core.c 173.8 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 "sched.h"
L
Linus Torvalds 已提交
9

10
#include <asm/switch_to.h>
11
#include <asm/tlb.h>
L
Linus Torvalds 已提交
12

13
#include "../workqueue_internal.h"
14
#include "../smpboot.h"
15

16
#define CREATE_TRACE_POINTS
17
#include <trace/events/sched.h>
18

19
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
20

21
#if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
I
Ingo Molnar 已提交
22 23
/*
 * Debugging: various feature bits
24 25 26 27
 *
 * 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
28
 */
P
Peter Zijlstra 已提交
29 30
#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |
I
Ingo Molnar 已提交
31
const_debug unsigned int sysctl_sched_features =
32
#include "features.h"
P
Peter Zijlstra 已提交
33 34
	0;
#undef SCHED_FEAT
35
#endif
P
Peter Zijlstra 已提交
36

37 38 39 40 41 42
/*
 * 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;

43 44 45 46 47 48 49 50
/*
 * 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
51
/*
I
Ingo Molnar 已提交
52
 * period over which we measure -rt task CPU usage in us.
P
Peter Zijlstra 已提交
53 54
 * default: 1s
 */
P
Peter Zijlstra 已提交
55
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
56

57
__read_mostly int scheduler_running;
58

P
Peter Zijlstra 已提交
59 60 61 62 63
/*
 * 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
64

65 66 67
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
68
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
69 70 71 72 73 74 75 76 77 78
	__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))) {
79
			rq_pin_lock(rq, rf);
80 81 82 83 84 85 86 87 88 89 90 91
			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.
 */
92
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
93 94 95 96 97 98
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
99
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
100 101 102 103 104 105 106 107 108 109 110 111
		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)
		 *
112
		 * If we observe the old CPU in task_rq_lock, the acquire of
113 114
		 * the old rq->lock will fully serialize against the stores.
		 *
I
Ingo Molnar 已提交
115
		 * If we observe the new CPU in task_rq_lock, the acquire will
116 117 118
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
119
			rq_pin_lock(rq, rf);
120 121 122
			return rq;
		}
		raw_spin_unlock(&rq->lock);
123
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
124 125 126 127 128 129

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

130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197
/*
 * 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
198 199
	if (sched_feat(WARN_DOUBLE_CLOCK))
		SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
200 201
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
202

203 204 205 206 207 208 209 210
	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
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228
#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);
229
	struct rq_flags rf;
P
Peter Zijlstra 已提交
230 231 232

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

233
	rq_lock(rq, &rf);
234
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
235
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
236
	rq_unlock(rq, &rf);
P
Peter Zijlstra 已提交
237 238 239 240

	return HRTIMER_NORESTART;
}

241
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
242

243
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
244 245 246
{
	struct hrtimer *timer = &rq->hrtick_timer;

247
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
248 249
}

250 251 252 253
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
254
{
255
	struct rq *rq = arg;
256
	struct rq_flags rf;
257

258
	rq_lock(rq, &rf);
P
Peter Zijlstra 已提交
259
	__hrtick_restart(rq);
260
	rq->hrtick_csd_pending = 0;
261
	rq_unlock(rq, &rf);
262 263
}

264 265 266 267 268
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
269
void hrtick_start(struct rq *rq, u64 delay)
270
{
271
	struct hrtimer *timer = &rq->hrtick_timer;
272 273 274 275 276 277 278 279 280
	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);
281

282
	hrtimer_set_expires(timer, time);
283 284

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
285
		__hrtick_restart(rq);
286
	} else if (!rq->hrtick_csd_pending) {
287
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
288 289
		rq->hrtick_csd_pending = 1;
	}
290 291
}

292 293 294 295 296 297
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
298
void hrtick_start(struct rq *rq, u64 delay)
299
{
W
Wanpeng Li 已提交
300 301 302 303 304
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
305 306
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
307 308
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
309

310
static void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
311
{
312 313
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
314

315 316 317 318
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
319

320 321
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
322
}
A
Andrew Morton 已提交
323
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
324 325 326 327
static inline void hrtick_clear(struct rq *rq)
{
}

328
static inline void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
329 330
{
}
A
Andrew Morton 已提交
331
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
332

333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350
/*
 * 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;								\
})

351
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
352 353 354 355 356 357 358 359 360 361
/*
 * 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);
}
362 363 364 365 366 367 368 369 370 371

/*
 * 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);
372
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
373 374 375 376 377 378 379 380 381 382 383 384 385 386

	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;
}

387 388 389 390 391 392
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
393 394 395 396 397 398 399

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
400 401
#endif

402 403 404 405 406 407 408 409 410 411
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
412
	 * barrier implied by the wakeup in wake_up_q().
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
	 */
	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
435
		/* Task can safely be re-inserted now: */
436 437 438 439 440 441 442 443 444 445 446 447
		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
448
/*
449
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
450 451 452 453 454
 *
 * 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.
 */
455
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
456
{
457
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
458 459
	int cpu;

460
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
461

462
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
463 464
		return;

465
	cpu = cpu_of(rq);
466

467
	if (cpu == smp_processor_id()) {
468
		set_tsk_need_resched(curr);
469
		set_preempt_need_resched();
I
Ingo Molnar 已提交
470
		return;
471
	}
I
Ingo Molnar 已提交
472

473
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
474
		smp_send_reschedule(cpu);
475 476
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
477 478
}

479
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
480 481 482 483
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

484
	raw_spin_lock_irqsave(&rq->lock, flags);
485 486
	if (cpu_online(cpu) || cpu == smp_processor_id())
		resched_curr(rq);
487
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
488
}
489

490
#ifdef CONFIG_SMP
491
#ifdef CONFIG_NO_HZ_COMMON
492
/*
I
Ingo Molnar 已提交
493 494
 * In the semi idle case, use the nearest busy CPU for migrating timers
 * from an idle CPU.  This is good for power-savings.
495 496
 *
 * We don't do similar optimization for completely idle system, as
I
Ingo Molnar 已提交
497 498
 * 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).
499
 */
500
int get_nohz_timer_target(void)
501
{
502
	int i, cpu = smp_processor_id();
503 504
	struct sched_domain *sd;

505
	if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
506 507
		return cpu;

508
	rcu_read_lock();
509
	for_each_domain(cpu, sd) {
510
		for_each_cpu(i, sched_domain_span(sd)) {
511 512 513
			if (cpu == i)
				continue;

514
			if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
515 516 517 518
				cpu = i;
				goto unlock;
			}
		}
519
	}
520

521 522
	if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
		cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
523 524
unlock:
	rcu_read_unlock();
525 526
	return cpu;
}
I
Ingo Molnar 已提交
527

528 529 530 531 532 533 534 535 536 537
/*
 * 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.
 */
538
static void wake_up_idle_cpu(int cpu)
539 540 541 542 543 544
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

