASK_RUNNING state, make sure this isn't used someplace that the
	 * current task can be in any other state. Note, idle is always in the
	 * TASK_RUNNING state.
	 */
	WARN_ON_ONCE(current->state);
	do {
		__schedule(false);
	} while (need_resched());
}

3524
#ifdef CONFIG_CONTEXT_TRACKING
3525
asmlinkage __visible void __sched schedule_user(void)
3526 3527 3528 3529 3530 3531
{
	/*
	 * 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.
3532 3533
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3534
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3535
	 * too frequently to make sense yet.
3536
	 */
3537
	enum ctx_state prev_state = exception_enter();
3538
	schedule();
3539
	exception_exit(prev_state);
3540 3541 3542
}
#endif

3543 3544 3545 3546 3547 3548 3549
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3550
	sched_preempt_enable_no_resched();
3551 3552 3553 3554
	schedule();
	preempt_disable();
}

3555
static void __sched notrace preempt_schedule_common(void)
3556 3557
{
	do {
3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570
		/*
		 * 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.
		 */
3571
		preempt_disable_notrace();
3572
		preempt_latency_start(1);
3573
		__schedule(true);
3574
		preempt_latency_stop(1);
3575
		preempt_enable_no_resched_notrace();
3576 3577 3578 3579 3580 3581 3582 3583

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

L
Linus Torvalds 已提交
3584 3585
#ifdef CONFIG_PREEMPT
/*
3586
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3587
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3588 3589
 * occur there and call schedule directly.
 */
3590
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3591 3592 3593
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3594
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3595
	 */
3596
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3597 3598
		return;

3599
	preempt_schedule_common();
L
Linus Torvalds 已提交
3600
}
3601
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3602
EXPORT_SYMBOL(preempt_schedule);
3603 3604

/**
3605
 * preempt_schedule_notrace - preempt_schedule called by tracing
3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617
 *
 * 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.
 */
3618
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3619 3620 3621 3622 3623 3624 3625
{
	enum ctx_state prev_ctx;

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

	do {
3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638
		/*
		 * 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.
		 */
3639
		preempt_disable_notrace();
3640
		preempt_latency_start(1);
3641 3642 3643 3644 3645 3646
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3647
		__schedule(true);
3648 3649
		exception_exit(prev_ctx);

3650
		preempt_latency_stop(1);
3651
		preempt_enable_no_resched_notrace();
3652 3653
	} while (need_resched());
}
3654
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3655

3656
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3657 3658

/*
3659
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3660 3661 3662 3663
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3664
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3665
{
3666
	enum ctx_state prev_state;
3667

3668
	/* Catch callers which need to be fixed */
3669
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3670

3671 3672
	prev_state = exception_enter();

3673
	do {
3674
		preempt_disable();
3675
		local_irq_enable();
3676
		__schedule(true);
3677
		local_irq_disable();
3678
		sched_preempt_enable_no_resched();
3679
	} while (need_resched());
3680 3681

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3682 3683
}

3684
int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3685
			  void *key)
L
Linus Torvalds 已提交
3686
{
P
Peter Zijlstra 已提交
3687
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3688 3689 3690
}
EXPORT_SYMBOL(default_wake_function);

3691 3692
#ifdef CONFIG_RT_MUTEXES

3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707
static inline int __rt_effective_prio(struct task_struct *pi_task, int prio)
{
	if (pi_task)
		prio = min(prio, pi_task->prio);

	return prio;
}

static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	struct task_struct *pi_task = rt_mutex_get_top_task(p);

	return __rt_effective_prio(pi_task, prio);
}

3708 3709
/*
 * rt_mutex_setprio - set the current priority of a task
3710 3711
 * @p: task to boost
 * @pi_task: donor task
3712 3713 3714 3715
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3716 3717
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3718
 */
3719
void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3720
{
3721
	int prio, oldprio, queued, running, queue_flag =
3722
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3723
	const struct sched_class *prev_class;
3724 3725
	struct rq_flags rf;
	struct rq *rq;
3726

3727 3728 3729 3730 3731 3732 3733 3734
	/* XXX used to be waiter->prio, not waiter->task->prio */
	prio = __rt_effective_prio(pi_task, p->normal_prio);

	/*
	 * If nothing changed; bail early.
	 */
	if (p->pi_top_task == pi_task && prio == p->prio && !dl_prio(prio))
		return;
3735

3736
	rq = __task_rq_lock(p, &rf);
3737
	update_rq_clock(rq);
3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754
	/*
	 * Set under pi_lock && rq->lock, such that the value can be used under
	 * either lock.
	 *
	 * Note that there is loads of tricky to make this pointer cache work
	 * right. rt_mutex_slowunlock()+rt_mutex_postunlock() work together to
	 * ensure a task is de-boosted (pi_task is set to NULL) before the
	 * task is allowed to run again (and can exit). This ensures the pointer
	 * points to a blocked task -- which guaratees the task is present.
	 */
	p->pi_top_task = pi_task;

	/*
	 * For FIFO/RR we only need to set prio, if that matches we're done.
	 */
	if (prio == p->prio && !dl_prio(prio))
		goto out_unlock;
3755

3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773
	/*
	 * 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;
	}

3774
	trace_sched_pi_setprio(p, pi_task);
3775
	oldprio = p->prio;
3776 3777 3778 3779

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

3780
	prev_class = p->sched_class;
3781
	queued = task_on_rq_queued(p);
3782
	running = task_current(rq, p);
3783
	if (queued)
3784
		dequeue_task(rq, p, queue_flag);
3785
	if (running)
3786
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3787

3788 3789 3790 3791 3792 3793 3794 3795 3796 3797
	/*
	 * 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)) {
3798 3799
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3800
			p->dl.dl_boosted = 1;
3801
			queue_flag |= ENQUEUE_REPLENISH;
3802 3803
		} else
			p->dl.dl_boosted = 0;
3804
		p->sched_class = &dl_sched_class;
3805 3806 3807 3808
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3809
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3810
		p->sched_class = &rt_sched_class;
3811 3812 3813
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3814 3815
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3816
		p->sched_class = &fair_sched_class;
3817
	}
I
Ingo Molnar 已提交
3818

3819 3820
	p->prio = prio;

3821
	if (queued)
3822
		enqueue_task(rq, p, queue_flag);
3823
	if (running)
3824
		set_curr_task(rq, p);
3825

P
Peter Zijlstra 已提交
3826
	check_class_changed(rq, p, prev_class, oldprio);
3827
out_unlock:
I
Ingo Molnar 已提交
3828 3829
	/* Avoid rq from going away on us: */
	preempt_disable();
3830
	__task_rq_unlock(rq, &rf);
3831 3832 3833

	balance_callback(rq);
	preempt_enable();
3834
}
3835 3836 3837 3838 3839
#else
static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	return prio;
}
3840
#endif
3841

3842
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3843
{
P
Peter Zijlstra 已提交
3844 3845
	bool queued, running;
	int old_prio, delta;
3846
	struct rq_flags rf;
3847
	struct rq *rq;
L
Linus Torvalds 已提交
3848

3849
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3850 3851 3852 3853 3854
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3855
	rq = task_rq_lock(p, &rf);
3856 3857
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3858 3859 3860 3861
	/*
	 * 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
3862
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3863
	 */
3864
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3865 3866 3867
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3868
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3869
	running = task_current(rq, p);
3870
	if (queued)
3871
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
3872 3873
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3874 3875

	p->static_prio = NICE_TO_PRIO(nice);
3876
	set_load_weight(p, true);
3877 3878 3879
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3880

3881
	if (queued) {
3882
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
L
Linus Torvalds 已提交
3883
		/*
3884 3885
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3886
		 */
3887
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3888
			resched_curr(rq);
L
Linus Torvalds 已提交
3889
	}
P
Peter Zijlstra 已提交
3890 3891
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3892
out_unlock:
3893
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3894 3895 3896
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3897 3898 3899 3900 3901
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3902
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3903
{
I
Ingo Molnar 已提交
3904
	/* Convert nice value [19,-20] to rlimit style value [1,40]: */
3905
	int nice_rlim = nice_to_rlimit(nice);
3906

3907
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3908 3909 3910
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3911 3912 3913 3914 3915 3916 3917 3918 3919
#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.
 */
3920
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3921
{
3922
	long nice, retval;
L
Linus Torvalds 已提交
3923 3924 3925 3926 3927 3928

	/*
	 * 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.
	 */
3929
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3930
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3931

3932
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3933 3934 3935
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949
	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.
 *
3950
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3951 3952 3953
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3954
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3955 3956 3957 3958 3959
{
	return p->prio - MAX_RT_PRIO;
}

/**
I
Ingo Molnar 已提交
3960
 * idle_cpu - is a given CPU idle currently?
L
Linus Torvalds 已提交
3961
 * @cpu: the processor in question.
3962 3963
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3964 3965 3966
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980
	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
3981 3982 3983
}