545
	if (set_nr_and_not_polling(rq->idle))
546
		smp_send_reschedule(cpu);
547 548
	else
		trace_sched_wake_idle_without_ipi(cpu);
549 550
}

551
static bool wake_up_full_nohz_cpu(int cpu)
552
{
553 554 555 556 557 558
	/*
	 * 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.
	 */
559 560
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
561
	if (tick_nohz_full_cpu(cpu)) {
562 563
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
564
			tick_nohz_full_kick_cpu(cpu);
565 566 567 568 569 570
		return true;
	}

	return false;
}

571 572 573 574 575
/*
 * 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.
 */
576 577
void wake_up_nohz_cpu(int cpu)
{
578
	if (!wake_up_full_nohz_cpu(cpu))
579 580 581
		wake_up_idle_cpu(cpu);
}

582
static inline bool got_nohz_idle_kick(void)
583
{
584
	int cpu = smp_processor_id();
585

586
	if (!(atomic_read(nohz_flags(cpu)) & NOHZ_BALANCE_KICK))
587 588 589 590 591 592 593 594 595
		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
	 */
596
	atomic_andnot(NOHZ_BALANCE_KICK, nohz_flags(cpu));
597
	return false;
598 599
}

600
#else /* CONFIG_NO_HZ_COMMON */
601

602
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
603
{
604
	return false;
P
Peter Zijlstra 已提交
605 606
}

607
#endif /* CONFIG_NO_HZ_COMMON */
608

609
#ifdef CONFIG_NO_HZ_FULL
610
bool sched_can_stop_tick(struct rq *rq)
611
{
612 613 614 615 616 617
	int fifo_nr_running;

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

618
	/*
619 620
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
621
	 */
622 623 624 625 626
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
627 628
	}

629 630 631 632 633 634 635 636 637 638 639 640 641 642
	/*
	 * 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)
643
		return false;
644

645
	return true;
646 647
}
#endif /* CONFIG_NO_HZ_FULL */
648

649
void sched_avg_update(struct rq *rq)
650
{
651 652
	s64 period = sched_avg_period();

653
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
654 655 656 657 658 659
		/*
		 * 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));
660 661 662
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
663 664
}

665
#endif /* CONFIG_SMP */
666

667 668
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
669
/*
670 671 672 673
 * 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.
674
 */
675
int walk_tg_tree_from(struct task_group *from,
676
			     tg_visitor down, tg_visitor up, void *data)
677 678
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
679
	int ret;
680

681 682
	parent = from;

683
down:
P
Peter Zijlstra 已提交
684 685
	ret = (*down)(parent, data);
	if (ret)
686
		goto out;
687 688 689 690 691 692 693
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
694
	ret = (*up)(parent, data);
695 696
	if (ret || parent == from)
		goto out;
697 698 699 700 701

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
702
out:
P
Peter Zijlstra 已提交
703
	return ret;
704 705
}

706
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
707
{
708
	return 0;
P
Peter Zijlstra 已提交
709
}
710 711
#endif

712
static void set_load_weight(struct task_struct *p, bool update_load)
713
{
N
Nikhil Rao 已提交
714 715 716
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
717 718 719
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
720
	if (idle_policy(p->policy)) {
721
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
722
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
723 724
		return;
	}
725

726 727 728 729 730 731 732 733 734 735
	/*
	 * 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];
	}
736 737
}

738
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
739
{
740 741 742
	if (!(flags & ENQUEUE_NOCLOCK))
		update_rq_clock(rq);

743 744
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
745

746
	p->sched_class->enqueue_task(rq, p, flags);
747 748
}

749
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
750
{
751 752 753
	if (!(flags & DEQUEUE_NOCLOCK))
		update_rq_clock(rq);

754 755
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
756

757
	p->sched_class->dequeue_task(rq, p, flags);
758 759
}

760
void activate_task(struct rq *rq, struct task_struct *p, int flags)
761 762 763 764
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

765
	enqueue_task(rq, p, flags);
766 767
}

768
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
769 770 771 772
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

773
	dequeue_task(rq, p, flags);
774 775
}

776
/*
I
Ingo Molnar 已提交
777
 * __normal_prio - return the priority that is based on the static prio
778 779 780
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
781
	return p->static_prio;
782 783
}

784 785 786 787 788 789 790
/*
 * 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.
 */
791
static inline int normal_prio(struct task_struct *p)
792 793 794
{
	int prio;

795 796 797
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
798 799 800 801 802 803 804 805 806 807 808 809 810
		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.
 */
811
static int effective_prio(struct task_struct *p)
812 813 814 815 816 817 818 819 820 821 822 823
{
	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
824 825 826
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
827 828
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
829
 */
830
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
831 832 833 834
{
	return cpu_curr(task_cpu(p)) == p;
}

835
/*
836 837 838 839 840
 * 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().
841
 */
842 843
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
844
				       int oldprio)
845 846 847
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
848
			prev_class->switched_from(rq, p);
849

P
Peter Zijlstra 已提交
850
		p->sched_class->switched_to(rq, p);
851
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
852
		p->sched_class->prio_changed(rq, p, oldprio);
853 854
}

855
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
856 857 858 859 860 861 862 863 864 865
{
	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) {
866
				resched_curr(rq);
867 868 869 870 871 872 873 874 875
				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.
	 */
876
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
877
		rq_clock_skip_update(rq, true);
878 879
}

L
Linus Torvalds 已提交
880
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899
/*
 * 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.
 */
900 901
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
902 903 904 905
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
906
	dequeue_task(rq, p, DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
907
	set_task_cpu(p, new_cpu);
908
	rq_unlock(rq, rf);
P
Peter Zijlstra 已提交
909 910 911

	rq = cpu_rq(new_cpu);

912
	rq_lock(rq, rf);
P
Peter Zijlstra 已提交
913 914
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
915
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
916 917 918 919 920 921 922 923 924 925 926
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
I
Ingo Molnar 已提交
927
 * Move (not current) task off this CPU, onto the destination CPU. We're doing
P
Peter Zijlstra 已提交
928 929 930 931 932 933 934
 * 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.
 */
935 936
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
937
{
938 939 940 941 942 943 944
	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
945 946

	/* Affinity changed (again). */
947
	if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
948
		return rq;
P
Peter Zijlstra 已提交
949

950
	update_rq_clock(rq);
951
	rq = move_queued_task(rq, rf, p, dest_cpu);
952 953

	return rq;
P
Peter Zijlstra 已提交
954 955 956 957 958 959 960 961 962 963
}

/*
 * 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;
964 965
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
966
	struct rq_flags rf;
P
Peter Zijlstra 已提交
967 968