/**
I
Ingo Molnar 已提交
3984
 * idle_task - return the idle task for a given CPU.
L
Linus Torvalds 已提交
3985
 * @cpu: the processor in question.
3986
 *
I
Ingo Molnar 已提交
3987
 * Return: The idle task for the CPU @cpu.
L
Linus Torvalds 已提交
3988
 */
3989
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3990 3991 3992 3993 3994 3995 3996
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3997 3998
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3999
 */
A
Alexey Dobriyan 已提交
4000
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
4001
{
4002
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
4003 4004
}

4005 4006 4007 4008 4009 4010
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

4011 4012
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
4013
{
4014 4015
	int policy = attr->sched_policy;

4016
	if (policy == SETPARAM_POLICY)
4017 4018
		policy = p->policy;

L
Linus Torvalds 已提交
4019
	p->policy = policy;
4020

4021 4022
	if (dl_policy(policy))
		__setparam_dl(p, attr);
4023
	else if (fair_policy(policy))
4024 4025
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

4026 4027 4028 4029 4030 4031
	/*
	 * __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;
4032
	p->normal_prio = normal_prio(p);
4033
	set_load_weight(p, true);
4034
}
4035

4036 4037
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
4038
			   const struct sched_attr *attr, bool keep_boost)
4039 4040
{
	__setscheduler_params(p, attr);
4041

4042
	/*
4043 4044
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
4045
	 */
4046
	p->prio = normal_prio(p);
4047
	if (keep_boost)
4048
		p->prio = rt_effective_prio(p, p->prio);
4049

4050 4051 4052
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
4053 4054 4055
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
4056
}
4057

4058
/*
I
Ingo Molnar 已提交
4059
 * Check the target process has a UID that matches the current process's:
4060 4061 4062 4063 4064 4065 4066 4067
 */
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);
4068 4069
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4070 4071 4072 4073
	rcu_read_unlock();
	return match;
}

4074 4075
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4076
				bool user, bool pi)
L
Linus Torvalds 已提交
4077
{
4078 4079
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4080
	int retval, oldprio, oldpolicy = -1, queued, running;
4081
	int new_effective_prio, policy = attr->sched_policy;
4082
	const struct sched_class *prev_class;
4083
	struct rq_flags rf;
4084
	int reset_on_fork;
4085
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4086
	struct rq *rq;
L
Linus Torvalds 已提交
4087

4088 4089
	/* The pi code expects interrupts enabled */
	BUG_ON(pi && in_interrupt());
L
Linus Torvalds 已提交
4090
recheck:
I
Ingo Molnar 已提交
4091
	/* Double check policy once rq lock held: */
4092 4093
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4094
		policy = oldpolicy = p->policy;
4095
	} else {
4096
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4097

4098
		if (!valid_policy(policy))
4099 4100 4101
			return -EINVAL;
	}

4102
	if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4103 4104
		return -EINVAL;

L
Linus Torvalds 已提交
4105 4106
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4107 4108
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4109
	 */
4110
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4111
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4112
		return -EINVAL;
4113 4114
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4115 4116
		return -EINVAL;

4117 4118 4119
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4120
	if (user && !capable(CAP_SYS_NICE)) {
4121
		if (fair_policy(policy)) {
4122
			if (attr->sched_nice < task_nice(p) &&
4123
			    !can_nice(p, attr->sched_nice))
4124 4125 4126
				return -EPERM;
		}

4127
		if (rt_policy(policy)) {
4128 4129
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4130

I
Ingo Molnar 已提交
4131
			/* Can't set/change the rt policy: */
4132 4133 4134
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

I
Ingo Molnar 已提交
4135
			/* Can't increase priority: */
4136 4137
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4138 4139
				return -EPERM;
		}
4140

4141 4142 4143 4144 4145 4146 4147 4148 4149
		 /*
		  * 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
4150
		/*
4151 4152
		 * 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
4153
		 */
4154
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4155
			if (!can_nice(p, task_nice(p)))
4156 4157
				return -EPERM;
		}
4158

I
Ingo Molnar 已提交
4159
		/* Can't change other user's priorities: */
4160
		if (!check_same_owner(p))
4161
			return -EPERM;
4162

I
Ingo Molnar 已提交
4163
		/* Normal users shall not reset the sched_reset_on_fork flag: */
4164 4165
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4166
	}
L
Linus Torvalds 已提交
4167

4168
	if (user) {
4169 4170 4171
		if (attr->sched_flags & SCHED_FLAG_SUGOV)
			return -EINVAL;

4172
		retval = security_task_setscheduler(p);
4173 4174 4175 4176
		if (retval)
			return retval;
	}

4177
	/*
I
Ingo Molnar 已提交
4178
	 * Make sure no PI-waiters arrive (or leave) while we are
4179
	 * changing the priority of the task:
4180
	 *
L
Lucas De Marchi 已提交
4181
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4182 4183
	 * runqueue lock must be held.
	 */
4184
	rq = task_rq_lock(p, &rf);
4185
	update_rq_clock(rq);
4186

4187
	/*
I
Ingo Molnar 已提交
4188
	 * Changing the policy of the stop threads its a very bad idea:
4189 4190
	 */
	if (p == rq->stop) {
4191
		task_rq_unlock(rq, p, &rf);
4192 4193 4194
		return -EINVAL;
	}

4195
	/*
4196 4197
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4198
	 */
4199
	if (unlikely(policy == p->policy)) {
4200
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4201 4202 4203
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4204
		if (dl_policy(policy) && dl_param_changed(p, attr))
4205
			goto change;
4206

4207
		p->sched_reset_on_fork = reset_on_fork;
4208
		task_rq_unlock(rq, p, &rf);
4209 4210
		return 0;
	}
4211
change:
4212

4213
	if (user) {
4214
#ifdef CONFIG_RT_GROUP_SCHED
4215 4216 4217 4218 4219
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4220 4221
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4222
			task_rq_unlock(rq, p, &rf);
4223 4224 4225
			return -EPERM;
		}
#endif
4226
#ifdef CONFIG_SMP
4227 4228
		if (dl_bandwidth_enabled() && dl_policy(policy) &&
				!(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4229 4230 4231 4232 4233 4234 4235
			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.
			 */
4236 4237
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4238
				task_rq_unlock(rq, p, &rf);
4239 4240 4241 4242 4243
				return -EPERM;
			}
		}
#endif
	}
4244

I
Ingo Molnar 已提交
4245
	/* Re-check policy now with rq lock held: */
L
Linus Torvalds 已提交
4246 4247
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4248
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4249 4250
		goto recheck;
	}
4251 4252 4253 4254 4255 4256

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4257
	if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4258
		task_rq_unlock(rq, p, &rf);
4259 4260 4261
		return -EBUSY;
	}

4262 4263 4264
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4265 4266 4267 4268 4269 4270 4271 4272
	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.
		 */
4273
		new_effective_prio = rt_effective_prio(p, newprio);
4274 4275
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4276 4277
	}

4278
	queued = task_on_rq_queued(p);
4279
	running = task_current(rq, p);
4280
	if (queued)
4281
		dequeue_task(rq, p, queue_flags);
4282
	if (running)
4283
		put_prev_task(rq, p);
4284

4285
	prev_class = p->sched_class;
4286
	__setscheduler(rq, p, attr, pi);
4287

4288
	if (queued) {
4289 4290 4291 4292
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4293 4294
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4295

4296
		enqueue_task(rq, p, queue_flags);
4297
	}
4298
	if (running)
4299
		set_curr_task(rq, p);
4300

P
Peter Zijlstra 已提交
4301
	check_class_changed(rq, p, prev_class, oldprio);
I
Ingo Molnar 已提交
4302 4303 4304

	/* Avoid rq from going away on us: */
	preempt_disable();
4305
	task_rq_unlock(rq, p, &rf);
4306

4307 4308
	if (pi)
		rt_mutex_adjust_pi(p);
4309

I
Ingo Molnar 已提交
4310
	/* Run balance callbacks after we've adjusted the PI chain: */
4311 4312
	balance_callback(rq);
	preempt_enable();
4313

L
Linus Torvalds 已提交
4314 4315
	return 0;
}
4316

4317 4318 4319 4320 4321 4322 4323 4324 4325
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),
	};

4326 4327
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4328 4329 4330 4331 4332
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4333
	return __sched_setscheduler(p, &attr, check, true);
4334
}
4335 4336 4337 4338 4339 4340
/**
 * 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.
 *
4341 4342
 * Return: 0 on success. An error code otherwise.
 *
4343 4344 4345
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4346
		       const struct sched_param *param)
4347
{
4348
	return _sched_setscheduler(p, policy, param, true);
4349
}
L
Linus Torvalds 已提交
4350 4351
EXPORT_SYMBOL_GPL(sched_setscheduler);

4352 4353
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4354
	return __sched_setscheduler(p, attr, true, true);
4355 4356 4357
}
EXPORT_SYMBOL_GPL(sched_setattr);

4358 4359 4360 4361 4362
int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
{
	return __sched_setscheduler(p, attr, false, true);
}

4363 4364 4365 4366 4367 4368 4369 4370 4371 4372
/**
 * 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.
4373 4374
 *
 * Return: 0 on success. An error code otherwise.
4375 4376
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4377
			       const struct sched_param *param)
4378
{
4379
	return _sched_setscheduler(p, policy, param, false);
4380
}
4381
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4382