	/*
I
Ingo Molnar 已提交
969 970
	 * 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
971 972 973 974 975 976 977 978
	 */
	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();
979 980

	raw_spin_lock(&p->pi_lock);
981
	rq_lock(rq, &rf);
982 983 984 985 986
	/*
	 * 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.
	 */
987 988
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
989
			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
990 991 992
		else
			p->wake_cpu = arg->dest_cpu;
	}
993
	rq_unlock(rq, &rf);
994 995
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
996 997 998 999
	local_irq_enable();
	return 0;
}

1000 1001 1002 1003 1004
/*
 * 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
1005 1006 1007 1008 1009
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1010 1011
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1012 1013 1014
	struct rq *rq = task_rq(p);
	bool queued, running;

1015
	lockdep_assert_held(&p->pi_lock);
1016 1017 1018 1019 1020 1021 1022 1023 1024 1025

	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);
1026
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1027 1028 1029 1030
	}
	if (running)
		put_prev_task(rq, p);

1031
	p->sched_class->set_cpus_allowed(p, new_mask);
1032 1033

	if (queued)
1034
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1035
	if (running)
1036
		set_curr_task(rq, p);
1037 1038
}

P
Peter Zijlstra 已提交
1039 1040 1041 1042 1043 1044 1045 1046 1047
/*
 * 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.
 */
1048 1049
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1050
{
1051
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1052
	unsigned int dest_cpu;
1053 1054
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1055 1056
	int ret = 0;

1057
	rq = task_rq_lock(p, &rf);
1058
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
1059

1060 1061 1062 1063 1064 1065 1066
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1067 1068 1069 1070 1071 1072 1073 1074 1075
	/*
	 * 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
1076 1077 1078
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1079
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1080 1081 1082 1083 1084 1085
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1086 1087 1088
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
I
Ingo Molnar 已提交
1089
		 * !active we want to ensure they are strict per-CPU threads.
1090 1091 1092 1093 1094 1095
		 */
		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
1096 1097 1098 1099
	/* 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;

1100
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1101 1102 1103
	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. */
1104
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1105 1106 1107
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1108 1109 1110 1111 1112
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1113
		rq = move_queued_task(rq, &rf, p, dest_cpu);
1114
	}
P
Peter Zijlstra 已提交
1115
out:
1116
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1117 1118 1119

	return ret;
}
1120 1121 1122 1123 1124

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
1125 1126
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1127
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1128
{
1129 1130 1131 1132 1133
#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
1134
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1135
			!p->on_rq);
1136

1137 1138 1139 1140 1141 1142 1143 1144 1145
	/*
	 * 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)));

1146
#ifdef CONFIG_LOCKDEP
1147 1148 1149 1150 1151
	/*
	 * 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
1152
	 * see task_group().
1153 1154 1155 1156
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1157 1158 1159
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1160 1161 1162 1163
	/*
	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
	 */
	WARN_ON_ONCE(!cpu_online(new_cpu));
1164 1165
#endif

1166
	trace_sched_migrate_task(p, new_cpu);
1167

1168
	if (task_cpu(p) != new_cpu) {
1169
		if (p->sched_class->migrate_task_rq)
1170
			p->sched_class->migrate_task_rq(p);
1171
		p->se.nr_migrations++;
1172
		perf_event_task_migrate(p);
1173
	}
I
Ingo Molnar 已提交
1174 1175

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1176 1177
}

1178 1179
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1180
	if (task_on_rq_queued(p)) {
1181
		struct rq *src_rq, *dst_rq;
1182
		struct rq_flags srf, drf;
1183 1184 1185 1186

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

1187 1188 1189
		rq_pin_lock(src_rq, &srf);
		rq_pin_lock(dst_rq, &drf);

1190
		p->on_rq = TASK_ON_RQ_MIGRATING;
1191 1192 1193
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1194
		p->on_rq = TASK_ON_RQ_QUEUED;
1195
		check_preempt_curr(dst_rq, p, 0);
1196 1197 1198 1199

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

1200 1201 1202 1203
	} 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
1204
		 * previous CPU our target instead of where it really is.
1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
		 */
		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;

1221 1222 1223
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1224 1225 1226
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1227 1228
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1229
	double_rq_lock(src_rq, dst_rq);
1230

1231 1232 1233 1234 1235 1236
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

1237
	if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1238 1239
		goto unlock;

1240
	if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1241 1242 1243 1244 1245 1246 1247 1248 1249
		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);
1250 1251
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273

	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;

1274 1275 1276 1277
	/*
	 * 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.
	 */
1278 1279 1280
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

1281
	if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
1282 1283
		goto out;

1284
	if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1285 1286
		goto out;

1287
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1288 1289 1290 1291 1292 1293
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1294 1295 1296
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1297 1298 1299 1300 1301 1302 1303
 * 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
1304 1305 1306 1307 1308 1309
 * 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
1310
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1311
{
1312
	int running, queued;
1313
	struct rq_flags rf;
R
Roland McGrath 已提交
1314
	unsigned long ncsw;
1315
	struct rq *rq;
L
Linus Torvalds 已提交
1316

1317 1318 1319 1320 1321 1322 1323 1324
	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);
1325

1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336
		/*
		 * 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
1337 1338 1339
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1340
			cpu_relax();
R
Roland McGrath 已提交
1341
		}
1342

1343 1344 1345 1346 1347
		/*
		 * 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.
		 */
1348
		rq = task_rq_lock(p, &rf);
1349
		trace_sched_wait_task(p);
1350
		running = task_running(rq, p);
1351
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1352
		ncsw = 0;
1353
		if (!match_state || p->state == match_state)
1354
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1355
		task_rq_unlock(rq, p, &rf);
1356

R
Roland McGrath 已提交
1357 1358 1359 1360 1361 1362
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1363 1364 1365 1366 1367 1368 1369 1370 1371 1372
		/*
		 * 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;
		}
1373

1374 1375 1376 1377 1378
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1379
		 * So if it was still runnable (but just not actively
1380 1381 1382
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1383
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1384
			ktime_t to = NSEC_PER_SEC / HZ;
1385 1386 1387

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1388 1389
			continue;
		}
1390

1391 1392 1393 1394 1395 1396 1397
		/*
		 * 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
1398 1399

	return ncsw;
L
Linus Torvalds 已提交
1400 1401 1402 1403 1404 1405 1406 1407 1408
}

/***
 * 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
1409
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1410 1411 1412 1413 1414
 * 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.
 */
1415
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1416 1417 1418 1419 1420 1421 1422 1423 1424
{
	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
1425
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1426

1427
/*
1428
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1429 1430 1431 1432 1433
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
1434
 *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
1435
 *    see __set_cpus_allowed_ptr(). At this point the newly online
I
Ingo Molnar 已提交
1436
 *    CPU isn't yet part of the sched domains, and balancing will not
1437 1438
 *    see it.
 *
I
Ingo Molnar 已提交
1439
 *  - on CPU-down we clear cpu_active() to mask the sched domains and
1440
 *    avoid the load balancer to place new tasks on the to be removed
I
Ingo Molnar 已提交
1441
 *    CPU. Existing tasks will remain running there and will be taken
1442 1443 1444 1445 1446 1447
 *    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.
1448
 */
1449 1450
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1451 1452
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1453 1454
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1455