I
Ingo Molnar 已提交
4383 4384
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4385 4386 4387
{
	struct sched_param lparam;
	struct task_struct *p;
4388
	int retval;
L
Linus Torvalds 已提交
4389 4390 4391 4392 4393

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4394 4395 4396

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4397
	p = find_process_by_pid(pid);
4398 4399 4400
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4401

L
Linus Torvalds 已提交
4402 4403 4404
	return retval;
}

4405 4406 4407
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
I
Ingo Molnar 已提交
4408
static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4409 4410 4411 4412 4413 4414 4415
{
	u32 size;
	int ret;

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

I
Ingo Molnar 已提交
4416
	/* Zero the full structure, so that a short copy will be nice: */
4417 4418 4419 4420 4421 4422
	memset(attr, 0, sizeof(*attr));

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

I
Ingo Molnar 已提交
4423 4424
	/* Bail out on silly large: */
	if (size > PAGE_SIZE)
4425 4426
		goto err_size;

I
Ingo Molnar 已提交
4427 4428
	/* ABI compatibility quirk: */
	if (!size)
4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462
		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;

	/*
I
Ingo Molnar 已提交
4463
	 * XXX: Do we want to be lenient like existing syscalls; or do we want
4464 4465
	 * to be strict and return an error on out-of-bounds values?
	 */
4466
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4467

4468
	return 0;
4469 4470 4471

err_size:
	put_user(sizeof(*attr), &uattr->size);
4472
	return -E2BIG;
4473 4474
}

L
Linus Torvalds 已提交
4475 4476 4477 4478 4479
/**
 * 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.
4480 4481
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4482
 */
I
Ingo Molnar 已提交
4483
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4484
{
4485 4486 4487
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4488 4489 4490 4491 4492 4493 4494
	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.
4495 4496
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4497
 */
4498
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4499
{
4500
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4501 4502
}

4503 4504 4505
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4506
 * @uattr: structure containing the extended parameters.
4507
 * @flags: for future extension.
4508
 */
4509 4510
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4511 4512 4513 4514 4515
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4516
	if (!uattr || pid < 0 || flags)
4517 4518
		return -EINVAL;

4519 4520 4521
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4522

4523
	if ((int)attr.sched_policy < 0)
4524
		return -EINVAL;
4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535

	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
4536 4537 4538
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4539 4540 4541
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4542
 */
4543
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4544
{
4545
	struct task_struct *p;
4546
	int retval;
L
Linus Torvalds 已提交
4547 4548

	if (pid < 0)
4549
		return -EINVAL;
L
Linus Torvalds 已提交
4550 4551

	retval = -ESRCH;
4552
	rcu_read_lock();
L
Linus Torvalds 已提交
4553 4554 4555 4556
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4557 4558
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4559
	}
4560
	rcu_read_unlock();
L
Linus Torvalds 已提交
4561 4562 4563 4564
	return retval;
}

/**
4565
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4566 4567
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4568 4569 4570
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4571
 */
4572
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4573
{
4574
	struct sched_param lp = { .sched_priority = 0 };
4575
	struct task_struct *p;
4576
	int retval;
L
Linus Torvalds 已提交
4577 4578

	if (!param || pid < 0)
4579
		return -EINVAL;
L
Linus Torvalds 已提交
4580

4581
	rcu_read_lock();
L
Linus Torvalds 已提交
4582 4583 4584 4585 4586 4587 4588 4589 4590
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4591 4592
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4593
	rcu_read_unlock();
L
Linus Torvalds 已提交
4594 4595 4596 4597 4598 4599 4600 4601 4602

	/*
	 * 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:
4603
	rcu_read_unlock();
L
Linus Torvalds 已提交
4604 4605 4606
	return retval;
}

4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629
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)
4630
				return -EFBIG;
4631 4632 4633 4634 4635
		}

		attr->size = usize;
	}

4636
	ret = copy_to_user(uattr, attr, attr->size);
4637 4638 4639
	if (ret)
		return -EFAULT;

4640
	return 0;
4641 4642 4643
}

/**
4644
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4645
 * @pid: the pid in question.
J
Juri Lelli 已提交
4646
 * @uattr: structure containing the extended parameters.
4647
 * @size: sizeof(attr) for fwd/bwd comp.
4648
 * @flags: for future extension.
4649
 */
4650 4651
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4652 4653 4654 4655 4656 4657 4658 4659
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4660
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673
		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;
4674 4675
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4676 4677 4678
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4679 4680
		attr.sched_priority = p->rt_priority;
	else
4681
		attr.sched_nice = task_nice(p);
4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4693
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4694
{
4695
	cpumask_var_t cpus_allowed, new_mask;
4696 4697
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4698

4699
	rcu_read_lock();
L
Linus Torvalds 已提交
4700 4701 4702

	p = find_process_by_pid(pid);
	if (!p) {
4703
		rcu_read_unlock();
L
Linus Torvalds 已提交
4704 4705 4706
		return -ESRCH;
	}

4707
	/* Prevent p going away */
L
Linus Torvalds 已提交
4708
	get_task_struct(p);
4709
	rcu_read_unlock();
L
Linus Torvalds 已提交
4710

4711 4712 4713 4714
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4715 4716 4717 4718 4719 4720 4721 4722
	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
4723
	retval = -EPERM;
E
Eric W. Biederman 已提交
4724 4725 4726 4727
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4728
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4729 4730 4731
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4732

4733
	retval = security_task_setscheduler(p);
4734
	if (retval)
4735
		goto out_free_new_mask;
4736

4737 4738 4739 4740

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

4741 4742 4743 4744 4745 4746 4747
	/*
	 * 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
4748 4749 4750
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4751
			retval = -EBUSY;
4752
			rcu_read_unlock();
4753
			goto out_free_new_mask;
4754
		}
4755
		rcu_read_unlock();
4756 4757
	}
#endif
P
Peter Zijlstra 已提交
4758
again:
4759
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4760

P
Paul Menage 已提交
4761
	if (!retval) {
4762 4763
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4764 4765 4766 4767 4768
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4769
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4770 4771 4772
			goto again;
		}
	}
4773
out_free_new_mask:
4774 4775 4776 4777
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4778 4779 4780 4781 4782
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4783
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4784
{
4785 4786 4787 4788 4789
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4790 4791 4792 4793
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
I
Ingo Molnar 已提交
4794
 * sys_sched_setaffinity - set the CPU affinity of a process
L
Linus Torvalds 已提交
4795 4796
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4797
 * @user_mask_ptr: user-space pointer to the new CPU mask
4798 4799
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4800
 */
4801 4802
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4803
{
4804
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4805 4806
	int retval;

4807 4808
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4809

4810 4811 4812 4813 4814
	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
4815 4816
}

4817
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4818
{
4819
	struct task_struct *p;
4820
	unsigned long flags;
L
Linus Torvalds 已提交
4821 4822
	int retval;

4823
	rcu_read_lock();
L
Linus Torvalds 已提交
4824 4825 4826 4827 4828 4829

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

4830 4831 4832 4833
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4834
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4835
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4836
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4837 4838

out_unlock:
4839
	rcu_read_unlock();
L
Linus Torvalds 已提交
4840

4841
	return retval;
L
Linus Torvalds 已提交
4842 4843 4844
}

/**
I
Ingo Molnar 已提交
4845
 * sys_sched_getaffinity - get the CPU affinity of a process
L
Linus Torvalds 已提交
4846 4847
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4848
 * @user_mask_ptr: user-space pointer to hold the current CPU mask
4849
 *
4850 4851
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4852
 */
4853 4854
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4855 4856
{
	int ret;
4857
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4858

A
Anton Blanchard 已提交
4859
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4860 4861
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4862 4863
		return -EINVAL;

4864 4865
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4866

4867 4868
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4869
		size_t retlen = min_t(size_t, len, cpumask_size());
4870 4871

		if (copy_to_user(user_mask_ptr, mask, retlen))
4872 4873
			ret = -EFAULT;
		else
4874
			ret = retlen;
4875 4876
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4877

4878
	return ret;
L
Linus Torvalds 已提交
4879 4880 4881 4882 4883
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4884 4885
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4886 4887
 *
 * Return: 0.
L
Linus Torvalds 已提交
4888
 */
4889
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4890
{
4891 4892 4893 4894 4895 4896
	struct rq_flags rf;
	struct rq *rq;

	local_irq_disable();
	rq = this_rq();
	rq_lock(rq, &rf);
L
Linus Torvalds 已提交
4897

4898
	schedstat_inc(rq->yld_count);
4899
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4900 4901 4902 4903 4904

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
4905 4906
	preempt_disable();
	rq_unlock(rq, &rf);
4907
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4908 4909 4910 4911 4912 4913

	schedule();

	return 0;
}

4914
#ifndef CONFIG_PREEMPT
4915
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4916
{
4917
	if (should_resched(0)) {
4918
		preempt_schedule_common();
L
Linus Torvalds 已提交
4919 4920
		return 1;
	}
4921
	rcu_all_qs();
L
Linus Torvalds 已提交
4922 4923
	return 0;
}
4924
EXPORT_SYMBOL(_cond_resched);
4925
#endif
L
Linus Torvalds 已提交
4926 4927

/*
4928
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4929 4930
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4931
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4932 4933 4934
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4935
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4936
{
4937
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4938 4939
	int ret = 0;

4940 4941
	lockdep_assert_held(lock);

4942
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4943
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4944
		if (resched)
4945
			preempt_schedule_common();
N
Nick Piggin 已提交
4946 4947
		else
			cpu_relax();
J
Jan Kara 已提交
4948
		ret = 1;
L
Linus Torvalds 已提交
4949 4950
		spin_lock(lock);
	}
J
Jan Kara 已提交
4951
	return ret;
L
Linus Torvalds 已提交
4952
}
4953
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4954

4955
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4956 4957 4958
{
	BUG_ON(!in_softirq());

4959
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4960
		local_bh_enable();
4961
		preempt_schedule_common();
L
Linus Torvalds 已提交
4962 4963 4964 4965 4966
		local_bh_disable();
		return 1;
	}
	return 0;
}
4967
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4968 4969 4970 4971

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4972 4973 4974 4975 4976 4977 4978 4979 4980
 * 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)
I
Ingo Molnar 已提交
4981
 *	yield();
P
Peter Zijlstra 已提交
4982 4983 4984 4985 4986 4987 4988 4989
 *
 * 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
4990 4991 4992 4993 4994 4995 4996 4997
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4998 4999 5000 5001
/**
 * 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
5002 5003
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5004 5005 5006 5007
 *
 * 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.
 *
5008
 * Return:
5009 5010 5011
 *	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.
5012
 */
5013
int __sched yield_to(struct task_struct *p, bool preempt)
5014 5015 5016 5017
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5018
	int yielded = 0;
5019 5020 5021 5022 5023 5024

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5025 5026 5027 5028 5029 5030 5031 5032 5033
	/*
	 * 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;
	}

5034
	double_rq_lock(rq, p_rq);
5035
	if (task_rq(p) != p_rq) {
5036 5037 5038 5039 5040
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5041
		goto out_unlock;
5042 5043

	if (curr->sched_class != p->sched_class)
5044
		goto out_unlock;
5045 5046

	if (task_running(p_rq, p) || p->state)
5047
		goto out_unlock;
5048 5049

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5050
	if (yielded) {
5051
		schedstat_inc(rq->yld_count);
5052 5053 5054 5055 5056
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5057
			resched_curr(p_rq);
5058
	}
5059

5060
out_unlock:
5061
	double_rq_unlock(rq, p_rq);
5062
out_irq:
5063 5064
	local_irq_restore(flags);

5065
	if (yielded > 0)
5066 5067 5068 5069 5070 5071
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086
int io_schedule_prepare(void)
{
	int old_iowait = current->in_iowait;

	current->in_iowait = 1;
	blk_schedule_flush_plug(current);

	return old_iowait;
}

void io_schedule_finish(int token)
{
	current->in_iowait = token;
}

L
Linus Torvalds 已提交
5087
/*
I
Ingo Molnar 已提交
5088
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5089 5090 5091 5092
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5093
	int token;
L
Linus Torvalds 已提交
5094 5095
	long ret;

5096
	token = io_schedule_prepare();
L
Linus Torvalds 已提交
5097
	ret = schedule_timeout(timeout);
5098
	io_schedule_finish(token);
5099

L
Linus Torvalds 已提交
5100 5101
	return ret;
}
5102
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5103

5104 5105 5106 5107 5108 5109 5110 5111 5112 5113
void io_schedule(void)
{
	int token;

	token = io_schedule_prepare();
	schedule();
	io_schedule_finish(token);
}
EXPORT_SYMBOL(io_schedule);

L
Linus Torvalds 已提交
5114 5115 5116 5117
/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5118 5119 5120
 * 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
5121
 */
5122
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5123 5124 5125 5126 5127 5128 5129 5130
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5131
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5132
	case SCHED_NORMAL:
5133
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5134
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5135 5136 5137 5138 5139 5140 5141 5142 5143 5144
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5145 5146 5147
 * 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
5148
 */
5149
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5150 5151 5152 5153 5154 5155 5156 5157
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5158
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5159
	case SCHED_NORMAL:
5160
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5161
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5162 5163 5164 5165 5166
		ret = 0;
	}
	return ret;
}

5167
static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
L
Linus Torvalds 已提交
5168
{
5169
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5170
	unsigned int time_slice;
5171
	struct rq_flags rf;
5172
	struct rq *rq;
5173
	int retval;
L
Linus Torvalds 已提交
5174 5175

	if (pid < 0)
5176
		return -EINVAL;
L
Linus Torvalds 已提交
5177 5178

	retval = -ESRCH;
5179
	rcu_read_lock();
L
Linus Torvalds 已提交
5180 5181 5182 5183 5184 5185 5186 5187
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5188
	rq = task_rq_lock(p, &rf);
5189 5190 5191
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5192
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5193

5194
	rcu_read_unlock();
5195 5196
	jiffies_to_timespec64(time_slice, t);
	return 0;
5197

L
Linus Torvalds 已提交
5198
out_unlock:
5199
	rcu_read_unlock();
L
Linus Torvalds 已提交
5200 5201 5202
	return retval;
}

5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213
/**
 * 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.
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
 */
5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
		struct timespec __user *, interval)
{
	struct timespec64 t;
	int retval = sched_rr_get_interval(pid, &t);

	if (retval == 0)
		retval = put_timespec64(&t, interval);

	return retval;
}

#ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE2(sched_rr_get_interval,
		       compat_pid_t, pid,
		       struct compat_timespec __user *, interval)
{
	struct timespec64 t;
	int retval = sched_rr_get_interval(pid, &t);

	if (retval == 0)
		retval = compat_put_timespec64(&t, interval);
	return retval;
}
#endif

5240
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5241 5242
{
	unsigned long free = 0;
5243
	int ppid;
5244

5245 5246
	if (!try_get_task_stack(p))
		return;
5247 5248 5249 5250

	printk(KERN_INFO "%-15.15s %c", p->comm, task_state_to_char(p));

	if (p->state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5251
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5252
#ifdef CONFIG_DEBUG_STACK_USAGE
5253
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5254
#endif
5255
	ppid = 0;
5256
	rcu_read_lock();
5257 5258
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5259
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5260
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5261
		task_pid_nr(p), ppid,
5262
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5263

5264
	print_worker_info(KERN_INFO, p);
5265
	show_stack(p, NULL);
5266
	put_task_stack(p);
L
Linus Torvalds 已提交
5267
}
5268
EXPORT_SYMBOL_GPL(sched_show_task);
L
Linus Torvalds 已提交
5269

5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291
static inline bool
state_filter_match(unsigned long state_filter, struct task_struct *p)
{
	/* no filter, everything matches */
	if (!state_filter)
		return true;

	/* filter, but doesn't match */
	if (!(p->state & state_filter))
		return false;

	/*
	 * When looking for TASK_UNINTERRUPTIBLE skip TASK_IDLE (allows
	 * TASK_KILLABLE).
	 */
	if (state_filter == TASK_UNINTERRUPTIBLE && p->state == TASK_IDLE)
		return false;

	return true;
}


I
Ingo Molnar 已提交
5292
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5293
{
5294
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5295

5296
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5297 5298
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5299
#else
P
Peter Zijlstra 已提交
5300 5301
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5302
#endif
5303
	rcu_read_lock();
5304
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5305 5306
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5307
		 * console might take a lot of time:
5308 5309 5310
		 * 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
5311 5312
		 */
		touch_nmi_watchdog();
5313
		touch_all_softlockup_watchdogs();
5314
		if (state_filter_match(state_filter, p))
5315
			sched_show_task(p);
5316
	}
L
Linus Torvalds 已提交
5317

I
Ingo Molnar 已提交
5318
#ifdef CONFIG_SCHED_DEBUG
5319 5320
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5321
#endif
5322
	rcu_read_unlock();
I
Ingo Molnar 已提交
5323 5324 5325
	/*
	 * Only show locks if all tasks are dumped:
	 */
5326
	if (!state_filter)
I
Ingo Molnar 已提交
5327
		debug_show_all_locks();
L
Linus Torvalds 已提交
5328 5329
}

5330 5331 5332
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
I
Ingo Molnar 已提交
5333
 * @cpu: CPU the idle task belongs to
5334 5335 5336 5337
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5338
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5339
{
5340
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5341 5342
	unsigned long flags;

5343 5344
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5345

5346
	__sched_fork(0, idle);
5347
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5348
	idle->se.exec_start = sched_clock();
5349
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5350

5351 5352
	kasan_unpoison_task_stack(idle);