1456
	/*
I
Ingo Molnar 已提交
1457 1458 1459
	 * 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.
1460 1461 1462 1463 1464 1465 1466 1467
	 */
	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;
1468
			if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1469 1470
				return dest_cpu;
		}
1471
	}
1472

1473 1474
	for (;;) {
		/* Any allowed, online CPU? */
1475
		for_each_cpu(dest_cpu, &p->cpus_allowed) {
1476 1477 1478
			if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu))
				continue;
			if (!cpu_online(dest_cpu))
1479 1480 1481
				continue;
			goto out;
		}
1482

1483
		/* No more Mr. Nice Guy. */
1484 1485
		switch (state) {
		case cpuset:
1486 1487 1488 1489 1490
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
I
Ingo Molnar 已提交
1491
			/* Fall-through */
1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510
		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()) {
1511
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1512 1513
					task_pid_nr(p), p->comm, cpu);
		}
1514 1515 1516 1517 1518
	}

	return dest_cpu;
}

1519
/*
1520
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1521
 */
1522
static inline
1523
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1524
{
1525 1526
	lockdep_assert_held(&p->pi_lock);

1527
	if (p->nr_cpus_allowed > 1)
1528
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1529
	else
1530
		cpu = cpumask_any(&p->cpus_allowed);
1531 1532 1533 1534

	/*
	 * 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
1535
	 * CPU.
1536 1537 1538 1539 1540 1541
	 *
	 * 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 ]
	 */
1542
	if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
P
Peter Zijlstra 已提交
1543
		     !cpu_online(cpu)))
1544
		cpu = select_fallback_rq(task_cpu(p), p);
1545 1546

	return cpu;
1547
}
1548 1549 1550 1551 1552 1553

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1554

1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
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;
	}
}

1585 1586 1587 1588 1589 1590 1591 1592
#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
1593
#endif /* CONFIG_SMP */
1594

P
Peter Zijlstra 已提交
1595
static void
1596
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1597
{
1598
	struct rq *rq;
1599

1600 1601 1602 1603
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1604

1605 1606
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1607 1608
		__schedstat_inc(rq->ttwu_local);
		__schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1609 1610 1611
	} else {
		struct sched_domain *sd;

1612
		__schedstat_inc(p->se.statistics.nr_wakeups_remote);
1613
		rcu_read_lock();
1614
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1615
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1616
				__schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1617 1618 1619
				break;
			}
		}
1620
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1621
	}
1622 1623

	if (wake_flags & WF_MIGRATED)
1624
		__schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1625 1626
#endif /* CONFIG_SMP */

1627 1628
	__schedstat_inc(rq->ttwu_count);
	__schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1629 1630

	if (wake_flags & WF_SYNC)
1631
		__schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1632 1633
}

1634
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1635
{
T
Tejun Heo 已提交
1636
	activate_task(rq, p, en_flags);
1637
	p->on_rq = TASK_ON_RQ_QUEUED;
1638

I
Ingo Molnar 已提交
1639
	/* If a worker is waking up, notify the workqueue: */
1640 1641
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1642 1643
}

1644 1645 1646
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1647
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1648
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1649 1650 1651
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1652 1653
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1654
#ifdef CONFIG_SMP
1655 1656
	if (p->sched_class->task_woken) {
		/*
1657 1658
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1659
		 */
1660
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1661
		p->sched_class->task_woken(rq, p);
1662
		rq_repin_lock(rq, rf);
1663
	}
T
Tejun Heo 已提交
1664

1665
	if (rq->idle_stamp) {
1666
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1667
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1668

1669 1670 1671
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1672
			rq->avg_idle = max;
1673

T
Tejun Heo 已提交
1674 1675 1676 1677 1678
		rq->idle_stamp = 0;
	}
#endif
}

1679
static void
1680
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1681
		 struct rq_flags *rf)
1682
{
1683
	int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1684

1685 1686
	lockdep_assert_held(&rq->lock);

1687 1688 1689
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1690 1691

	if (wake_flags & WF_MIGRATED)
1692
		en_flags |= ENQUEUE_MIGRATED;
1693 1694
#endif

1695
	ttwu_activate(rq, p, en_flags);
1696
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1697 1698 1699 1700 1701 1702 1703 1704 1705 1706
}

/*
 * 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)
{
1707
	struct rq_flags rf;
1708 1709 1710
	struct rq *rq;
	int ret = 0;

1711
	rq = __task_rq_lock(p, &rf);
1712
	if (task_on_rq_queued(p)) {
1713 1714
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1715
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1716 1717
		ret = 1;
	}
1718
	__task_rq_unlock(rq, &rf);
1719 1720 1721 1722

	return ret;
}

1723
#ifdef CONFIG_SMP
1724
void sched_ttwu_pending(void)
1725 1726
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1727
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1728
	struct task_struct *p, *t;
1729
	struct rq_flags rf;
1730

1731 1732 1733
	if (!llist)
		return;

1734
	rq_lock_irqsave(rq, &rf);
1735
	update_rq_clock(rq);
1736

1737 1738
	llist_for_each_entry_safe(p, t, llist, wake_entry)
		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1739

1740
	rq_unlock_irqrestore(rq, &rf);
1741 1742 1743 1744
}

void scheduler_ipi(void)
{
1745 1746 1747 1748 1749
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1750
	preempt_fold_need_resched();
1751

1752
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768
		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
1769
	sched_ttwu_pending();
1770 1771 1772 1773

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1774
	if (unlikely(got_nohz_idle_kick())) {
1775
		this_rq()->idle_balance = 1;
1776
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1777
	}
1778
	irq_exit();
1779 1780
}

P
Peter Zijlstra 已提交
1781
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1782
{
1783 1784
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1785 1786
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1787 1788 1789 1790 1791 1792
	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);
	}
1793
}
1794

1795 1796 1797
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
1798
	struct rq_flags rf;
1799

1800 1801 1802 1803
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1804 1805 1806 1807

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
1808
		rq_lock_irqsave(rq, &rf);
1809 1810
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
I
Ingo Molnar 已提交
1811
		/* Else CPU is not idle, do nothing here: */
1812
		rq_unlock_irqrestore(rq, &rf);
1813
	}
1814 1815 1816

out:
	rcu_read_unlock();
1817 1818
}

1819
bool cpus_share_cache(int this_cpu, int that_cpu)
1820 1821 1822
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1823
#endif /* CONFIG_SMP */
1824

1825
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1826 1827
{
	struct rq *rq = cpu_rq(cpu);
1828
	struct rq_flags rf;
1829

1830
#if defined(CONFIG_SMP)
1831
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
I
Ingo Molnar 已提交
1832
		sched_clock_cpu(cpu); /* Sync clocks across CPUs */
P
Peter Zijlstra 已提交
1833
		ttwu_queue_remote(p, cpu, wake_flags);
1834 1835 1836 1837
		return;
	}
#endif

1838
	rq_lock(rq, &rf);
1839
	update_rq_clock(rq);
1840
	ttwu_do_activate(rq, p, wake_flags, &rf);
1841
	rq_unlock(rq, &rf);
T
Tejun Heo 已提交
1842 1843
}

1844 1845 1846 1847 1848 1849
/*
 * 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
1850 1851
 * migrates, all its activity on its old CPU [c0] happens-before any subsequent
 * execution on its new CPU [c1].
1852 1853 1854 1855 1856 1857 1858 1859 1860 1861
 *
 * 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
1862
 * Note: the CPU doing B need not be c0 or c1
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893
 *
 * 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)
1894
 *   2) smp_cond_load_acquire(!X->on_cpu)
1895 1896 1897 1898 1899 1900 1901 1902 1903 1904
 *
 * 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);
 *
1905
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930
 *                    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,
1931
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1932 1933 1934
 *
 */

T
Tejun Heo 已提交
1935
/**
L
Linus Torvalds 已提交
1936
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1937
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1938
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1939
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1940
 *
1941
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1942
 *
1943 1944 1945 1946 1947 1948 1949
 * 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
1950
 */
1951 1952
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1953 1954
{
	unsigned long flags;
1955
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1956

1957 1958 1959 1960 1961 1962
	/*
	 * 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.
	 */
1963
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1964
	smp_mb__after_spinlock();
P
Peter Zijlstra 已提交
1965
	if (!(p->state & state))
L
Linus Torvalds 已提交
1966 1967
		goto out;

1968 1969
	trace_sched_waking(p);

I
Ingo Molnar 已提交
1970 1971
	/* We're going to change ->state: */
	success = 1;
L
Linus Torvalds 已提交
1972 1973
	cpu = task_cpu(p);

1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995
	/*
	 * 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();
1996 1997
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
1998 1999

#ifdef CONFIG_SMP
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018
	/*
	 * 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
2019
	/*
I
Ingo Molnar 已提交
2020
	 * If the owning (remote) CPU is still in the middle of schedule() with
2021
	 * this task as prev, wait until its done referencing the task.
2022
	 *
2023
	 * Pairs with the smp_store_release() in finish_task().
2024 2025 2026
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2027
	 */
2028
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2029

2030
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2031
	p->state = TASK_WAKING;
2032

2033
	if (p->in_iowait) {
2034
		delayacct_blkio_end(p);
2035 2036 2037
		atomic_dec(&task_rq(p)->nr_iowait);
	}

2038
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2039 2040
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2041
		set_task_cpu(p, cpu);
2042
	}
2043 2044 2045 2046

#else /* CONFIG_SMP */

	if (p->in_iowait) {
2047
		delayacct_blkio_end(p);
2048 2049 2050
		atomic_dec(&task_rq(p)->nr_iowait);
	}

L
Linus Torvalds 已提交
2051 2052
#endif /* CONFIG_SMP */

2053
	ttwu_queue(p, cpu, wake_flags);
2054
stat:
2055
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2056
out:
2057
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2058 2059 2060 2061

	return success;
}

T
Tejun Heo 已提交
2062 2063 2064
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2065
 * @rf: request-queue flags for pinning
T
Tejun Heo 已提交
2066
 *
2067
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2068
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2069
 * the current task.
T
Tejun Heo 已提交
2070
 */
2071
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2072 2073 2074
{
	struct rq *rq = task_rq(p);

2075 2076 2077 2078
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2079 2080
	lockdep_assert_held(&rq->lock);

2081
	if (!raw_spin_trylock(&p->pi_lock)) {
2082 2083 2084 2085 2086 2087
		/*
		 * 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.
		 */
2088
		rq_unlock(rq, rf);
2089
		raw_spin_lock(&p->pi_lock);
2090
		rq_relock(rq, rf);
2091 2092
	}

T
Tejun Heo 已提交
2093
	if (!(p->state & TASK_NORMAL))
2094
		goto out;
T
Tejun Heo 已提交
2095

2096 2097
	trace_sched_waking(p);

2098 2099
	if (!task_on_rq_queued(p)) {
		if (p->in_iowait) {
2100
			delayacct_blkio_end(p);
2101 2102
			atomic_dec(&rq->nr_iowait);
		}
2103
		ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2104
	}
P
Peter Zijlstra 已提交
2105

2106
	ttwu_do_wakeup(rq, p, 0, rf);
2107
	ttwu_stat(p, smp_processor_id(), 0);
2108 2109
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2110 2111
}

2112 2113 2114 2115 2116
/**
 * 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
2117 2118 2119
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2120 2121 2122 2123
 *
 * 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.
 */
2124
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2125
{
2126
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2127 2128 2129
}
EXPORT_SYMBOL(wake_up_process);

2130
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2131 2132 2133 2134 2135 2136 2137
{
	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
2138 2139 2140
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2141
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2142
{
P
Peter Zijlstra 已提交
2143 2144 2145
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2146 2147
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2148
	p->se.prev_sum_exec_runtime	= 0;
2149
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2150
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2151
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2152

2153 2154 2155 2156
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2157
#ifdef CONFIG_SCHEDSTATS
2158
	/* Even if schedstat is disabled, there should not be garbage */
2159
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2160
#endif
N
Nick Piggin 已提交
2161

2162
	RB_CLEAR_NODE(&p->dl.rb_node);
2163
	init_dl_task_timer(&p->dl);
2164
	init_dl_inactive_task_timer(&p->dl);
2165
	__dl_clear_params(p);
2166

P
Peter Zijlstra 已提交
2167
	INIT_LIST_HEAD(&p->rt.run_list);
2168 2169 2170 2171
	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
2172

2173 2174 2175
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2176 2177 2178

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2179
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2180 2181 2182
		p->mm->numa_scan_seq = 0;
	}

2183 2184 2185 2186 2187
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2188 2189
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2190
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2191
	p->numa_work.next = &p->numa_work;
2192
	p->numa_faults = NULL;
2193 2194
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2195 2196

	p->numa_group = NULL;
2197
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2198 2199
}

2200 2201
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2202
#ifdef CONFIG_NUMA_BALANCING
2203

2204 2205 2206
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2207
		static_branch_enable(&sched_numa_balancing);
2208
	else
2209
		static_branch_disable(&sched_numa_balancing);
2210
}
2211 2212 2213 2214 2215 2216 2217

#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;
2218
	int state = static_branch_likely(&sched_numa_balancing);
2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233