5353 5354 5355 5356 5357 5358 5359 5360 5361
#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
5362 5363
	/*
	 * We're having a chicken and egg problem, even though we are
I
Ingo Molnar 已提交
5364
	 * holding rq->lock, the CPU isn't yet set to this CPU so the
5365 5366 5367 5368 5369 5370 5371 5372
	 * 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
5373
	__set_task_cpu(idle, cpu);
5374
	rcu_read_unlock();
L
Linus Torvalds 已提交
5375 5376

	rq->curr = rq->idle = idle;
5377
	idle->on_rq = TASK_ON_RQ_QUEUED;
5378
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5379
	idle->on_cpu = 1;
5380
#endif
5381 5382
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5383 5384

	/* Set the preempt count _outside_ the spinlocks! */
5385
	init_idle_preempt_count(idle, cpu);
5386

I
Ingo Molnar 已提交
5387 5388 5389 5390
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5391
	ftrace_graph_init_idle_task(idle, cpu);
5392
	vtime_init_idle(idle, cpu);
5393
#ifdef CONFIG_SMP
5394 5395
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5396 5397
}

5398 5399
#ifdef CONFIG_SMP

5400 5401 5402
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
5403
	int ret = 1;
5404

5405 5406 5407
	if (!cpumask_weight(cur))
		return ret;

5408
	ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5409 5410 5411 5412

	return ret;
}

5413 5414 5415 5416 5417 5418 5419
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
I
Ingo Molnar 已提交
5420
	 * to a new cpuset; we don't want to change their CPU
5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432
	 * 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;
	}

	if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span,
5433 5434
					      cs_cpus_allowed))
		ret = dl_task_can_attach(p, cs_cpus_allowed);
5435 5436 5437 5438 5439

out:
	return ret;
}

5440
bool sched_smp_initialized __read_mostly;
5441

5442 5443 5444 5445 5446 5447 5448 5449 5450 5451
#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;

5452
	if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5453 5454 5455 5456
		return -EINVAL;

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

5457
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5458 5459
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5460 5461 5462 5463 5464 5465 5466

/*
 * 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)
{
5467
	bool queued, running;
5468 5469
	struct rq_flags rf;
	struct rq *rq;
5470

5471
	rq = task_rq_lock(p, &rf);
5472
	queued = task_on_rq_queued(p);
5473 5474
	running = task_current(rq, p);

5475
	if (queued)
5476
		dequeue_task(rq, p, DEQUEUE_SAVE);
5477
	if (running)
5478
		put_prev_task(rq, p);
5479 5480 5481

	p->numa_preferred_nid = nid;

5482
	if (queued)
5483
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5484
	if (running)
5485
		set_curr_task(rq, p);
5486
	task_rq_unlock(rq, p, &rf);
5487
}
P
Peter Zijlstra 已提交
5488
#endif /* CONFIG_NUMA_BALANCING */
5489

L
Linus Torvalds 已提交
5490
#ifdef CONFIG_HOTPLUG_CPU
5491
/*
I
Ingo Molnar 已提交
5492
 * Ensure that the idle task is using init_mm right before its CPU goes
5493
 * offline.
5494
 */
5495
void idle_task_exit(void)
L
Linus Torvalds 已提交
5496
{
5497
	struct mm_struct *mm = current->active_mm;
5498

5499
	BUG_ON(cpu_online(smp_processor_id()));
5500

5501
	if (mm != &init_mm) {
5502
		switch_mm(mm, &init_mm, current);
5503 5504
		finish_arch_post_lock_switch();
	}
5505
	mmdrop(mm);
L
Linus Torvalds 已提交
5506 5507 5508
}

/*
5509 5510
 * 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
5511 5512 5513
 * 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.
5514 5515
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5516
 */
5517
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5518
{
5519
	long delta = calc_load_fold_active(rq, 1);
5520 5521
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5522 5523
}

5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539
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,
};

5540
/*
5541 5542 5543 5544 5545 5546
 * 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
5547
 */
5548
static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
L
Linus Torvalds 已提交
5549
{
5550
	struct rq *rq = dead_rq;
5551
	struct task_struct *next, *stop = rq->stop;
5552
	struct rq_flags orf = *rf;
5553
	int dest_cpu;
L
Linus Torvalds 已提交
5554 5555

	/*
5556 5557 5558 5559 5560 5561 5562
	 * 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
5563
	 */
5564
	rq->stop = NULL;
5565

5566 5567 5568 5569 5570 5571 5572
	/*
	 * 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);

5573
	for (;;) {
5574 5575
		/*
		 * There's this thread running, bail when that's the only
I
Ingo Molnar 已提交
5576
		 * remaining thread:
5577 5578
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5579
			break;
5580

5581
		/*
I
Ingo Molnar 已提交
5582
		 * pick_next_task() assumes pinned rq->lock:
5583
		 */
5584
		next = pick_next_task(rq, &fake_task, rf);
5585
		BUG_ON(!next);
V
Viresh Kumar 已提交
5586
		put_prev_task(rq, next);
5587

W
Wanpeng Li 已提交
5588 5589 5590 5591 5592 5593 5594 5595 5596
		/*
		 * 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.
		 */
5597
		rq_unlock(rq, rf);
W
Wanpeng Li 已提交
5598
		raw_spin_lock(&next->pi_lock);
5599
		rq_relock(rq, rf);
W
Wanpeng Li 已提交
5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610

		/*
		 * 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;
		}

5611
		/* Find suitable destination for @next, with force if needed. */
5612
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5613
		rq = __migrate_task(rq, rf, next, dest_cpu);
5614
		if (rq != dead_rq) {
5615
			rq_unlock(rq, rf);
5616
			rq = dead_rq;
5617 5618
			*rf = orf;
			rq_relock(rq, rf);
5619
		}
W
Wanpeng Li 已提交
5620
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5621
	}
5622

5623
	rq->stop = stop;
5624
}
L
Linus Torvalds 已提交
5625 5626
#endif /* CONFIG_HOTPLUG_CPU */

5627
void set_rq_online(struct rq *rq)
5628 5629 5630 5631
{
	if (!rq->online) {
		const struct sched_class *class;

5632
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5633 5634 5635 5636 5637 5638 5639 5640 5641
		rq->online = 1;

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

5642
void set_rq_offline(struct rq *rq)
5643 5644 5645 5646 5647 5648 5649 5650 5651
{
	if (rq->online) {
		const struct sched_class *class;

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

5652
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5653 5654 5655 5656
		rq->online = 0;
	}
}

5657
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5658
{
5659
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5660

5661 5662 5663
	rq->age_stamp = sched_clock_cpu(cpu);
}

I
Ingo Molnar 已提交
5664 5665 5666 5667
/*
 * used to mark begin/end of suspend/resume:
 */
static int num_cpus_frozen;
5668

L
Linus Torvalds 已提交
5669
/*
5670 5671 5672
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
5673 5674 5675
 *
 * 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
5676
 */
5677
static void cpuset_cpu_active(void)
5678
{
5679
	if (cpuhp_tasks_frozen) {
5680 5681 5682 5683 5684 5685
		/*
		 * 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.
		 */
5686 5687
		partition_sched_domains(1, NULL, NULL);
		if (--num_cpus_frozen)
5688
			return;
5689 5690 5691 5692 5693
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
5694
		cpuset_force_rebuild();
5695
	}
5696
	cpuset_update_active_cpus();
5697
}
5698

5699
static int cpuset_cpu_inactive(unsigned int cpu)
5700
{
5701
	if (!cpuhp_tasks_frozen) {
5702
		if (dl_cpu_busy(cpu))
5703
			return -EBUSY;
5704
		cpuset_update_active_cpus();
5705
	} else {
5706 5707
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
5708
	}
5709
	return 0;
5710 5711
}

5712
int sched_cpu_activate(unsigned int cpu)
5713
{
5714
	struct rq *rq = cpu_rq(cpu);
5715
	struct rq_flags rf;
5716

5717
	set_cpu_active(cpu, true);
5718

5719
	if (sched_smp_initialized) {
5720
		sched_domains_numa_masks_set(cpu);
5721
		cpuset_cpu_active();
5722
	}
5723 5724 5725 5726 5727

	/*
	 * Put the rq online, if not already. This happens:
	 *
	 * 1) In the early boot process, because we build the real domains
I
Ingo Molnar 已提交
5728
	 *    after all CPUs have been brought up.
5729 5730 5731 5732
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
5733
	rq_lock_irqsave(rq, &rf);
5734 5735 5736 5737
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
5738
	rq_unlock_irqrestore(rq, &rf);
5739 5740 5741

	update_max_interval();

5742
	return 0;
5743 5744
}

5745
int sched_cpu_deactivate(unsigned int cpu)
5746 5747 5748
{
	int ret;

5749
	set_cpu_active(cpu, false);
5750 5751 5752 5753 5754 5755 5756
	/*
	 * 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.
	 *
	 * Do sync before park smpboot threads to take care the rcu boost case.
	 */
5757
	synchronize_rcu_mult(call_rcu, call_rcu_sched);
5758 5759 5760 5761 5762 5763 5764 5765

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
5766
	}
5767 5768
	sched_domains_numa_masks_clear(cpu);
	return 0;
5769 5770
}

5771 5772 5773 5774 5775 5776 5777 5778
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();
}

5779 5780 5781
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
5782
	sched_rq_cpu_starting(cpu);
5783
	return 0;
5784 5785
}

5786 5787 5788 5789
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5790
	struct rq_flags rf;
5791 5792 5793

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
5794 5795

	rq_lock_irqsave(rq, &rf);
5796 5797 5798 5799
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
5800
	migrate_tasks(rq, &rf);
5801
	BUG_ON(rq->nr_running != 1);
5802 5803
	rq_unlock_irqrestore(rq, &rf);

5804 5805
	calc_load_migrate(rq);
	update_max_interval();
5806
	nohz_balance_exit_idle(cpu);
5807
	hrtick_clear(rq);
5808 5809 5810 5811
	return 0;
}
#endif

P
Peter Zijlstra 已提交
5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827
#ifdef CONFIG_SCHED_SMT
DEFINE_STATIC_KEY_FALSE(sched_smt_present);

static void sched_init_smt(void)
{
	/*
	 * We've enumerated all CPUs and will assume that if any CPU
	 * has SMT siblings, CPU0 will too.
	 */
	if (cpumask_weight(cpu_smt_mask(0)) > 1)
		static_branch_enable(&sched_smt_present);
}
#else
static inline void sched_init_smt(void) { }
#endif

L
Linus Torvalds 已提交
5828 5829
void __init sched_init_smp(void)
{
5830 5831
	sched_init_numa();

5832 5833
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
I
Ingo Molnar 已提交
5834
	 * CPU masks are stable and all blatant races in the below code cannot
5835 5836
	 * happen.
	 */
5837
	mutex_lock(&sched_domains_mutex);
P
Peter Zijlstra 已提交
5838
	sched_init_domains(cpu_active_mask);
5839
	mutex_unlock(&sched_domains_mutex);
5840

5841
	/* Move init over to a non-isolated CPU */
5842
	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5843
		BUG();
I
Ingo Molnar 已提交
5844
	sched_init_granularity();
5845

5846
	init_sched_rt_class();
5847
	init_sched_dl_class();
P
Peter Zijlstra 已提交
5848 5849 5850

	sched_init_smt();

5851
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
5852
}
5853 5854 5855

static int __init migration_init(void)
{
5856
	sched_rq_cpu_starting(smp_processor_id());
5857
	return 0;
L
Linus Torvalds 已提交
5858
}
5859 5860
early_initcall(migration_init);

L
Linus Torvalds 已提交
5861 5862 5863
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
5864
	sched_init_granularity();
L
Linus Torvalds 已提交
5865 5866 5867 5868 5869 5870 5871 5872 5873 5874
}
#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);
}

5875
#ifdef CONFIG_CGROUP_SCHED
5876 5877 5878 5879
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
5880
struct task_group root_task_group;
5881
LIST_HEAD(task_groups);
5882 5883 5884

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
5885
#endif
P
Peter Zijlstra 已提交
5886

5887
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5888
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
5889

L
Linus Torvalds 已提交
5890 5891
void __init sched_init(void)
{
I
Ingo Molnar 已提交
5892
	int i, j;
5893 5894
	unsigned long alloc_size = 0, ptr;

5895
	sched_clock_init();
5896
	wait_bit_init();
5897

5898 5899 5900 5901 5902 5903 5904
#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) {
5905
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5906 5907

#ifdef CONFIG_FAIR_GROUP_SCHED
5908
		root_task_group.se = (struct sched_entity **)ptr;
5909 5910
		ptr += nr_cpu_ids * sizeof(void **);

5911
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5912
		ptr += nr_cpu_ids * sizeof(void **);
5913

5914
#endif /* CONFIG_FAIR_GROUP_SCHED */
5915
#ifdef CONFIG_RT_GROUP_SCHED
5916
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5917 5918
		ptr += nr_cpu_ids * sizeof(void **);

5919
		root_task_group.rt_rq = (struct rt_rq **)ptr;
5920 5921
		ptr += nr_cpu_ids * sizeof(void **);

5922
#endif /* CONFIG_RT_GROUP_SCHED */
5923
	}
5924
#ifdef CONFIG_CPUMASK_OFFSTACK
5925 5926 5927
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5928 5929
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5930
	}
5931
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
5932

I
Ingo Molnar 已提交
5933 5934
	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
5935

G
Gregory Haskins 已提交
5936 5937 5938 5939
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

5940
#ifdef CONFIG_RT_GROUP_SCHED
5941
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
5942
			global_rt_period(), global_rt_runtime());
5943
#endif /* CONFIG_RT_GROUP_SCHED */
5944

D
Dhaval Giani 已提交
5945
#ifdef CONFIG_CGROUP_SCHED
5946 5947
	task_group_cache = KMEM_CACHE(task_group, 0);

5948 5949
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
5950
	INIT_LIST_HEAD(&root_task_group.siblings);
5951
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
5952
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
5953

5954
	for_each_possible_cpu(i) {
5955
		struct rq *rq;
L
Linus Torvalds 已提交
5956 5957

		rq = cpu_rq(i);
5958
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
5959
		rq->nr_running = 0;
5960 5961
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
5962
		init_cfs_rq(&rq->cfs);
5963 5964
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
5965
#ifdef CONFIG_FAIR_GROUP_SCHED
5966
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
5967
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
5968
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
5969
		/*
I
Ingo Molnar 已提交
5970
		 * How much CPU bandwidth does root_task_group get?
D
Dhaval Giani 已提交
5971 5972
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
I
Ingo Molnar 已提交
5973 5974
		 * gets 100% of the CPU resources in the system. This overall
		 * system CPU resource is divided among the tasks of
5975
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
5976 5977 5978
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
5979
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
5980
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
I
Ingo Molnar 已提交
5981
		 * then A0's share of the CPU resource is:
D
Dhaval Giani 已提交
5982
		 *
5983
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
5984
		 *
5985 5986
		 * 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
5987
		 */
5988
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
5989
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
5990 5991 5992
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
5993
#ifdef CONFIG_RT_GROUP_SCHED
5994
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
5995
#endif
L
Linus Torvalds 已提交
5996

I
Ingo Molnar 已提交
5997 5998
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
5999

L
Linus Torvalds 已提交
6000
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6001
		rq->sd = NULL;
G
Gregory Haskins 已提交
6002
		rq->rd = NULL;
6003
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6004
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
6005
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6006
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6007
		rq->push_cpu = 0;
6008
		rq->cpu = i;
6009
		rq->online = 0;
6010 6011
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6012
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6013 6014 6015

		INIT_LIST_HEAD(&rq->cfs_tasks);

6016
		rq_attach_root(rq, &def_root_domain);
6017
#ifdef CONFIG_NO_HZ_COMMON
6018
		rq->last_load_update_tick = jiffies;
6019
		rq->nohz_flags = 0;
6020
#endif
6021 6022 6023
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
6024
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
6025
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
6026 6027 6028
		atomic_set(&rq->nr_iowait, 0);
	}

6029
	set_load_weight(&init_task, false);
6030

L
Linus Torvalds 已提交
6031 6032 6033
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
V
Vegard Nossum 已提交
6034
	mmgrab(&init_mm);
L
Linus Torvalds 已提交
6035 6036 6037 6038 6039 6040 6041 6042 6043
	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());
6044 6045 6046

	calc_load_update = jiffies + LOAD_FREQ;

6047
#ifdef CONFIG_SMP
6048
	idle_thread_set_boot_cpu();
6049
	set_cpu_rq_start_time(smp_processor_id());
6050 6051
#endif
	init_sched_fair_class();
6052

6053 6054
	init_schedstats();

6055
	scheduler_running = 1;
L
Linus Torvalds 已提交
6056 6057
}

6058
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6059 6060
static inline int preempt_count_equals(int preempt_offset)
{
6061
	int nested = preempt_count() + rcu_preempt_depth();
6062

A
Arnd Bergmann 已提交
6063
	return (nested == preempt_offset);
6064 6065
}

6066
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6067
{
P
Peter Zijlstra 已提交
6068 6069 6070 6071 6072
	/*
	 * 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.
	 */
6073
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
6074 6075 6076 6077
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
6078
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
6079

6080 6081 6082 6083 6084
	___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
6085
{
I
Ingo Molnar 已提交
6086 6087 6088
	/* Ratelimiting timestamp: */
	static unsigned long prev_jiffy;

6089
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
6090

I
Ingo Molnar 已提交
6091 6092 6093
	/* WARN_ON_ONCE() by default, no rate limit required: */
	rcu_sleep_check();

6094 6095
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
6096 6097
	    system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
	    oops_in_progress)
I
Ingo Molnar 已提交
6098
		return;
6099

I
Ingo Molnar 已提交
6100 6101 6102 6103
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

I
Ingo Molnar 已提交
6104
	/* Save this before calling printk(), since that will clobber it: */
6105 6106
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
6107 6108 6109 6110 6111 6112 6113
	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
6114