	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
2234

2235 2236
#ifdef CONFIG_SCHEDSTATS

2237
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2238
static bool __initdata __sched_schedstats = false;
2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261

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;

2262 2263 2264 2265 2266
	/*
	 * 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.
	 */
2267
	if (!strcmp(str, "enable")) {
2268
		__sched_schedstats = true;
2269 2270
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2271
		__sched_schedstats = false;
2272 2273 2274 2275 2276 2277 2278 2279 2280 2281
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2282 2283 2284 2285 2286
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306
#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;
}
2307 2308 2309 2310
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2311 2312 2313 2314

/*
 * fork()/clone()-time setup:
 */
2315
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2316
{
2317
	unsigned long flags;
I
Ingo Molnar 已提交
2318 2319
	int cpu = get_cpu();

2320
	__sched_fork(clone_flags, p);
2321
	/*
2322
	 * We mark the process as NEW here. This guarantees that
2323 2324 2325
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2326
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2327

2328 2329 2330 2331 2332
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2333 2334 2335 2336
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2337
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2338
			p->policy = SCHED_NORMAL;
2339
			p->static_prio = NICE_TO_PRIO(0);
2340 2341 2342 2343 2344
			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);
2345
		set_load_weight(p, false);
2346

2347 2348 2349 2350 2351 2352
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2353

2354 2355 2356 2357 2358 2359
	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
2360
		p->sched_class = &fair_sched_class;
2361
	}
2362

2363
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2364

2365 2366 2367 2368 2369 2370 2371
	/*
	 * 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.
	 */
2372
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2373
	/*
I
Ingo Molnar 已提交
2374
	 * We're setting the CPU for the first time, we don't migrate,
2375 2376 2377 2378 2379
	 * so use __set_task_cpu().
	 */
	__set_task_cpu(p, cpu);
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2380
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2381

2382
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2383
	if (likely(sched_info_on()))
2384
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2385
#endif
P
Peter Zijlstra 已提交
2386 2387
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2388
#endif
2389
	init_task_preempt_count(p);
2390
#ifdef CONFIG_SMP
2391
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2392
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2393
#endif
2394

N
Nick Piggin 已提交
2395
	put_cpu();
2396
	return 0;
L
Linus Torvalds 已提交
2397 2398
}

2399 2400 2401
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
2402
		return BW_UNIT;
2403 2404 2405 2406 2407 2408 2409 2410 2411

	/*
	 * 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;

2412
	return div64_u64(runtime << BW_SHIFT, period);
2413 2414
}

L
Linus Torvalds 已提交
2415 2416 2417 2418 2419 2420 2421
/*
 * 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.
 */
2422
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2423
{
2424
	struct rq_flags rf;
I
Ingo Molnar 已提交
2425
	struct rq *rq;
2426

2427
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2428
	p->state = TASK_RUNNING;
2429 2430 2431 2432
#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
2433
	 *  - any previously selected CPU might disappear through hotplug
2434 2435 2436
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2437
	 */
2438
	p->recent_used_cpu = task_cpu(p);
2439
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2440
#endif
2441
	rq = __task_rq_lock(p, &rf);
2442
	update_rq_clock(rq);
2443
	post_init_entity_util_avg(&p->se);
2444

2445
	activate_task(rq, p, ENQUEUE_NOCLOCK);
2446
	p->on_rq = TASK_ON_RQ_QUEUED;
2447
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2448
	check_preempt_curr(rq, p, WF_FORK);
2449
#ifdef CONFIG_SMP
2450 2451 2452 2453 2454
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2455
		rq_unpin_lock(rq, &rf);
2456
		p->sched_class->task_woken(rq, p);
2457
		rq_repin_lock(rq, &rf);
2458
	}
2459
#endif
2460
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2461 2462
}

2463 2464
#ifdef CONFIG_PREEMPT_NOTIFIERS

2465 2466
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478
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);

2479
/**
2480
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2481
 * @notifier: notifier struct to register
2482 2483 2484
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2485 2486 2487
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2488 2489 2490 2491 2492 2493
	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
2494
 * @notifier: notifier struct to unregister
2495
 *
2496
 * This is *not* safe to call from within a preemption notifier.
2497 2498 2499 2500 2501 2502 2503
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2504
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2505 2506 2507
{
	struct preempt_notifier *notifier;

2508
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2509 2510 2511
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2512 2513 2514 2515 2516 2517
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);
}

2518
static void
2519 2520
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2521 2522 2523
{
	struct preempt_notifier *notifier;

2524
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2525 2526 2527
		notifier->ops->sched_out(notifier, next);
}

2528 2529 2530 2531 2532 2533 2534 2535
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);
}

2536
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2537

2538
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2539 2540 2541
{
}

2542
static inline void
2543 2544 2545 2546 2547
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2548
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2549

2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577
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
}

2578 2579
static inline void
prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
2580
{
2581 2582 2583 2584 2585 2586 2587 2588
	/*
	 * 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:
	 */
	rq_unpin_lock(rq, rf);
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
2589 2590
#ifdef CONFIG_DEBUG_SPINLOCK
	/* this is a valid case when another task releases the spinlock */
2591
	rq->lock.owner = next;
2592
#endif
2593 2594 2595 2596
}

static inline void finish_lock_switch(struct rq *rq)
{
2597 2598 2599 2600 2601 2602 2603 2604 2605
	/*
	 * 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);
}

2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617
/*
 * NOP if the arch has not defined these:
 */

#ifndef prepare_arch_switch
# define prepare_arch_switch(next)	do { } while (0)
#endif

#ifndef finish_arch_post_lock_switch
# define finish_arch_post_lock_switch()	do { } while (0)
#endif

2618 2619 2620
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2621
 * @prev: the current task that is being switched out
2622 2623 2624 2625 2626 2627 2628 2629 2630
 * @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.
 */
2631 2632 2633
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2634
{
2635
	sched_info_switch(rq, prev, next);
2636
	perf_event_task_sched_out(prev, next);
2637
	fire_sched_out_preempt_notifiers(prev, next);
2638
	prepare_task(next);
2639 2640 2641
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2642 2643 2644 2645
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2646 2647 2648 2649
 * 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
2650 2651
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2652
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2653 2654
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2655 2656 2657 2658 2659
 *
 * 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
2660
 */
2661
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2662 2663
	__releases(rq->lock)
{
2664
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2665
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2666
	long prev_state;
L
Linus Torvalds 已提交
2667

2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678
	/*
	 * 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.
	 */
2679 2680 2681 2682
	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);
2683

L
Linus Torvalds 已提交
2684 2685 2686 2687
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2688
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2689 2690
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2691 2692
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
2693
	 * finish_task), otherwise a concurrent wakeup can get prev
2694 2695
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2696
	 */
O
Oleg Nesterov 已提交
2697
	prev_state = prev->state;
2698
	vtime_task_switch(prev);
2699
	perf_event_task_sched_in(prev, current);
2700 2701
	finish_task(prev);
	finish_lock_switch(rq);
2702
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2703