6115 6116 6117
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
6118 6119 6120
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
6121 6122
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
6123
		pr_err("Preemption disabled at:");
6124
		print_ip_sym(preempt_disable_ip);
6125 6126
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
6127
	dump_stack();
6128
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
6129
}
6130
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
6131 6132 6133
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6134
void normalize_rt_tasks(void)
6135
{
6136
	struct task_struct *g, *p;
6137 6138 6139
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
6140

6141
	read_lock(&tasklist_lock);
6142
	for_each_process_thread(g, p) {
6143 6144 6145
		/*
		 * Only normalize user tasks:
		 */
6146
		if (p->flags & PF_KTHREAD)
6147 6148
			continue;

6149 6150 6151 6152
		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
6153

6154
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
6155 6156 6157 6158
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
6159
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
6160
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6161
			continue;
I
Ingo Molnar 已提交
6162
		}
L
Linus Torvalds 已提交
6163

6164
		__sched_setscheduler(p, &attr, false, false);
6165
	}
6166
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
6167 6168 6169
}

#endif /* CONFIG_MAGIC_SYSRQ */
6170

6171
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6172
/*
6173
 * These functions are only useful for the IA64 MCA handling, or kdb.
6174 6175 6176 6177 6178 6179 6180 6181 6182
 *
 * 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.
 */

/**
I
Ingo Molnar 已提交
6183
 * curr_task - return the current task for a given CPU.
6184 6185 6186
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6187 6188
 *
 * Return: The current task for @cpu.
6189
 */
6190
struct task_struct *curr_task(int cpu)
6191 6192 6193 6194
{
	return cpu_curr(cpu);
}

6195 6196 6197
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6198
/**
I
Ingo Molnar 已提交
6199
 * set_curr_task - set the current task for a given CPU.
6200 6201 6202 6203
 * @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
6204
 * are serviced on a separate stack. It allows the architecture to switch the
I
Ingo Molnar 已提交
6205
 * notion of the current task on a CPU in a non-blocking manner. This function
6206 6207 6208 6209 6210 6211 6212
 * 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!
 */
6213
void ia64_set_curr_task(int cpu, struct task_struct *p)
6214 6215 6216 6217 6218
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6219

D
Dhaval Giani 已提交
6220
#ifdef CONFIG_CGROUP_SCHED
6221 6222 6223
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6224
static void sched_free_group(struct task_group *tg)
6225 6226 6227
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6228
	autogroup_free(tg);
6229
	kmem_cache_free(task_group_cache, tg);
6230 6231 6232
}

/* allocate runqueue etc for a new task group */
6233
struct task_group *sched_create_group(struct task_group *parent)
6234 6235 6236
{
	struct task_group *tg;

6237
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6238 6239 6240
	if (!tg)
		return ERR_PTR(-ENOMEM);

6241
	if (!alloc_fair_sched_group(tg, parent))
6242 6243
		goto err;

6244
	if (!alloc_rt_sched_group(tg, parent))
6245 6246
		goto err;

6247 6248 6249
	return tg;

err:
6250
	sched_free_group(tg);
6251 6252 6253 6254 6255 6256 6257
	return ERR_PTR(-ENOMEM);
}

void sched_online_group(struct task_group *tg, struct task_group *parent)
{
	unsigned long flags;

6258
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6259
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6260

I
Ingo Molnar 已提交
6261 6262
	/* Root should already exist: */
	WARN_ON(!parent);
P
Peter Zijlstra 已提交
6263 6264 6265

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6266
	list_add_rcu(&tg->siblings, &parent->children);
6267
	spin_unlock_irqrestore(&task_group_lock, flags);
6268 6269

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
6270 6271
}

6272
/* rcu callback to free various structures associated with a task group */
6273
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6274
{
I
Ingo Molnar 已提交
6275
	/* Now it should be safe to free those cfs_rqs: */
6276
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6277 6278
}

6279
void sched_destroy_group(struct task_group *tg)
6280
{
I
Ingo Molnar 已提交
6281
	/* Wait for possible concurrent references to cfs_rqs complete: */
6282
	call_rcu(&tg->rcu, sched_free_group_rcu);
6283 6284 6285
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6286
{
6287
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
6288

I
Ingo Molnar 已提交
6289
	/* End participation in shares distribution: */
6290
	unregister_fair_sched_group(tg);
6291 6292

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6293
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6294
	list_del_rcu(&tg->siblings);
6295
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6296 6297
}

6298
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
6299
{
P
Peter Zijlstra 已提交
6300
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6301

6302 6303 6304 6305 6306 6307
	/*
	 * 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
6308 6309 6310 6311
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6312
#ifdef CONFIG_FAIR_GROUP_SCHED
6313 6314
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
6315
	else
P
Peter Zijlstra 已提交
6316
#endif
6317
		set_task_rq(tsk, task_cpu(tsk));
6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328
}

/*
 * 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)
{
6329 6330
	int queued, running, queue_flags =
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6331 6332 6333 6334
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);
6335
	update_rq_clock(rq);
6336 6337 6338 6339 6340

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

	if (queued)
6341
		dequeue_task(rq, tsk, queue_flags);
6342
	if (running)
6343 6344 6345
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
6346

6347
	if (queued)
6348
		enqueue_task(rq, tsk, queue_flags);
6349
	if (running)
6350
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6351

6352
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
6353
}
6354

6355
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6356
{
6357
	return css ? container_of(css, struct task_group, css) : NULL;
6358 6359
}

6360 6361
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6362
{
6363 6364
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
6365

6366
	if (!parent) {
6367
		/* This is early initialization for the top cgroup */
6368
		return &root_task_group.css;
6369 6370
	}

6371
	tg = sched_create_group(parent);
6372 6373 6374 6375 6376 6377
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388
/* Expose task group only after completing cgroup initialization */
static int cpu_cgroup_css_online(struct cgroup_subsys_state *css)
{
	struct task_group *tg = css_tg(css);
	struct task_group *parent = css_tg(css->parent);

	if (parent)
		sched_online_group(tg, parent);
	return 0;
}

6389
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6390
{
6391
	struct task_group *tg = css_tg(css);
6392

6393
	sched_offline_group(tg);
6394 6395
}

6396
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6397
{
6398
	struct task_group *tg = css_tg(css);
6399

6400 6401 6402 6403
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
6404 6405
}

6406 6407 6408 6409
/*
 * 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.
 */
6410
static void cpu_cgroup_fork(struct task_struct *task)
6411
{
6412 6413 6414 6415 6416
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

6417
	update_rq_clock(rq);
6418 6419 6420
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
6421 6422
}

6423
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6424
{
6425
	struct task_struct *task;
6426
	struct cgroup_subsys_state *css;
6427
	int ret = 0;
6428

6429
	cgroup_taskset_for_each(task, css, tset) {
6430
#ifdef CONFIG_RT_GROUP_SCHED
6431
		if (!sched_rt_can_attach(css_tg(css), task))
6432
			return -EINVAL;
6433
#else
6434 6435 6436
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
6437
#endif
6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453
		/*
		 * 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;
6454
	}
6455
	return ret;
6456
}
6457

6458
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6459
{
6460
	struct task_struct *task;
6461
	struct cgroup_subsys_state *css;
6462

6463
	cgroup_taskset_for_each(task, css, tset)
6464
		sched_move_task(task);
6465 6466
}

6467
#ifdef CONFIG_FAIR_GROUP_SCHED
6468 6469
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
6470
{
6471
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
6472 6473
}

6474 6475
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
6476
{
6477
	struct task_group *tg = css_tg(css);
6478

6479
	return (u64) scale_load_down(tg->shares);
6480
}
6481 6482

#ifdef CONFIG_CFS_BANDWIDTH
6483 6484
static DEFINE_MUTEX(cfs_constraints_mutex);

6485 6486 6487
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

6488 6489
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

6490 6491
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
6492
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
6493
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513

	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;

6514 6515 6516 6517 6518
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
6519 6520 6521 6522 6523
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

6524
	runtime_enabled = quota != RUNTIME_INF;
6525
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6526 6527 6528 6529 6530 6531
	/*
	 * 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();
6532 6533 6534
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
6535

P
Paul Turner 已提交
6536
	__refill_cfs_bandwidth_runtime(cfs_b);
I
Ingo Molnar 已提交
6537 6538

	/* Restart the period timer (if active) to handle new period expiry: */
P
Peter Zijlstra 已提交
6539 6540
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
I
Ingo Molnar 已提交
6541

6542 6543
	raw_spin_unlock_irq(&cfs_b->lock);

6544
	for_each_online_cpu(i) {
6545
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6546
		struct rq *rq = cfs_rq->rq;
6547
		struct rq_flags rf;
6548

6549
		rq_lock_irq(rq, &rf);
6550
		cfs_rq->runtime_enabled = runtime_enabled;
6551
		cfs_rq->runtime_remaining = 0;
6552

6553
		if (cfs_rq->throttled)
6554
			unthrottle_cfs_rq(cfs_rq);
6555
		rq_unlock_irq(rq, &rf);
6556
	}
6557 6558
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
6559 6560
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
6561
	put_online_cpus();
6562

6563
	return ret;
6564 6565 6566 6567 6568 6569
}

int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
{
	u64 quota, period;