2704
	fire_sched_in_preempt_notifiers(current);
2705
	/*
2706 2707 2708 2709 2710 2711 2712 2713 2714 2715
	 * 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.
2716
	 */
2717 2718
	if (mm) {
		membarrier_mm_sync_core_before_usermode(mm);
L
Linus Torvalds 已提交
2719
		mmdrop(mm);
2720
	}
2721
	if (unlikely(prev_state == TASK_DEAD)) {
2722 2723 2724
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2725 2726 2727
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2728
		 */
2729
		kprobe_flush_task(prev);
2730 2731 2732 2733

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

L
Linus Torvalds 已提交
2734
		put_task_struct(prev);
2735
	}
2736

2737
	tick_nohz_task_switch();
2738
	return rq;
L
Linus Torvalds 已提交
2739 2740
}

2741 2742 2743
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2744
static void __balance_callback(struct rq *rq)
2745
{
2746 2747 2748
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2749

2750 2751 2752 2753 2754 2755 2756 2757
	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;
2758

2759
		func(rq);
2760
	}
2761 2762 2763 2764 2765 2766 2767
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2768 2769 2770
}

#else
2771

2772
static inline void balance_callback(struct rq *rq)
2773
{
L
Linus Torvalds 已提交
2774 2775
}

2776 2777
#endif

L
Linus Torvalds 已提交
2778 2779 2780 2781
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2782
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2783 2784
	__releases(rq->lock)
{
2785
	struct rq *rq;
2786

2787 2788 2789 2790 2791 2792 2793 2794 2795
	/*
	 * 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).
	 */

2796
	rq = finish_task_switch(prev);
2797
	balance_callback(rq);
2798
	preempt_enable();
2799

L
Linus Torvalds 已提交
2800
	if (current->set_child_tid)
2801
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2802 2803 2804
}

/*
2805
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2806
 */
2807
static __always_inline struct rq *
2808
context_switch(struct rq *rq, struct task_struct *prev,
2809
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2810
{
I
Ingo Molnar 已提交
2811
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2812

2813
	prepare_task_switch(rq, prev, next);
2814

I
Ingo Molnar 已提交
2815 2816
	mm = next->mm;
	oldmm = prev->active_mm;
2817 2818 2819 2820 2821
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2822
	arch_start_context_switch(prev);
2823

2824 2825 2826 2827 2828 2829 2830
	/*
	 * 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.
	 */
2831
	if (!mm) {
L
Linus Torvalds 已提交
2832
		next->active_mm = oldmm;
V
Vegard Nossum 已提交
2833
		mmgrab(oldmm);
L
Linus Torvalds 已提交
2834 2835
		enter_lazy_tlb(oldmm, next);
	} else
2836
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2837

2838
	if (!prev->mm) {
L
Linus Torvalds 已提交
2839 2840 2841
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2842

2843
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2844

2845
	prepare_lock_switch(rq, next, rf);
L
Linus Torvalds 已提交
2846 2847 2848

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2849
	barrier();
2850 2851

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2852 2853 2854
}

/*
2855
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2856 2857
 *
 * externally visible scheduler statistics: current number of runnable
2858
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2859 2860 2861 2862 2863 2864 2865 2866 2867
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2868
}
L
Linus Torvalds 已提交
2869

2870
/*
I
Ingo Molnar 已提交
2871
 * Check if only the current task is running on the CPU.
2872 2873 2874 2875 2876 2877 2878 2879 2880 2881
 *
 * 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)
2882 2883 2884
 */
bool single_task_running(void)
{
2885
	return raw_rq()->nr_running == 1;
2886 2887 2888
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2889
unsigned long long nr_context_switches(void)
2890
{
2891 2892
	int i;
	unsigned long long sum = 0;
2893

2894
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2895
		sum += cpu_rq(i)->nr_switches;
2896

L
Linus Torvalds 已提交
2897 2898
	return sum;
}
2899

2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929
/*
 * 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
2930 2931 2932
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2933

2934
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2935
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2936

L
Linus Torvalds 已提交
2937 2938
	return sum;
}
2939

2940 2941 2942 2943 2944 2945 2946
/*
 * 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.
 */

2947
unsigned long nr_iowait_cpu(int cpu)
2948
{
2949
	struct rq *this = cpu_rq(cpu);
2950 2951
	return atomic_read(&this->nr_iowait);
}
2952

2953 2954
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2955 2956 2957
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2958 2959
}

I
Ingo Molnar 已提交
2960
#ifdef CONFIG_SMP
2961

2962
/*
P
Peter Zijlstra 已提交
2963 2964
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2965
 */
P
Peter Zijlstra 已提交
2966
void sched_exec(void)
2967
{
P
Peter Zijlstra 已提交
2968
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2969
	unsigned long flags;
2970
	int dest_cpu;
2971

2972
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2973
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2974 2975
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2976

2977
	if (likely(cpu_active(dest_cpu))) {
2978
		struct migration_arg arg = { p, dest_cpu };
2979

2980 2981
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2982 2983
		return;
	}
2984
unlock:
2985
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2986
}
I
Ingo Molnar 已提交
2987

L
Linus Torvalds 已提交
2988 2989 2990
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2991
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2992 2993

EXPORT_PER_CPU_SYMBOL(kstat);
2994
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2995

2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012
/*
 * 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);
}

3013 3014 3015 3016 3017 3018 3019
/*
 * 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)
{
3020
	struct rq_flags rf;
3021
	struct rq *rq;
3022
	u64 ns;
3023

3024 3025
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
3026
	 * 64-bit doesn't need locks to atomically read a 64-bit value.
3027 3028 3029
	 * 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
3030 3031
	 * 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
3032
	 * indistinguishable from the read occurring a few cycles earlier.
3033 3034
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3035
	 */
3036
	if (!p->on_cpu || !task_on_rq_queued(p))
3037 3038 3039
		return p->se.sum_exec_runtime;
#endif

3040
	rq = task_rq_lock(p, &rf);
3041 3042 3043 3044 3045 3046
	/*
	 * 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)) {
3047
		prefetch_curr_exec_start(p);
3048 3049 3050 3051
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3052
	task_rq_unlock(rq, p, &rf);
3053 3054 3055

	return ns;
}
3056

3057 3058 3059 3060 3061 3062 3063 3064
/*
 * 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
3065
	struct task_struct *curr = rq->curr;
3066
	struct rq_flags rf;
3067 3068

	sched_clock_tick();
I
Ingo Molnar 已提交
3069

3070 3071
	rq_lock(rq, &rf);