6570
	period = ktime_to_ns(tg->cfs_bandwidth.period);
6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582
	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;

6583
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6584 6585
		return -1;

6586
	quota_us = tg->cfs_bandwidth.quota;
6587 6588 6589 6590 6591 6592 6593 6594 6595 6596
	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;
6597
	quota = tg->cfs_bandwidth.quota;
6598 6599 6600 6601 6602 6603 6604 6605

	return tg_set_cfs_bandwidth(tg, period, quota);
}

long tg_get_cfs_period(struct task_group *tg)
{
	u64 cfs_period_us;

6606
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6607 6608 6609 6610 6611
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

6612 6613
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
6614
{
6615
	return tg_get_cfs_quota(css_tg(css));
6616 6617
}

6618 6619
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
6620
{
6621
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6622 6623
}

6624 6625
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6626
{
6627
	return tg_get_cfs_period(css_tg(css));
6628 6629
}

6630 6631
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
6632
{
6633
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
6634 6635
}

6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667
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;
6668
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6669 6670 6671 6672 6673
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
6674
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6675 6676

		quota = normalize_cfs_quota(tg, d);
6677
		parent_quota = parent_b->hierarchical_quota;
6678 6679

		/*
I
Ingo Molnar 已提交
6680 6681
		 * Ensure max(child_quota) <= parent_quota, inherit when no
		 * limit is set:
6682 6683 6684 6685 6686 6687
		 */
		if (quota == RUNTIME_INF)
			quota = parent_quota;
		else if (parent_quota != RUNTIME_INF && quota > parent_quota)
			return -EINVAL;
	}
6688
	cfs_b->hierarchical_quota = quota;
6689 6690 6691 6692 6693 6694

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
6695
	int ret;
6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706
	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);
	}

6707 6708 6709 6710 6711
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6712
}
6713

6714
static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6715
{
6716
	struct task_group *tg = css_tg(seq_css(sf));
6717
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6718

6719 6720 6721
	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);
6722 6723 6724

	return 0;
}
6725
#endif /* CONFIG_CFS_BANDWIDTH */
6726
#endif /* CONFIG_FAIR_GROUP_SCHED */
6727

6728
#ifdef CONFIG_RT_GROUP_SCHED
6729 6730
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
6731
{
6732
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
6733 6734
}

6735 6736
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
6737
{
6738
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
6739
}
6740

6741 6742
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
6743
{
6744
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
6745 6746
}

6747 6748
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6749
{
6750
	return sched_group_rt_period(css_tg(css));
6751
}
6752
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
6753

6754
static struct cftype cpu_legacy_files[] = {
6755
#ifdef CONFIG_FAIR_GROUP_SCHED
6756 6757
	{
		.name = "shares",
6758 6759
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
6760
	},
6761
#endif
6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772
#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,
	},
6773 6774
	{
		.name = "stat",
6775
		.seq_show = cpu_cfs_stat_show,
6776
	},
6777
#endif
6778
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6779
	{
P
Peter Zijlstra 已提交
6780
		.name = "rt_runtime_us",
6781 6782
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
6783
	},
6784 6785
	{
		.name = "rt_period_us",
6786 6787
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
6788
	},
6789
#endif
I
Ingo Molnar 已提交
6790
	{ }	/* Terminate */
6791 6792
};

6793 6794
static int cpu_extra_stat_show(struct seq_file *sf,
			       struct cgroup_subsys_state *css)
6795 6796 6797
{
#ifdef CONFIG_CFS_BANDWIDTH
	{
6798
		struct task_group *tg = css_tg(css);
6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955
		struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
		u64 throttled_usec;

		throttled_usec = cfs_b->throttled_time;
		do_div(throttled_usec, NSEC_PER_USEC);

		seq_printf(sf, "nr_periods %d\n"
			   "nr_throttled %d\n"
			   "throttled_usec %llu\n",
			   cfs_b->nr_periods, cfs_b->nr_throttled,
			   throttled_usec);
	}
#endif
	return 0;
}

#ifdef CONFIG_FAIR_GROUP_SCHED
static u64 cpu_weight_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
{
	struct task_group *tg = css_tg(css);
	u64 weight = scale_load_down(tg->shares);

	return DIV_ROUND_CLOSEST_ULL(weight * CGROUP_WEIGHT_DFL, 1024);
}

static int cpu_weight_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cft, u64 weight)
{
	/*
	 * cgroup weight knobs should use the common MIN, DFL and MAX
	 * values which are 1, 100 and 10000 respectively.  While it loses
	 * a bit of range on both ends, it maps pretty well onto the shares
	 * value used by scheduler and the round-trip conversions preserve
	 * the original value over the entire range.
	 */
	if (weight < CGROUP_WEIGHT_MIN || weight > CGROUP_WEIGHT_MAX)
		return -ERANGE;

	weight = DIV_ROUND_CLOSEST_ULL(weight * 1024, CGROUP_WEIGHT_DFL);

	return sched_group_set_shares(css_tg(css), scale_load(weight));
}

static s64 cpu_weight_nice_read_s64(struct cgroup_subsys_state *css,
				    struct cftype *cft)
{
	unsigned long weight = scale_load_down(css_tg(css)->shares);
	int last_delta = INT_MAX;
	int prio, delta;

	/* find the closest nice value to the current weight */
	for (prio = 0; prio < ARRAY_SIZE(sched_prio_to_weight); prio++) {
		delta = abs(sched_prio_to_weight[prio] - weight);
		if (delta >= last_delta)
			break;
		last_delta = delta;
	}

	return PRIO_TO_NICE(prio - 1 + MAX_RT_PRIO);
}

static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
				     struct cftype *cft, s64 nice)
{
	unsigned long weight;

	if (nice < MIN_NICE || nice > MAX_NICE)
		return -ERANGE;

	weight = sched_prio_to_weight[NICE_TO_PRIO(nice) - MAX_RT_PRIO];
	return sched_group_set_shares(css_tg(css), scale_load(weight));
}
#endif

static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
						  long period, long quota)
{
	if (quota < 0)
		seq_puts(sf, "max");
	else
		seq_printf(sf, "%ld", quota);

	seq_printf(sf, " %ld\n", period);
}

/* caller should put the current value in *@periodp before calling */
static int __maybe_unused cpu_period_quota_parse(char *buf,
						 u64 *periodp, u64 *quotap)
{
	char tok[21];	/* U64_MAX */

	if (!sscanf(buf, "%s %llu", tok, periodp))
		return -EINVAL;

	*periodp *= NSEC_PER_USEC;

	if (sscanf(tok, "%llu", quotap))
		*quotap *= NSEC_PER_USEC;
	else if (!strcmp(tok, "max"))
		*quotap = RUNTIME_INF;
	else
		return -EINVAL;

	return 0;
}

#ifdef CONFIG_CFS_BANDWIDTH
static int cpu_max_show(struct seq_file *sf, void *v)
{
	struct task_group *tg = css_tg(seq_css(sf));

	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
	return 0;
}

static ssize_t cpu_max_write(struct kernfs_open_file *of,
			     char *buf, size_t nbytes, loff_t off)
{
	struct task_group *tg = css_tg(of_css(of));
	u64 period = tg_get_cfs_period(tg);
	u64 quota;
	int ret;

	ret = cpu_period_quota_parse(buf, &period, &quota);
	if (!ret)
		ret = tg_set_cfs_bandwidth(tg, period, quota);
	return ret ?: nbytes;
}
#endif

static struct cftype cpu_files[] = {
#ifdef CONFIG_FAIR_GROUP_SCHED
	{
		.name = "weight",
		.flags = CFTYPE_NOT_ON_ROOT,
		.read_u64 = cpu_weight_read_u64,
		.write_u64 = cpu_weight_write_u64,
	},
	{
		.name = "weight.nice",
		.flags = CFTYPE_NOT_ON_ROOT,
		.read_s64 = cpu_weight_nice_read_s64,
		.write_s64 = cpu_weight_nice_write_s64,
	},
#endif
#ifdef CONFIG_CFS_BANDWIDTH
	{
		.name = "max",
		.flags = CFTYPE_NOT_ON_ROOT,
		.seq_show = cpu_max_show,
		.write = cpu_max_write,
	},
#endif
	{ }	/* terminate */
};

6956
struct cgroup_subsys cpu_cgrp_subsys = {
6957
	.css_alloc	= cpu_cgroup_css_alloc,
6958
	.css_online	= cpu_cgroup_css_online,
6959
	.css_released	= cpu_cgroup_css_released,
6960
	.css_free	= cpu_cgroup_css_free,
6961
	.css_extra_stat_show = cpu_extra_stat_show,
6962
	.fork		= cpu_cgroup_fork,
6963 6964
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
6965
	.legacy_cftypes	= cpu_legacy_files,
6966
	.dfl_cftypes	= cpu_files,
6967
	.early_init	= true,
6968
	.threaded	= true,
6969 6970
};

6971
#endif	/* CONFIG_CGROUP_SCHED */
6972

6973 6974 6975 6976 6977
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018

/*
 * 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,
};