3072
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3073
	curr->sched_class->task_tick(rq, curr, 0);
3074
	cpu_load_update_active(rq);
3075
	calc_global_load_tick(rq);
3076 3077

	rq_unlock(rq, &rf);
3078

3079
	perf_event_task_tick();
3080

3081
#ifdef CONFIG_SMP
3082
	rq->idle_balance = idle_cpu(cpu);
3083
	trigger_load_balance(rq);
3084
#endif
L
Linus Torvalds 已提交
3085 3086
}

3087
#ifdef CONFIG_NO_HZ_FULL
3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177

struct tick_work {
	int			cpu;
	struct delayed_work	work;
};

static struct tick_work __percpu *tick_work_cpu;

static void sched_tick_remote(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct tick_work *twork = container_of(dwork, struct tick_work, work);
	int cpu = twork->cpu;
	struct rq *rq = cpu_rq(cpu);
	struct rq_flags rf;

	/*
	 * Handle the tick only if it appears the remote CPU is running in full
	 * dynticks mode. The check is racy by nature, but missing a tick or
	 * having one too much is no big deal because the scheduler tick updates
	 * statistics and checks timeslices in a time-independent way, regardless
	 * of when exactly it is running.
	 */
	if (!idle_cpu(cpu) && tick_nohz_tick_stopped_cpu(cpu)) {
		struct task_struct *curr;
		u64 delta;

		rq_lock_irq(rq, &rf);
		update_rq_clock(rq);
		curr = rq->curr;
		delta = rq_clock_task(rq) - curr->se.exec_start;

		/*
		 * Make sure the next tick runs within a reasonable
		 * amount of time.
		 */
		WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
		curr->sched_class->task_tick(rq, curr, 0);
		rq_unlock_irq(rq, &rf);
	}

	/*
	 * Run the remote tick once per second (1Hz). This arbitrary
	 * frequency is large enough to avoid overload but short enough
	 * to keep scheduler internal stats reasonably up to date.
	 */
	queue_delayed_work(system_unbound_wq, dwork, HZ);
}

static void sched_tick_start(int cpu)
{
	struct tick_work *twork;

	if (housekeeping_cpu(cpu, HK_FLAG_TICK))
		return;

	WARN_ON_ONCE(!tick_work_cpu);

	twork = per_cpu_ptr(tick_work_cpu, cpu);
	twork->cpu = cpu;
	INIT_DELAYED_WORK(&twork->work, sched_tick_remote);
	queue_delayed_work(system_unbound_wq, &twork->work, HZ);
}

#ifdef CONFIG_HOTPLUG_CPU
static void sched_tick_stop(int cpu)
{
	struct tick_work *twork;

	if (housekeeping_cpu(cpu, HK_FLAG_TICK))
		return;

	WARN_ON_ONCE(!tick_work_cpu);

	twork = per_cpu_ptr(tick_work_cpu, cpu);
	cancel_delayed_work_sync(&twork->work);
}
#endif /* CONFIG_HOTPLUG_CPU */

int __init sched_tick_offload_init(void)
{
	tick_work_cpu = alloc_percpu(struct tick_work);
	BUG_ON(!tick_work_cpu);

	return 0;
}

#else /* !CONFIG_NO_HZ_FULL */
static inline void sched_tick_start(int cpu) { }
static inline void sched_tick_stop(int cpu) { }
3178
#endif
L
Linus Torvalds 已提交
3179

3180 3181
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195
/*
 * 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);
	}
}
3196

3197
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3198
{
3199
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3200 3201 3202
	/*
	 * Underflow?
	 */
3203 3204
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3205
#endif
3206
	__preempt_count_add(val);
3207
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3208 3209 3210
	/*
	 * Spinlock count overflowing soon?
	 */
3211 3212
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3213
#endif
3214
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3215
}
3216
EXPORT_SYMBOL(preempt_count_add);
3217
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3218

3219 3220 3221 3222 3223 3224 3225 3226 3227 3228
/*
 * 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());
}

3229
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3230
{
3231
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3232 3233 3234
	/*
	 * Underflow?
	 */
3235
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3236
		return;
L
Linus Torvalds 已提交
3237 3238 3239
	/*
	 * Is the spinlock portion underflowing?
	 */
3240 3241 3242
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3243
#endif
3244

3245
	preempt_latency_stop(val);
3246
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3247
}
3248
EXPORT_SYMBOL(preempt_count_sub);
3249
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3250

3251 3252 3253
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3254 3255
#endif

3256 3257 3258 3259 3260 3261 3262 3263 3264
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
3265
/*
I
Ingo Molnar 已提交
3266
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3267
 */
I
Ingo Molnar 已提交
3268
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3269
{
3270 3271 3272
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3273 3274 3275
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3276 3277
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3278

I
Ingo Molnar 已提交
3279
	debug_show_held_locks(prev);
3280
	print_modules();
I
Ingo Molnar 已提交
3281 3282
	if (irqs_disabled())
		print_irqtrace_events(prev);
3283 3284
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3285
		pr_err("Preemption disabled at:");
3286
		print_ip_sym(preempt_disable_ip);
3287 3288
		pr_cont("\n");
	}
3289 3290 3291
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3292
	dump_stack();
3293
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3294
}
L
Linus Torvalds 已提交
3295

I
Ingo Molnar 已提交
3296 3297 3298 3299 3300
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3301
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3302 3303
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3304
#endif
3305

3306
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3307
		__schedule_bug(prev);
3308 3309
		preempt_count_set(PREEMPT_DISABLED);
	}
3310
	rcu_sleep_check();
I
Ingo Molnar 已提交
3311

L
Linus Torvalds 已提交
3312 3313
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3314
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3315 3316 3317 3318 3319 3320
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3321
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3322
{
3323
	const struct sched_class *class;
I
Ingo Molnar 已提交
3324
	struct task_struct *p;
L
Linus Torvalds 已提交
3325 3326

	/*
3327 3328 3329 3330
	 * 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
3331
	 */
3332 3333 3334 3335
	if (likely((prev->sched_class == &idle_sched_class ||
		    prev->sched_class == &fair_sched_class) &&
		   rq->nr_running == rq->cfs.h_nr_running)) {

3336
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3337 3338 3339
		if (unlikely(p == RETRY_TASK))
			goto again;

I
Ingo Molnar 已提交
3340
		/* Assumes fair_sched_class->next == idle_sched_class */
3341
		if (unlikely(!p))
3342
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3343 3344

		return p;
L
Linus Torvalds 已提交
3345 3346
	}

3347
again:
3348
	for_each_class(class) {
3349
		p = class->pick_next_task(rq, prev, rf);
3350 3351 3352
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3353
			return p;
3354
		}
I
Ingo Molnar 已提交
3355
	}
3356

I
Ingo Molnar 已提交
3357 3358
	/* The idle class should always have a runnable task: */
	BUG();
I
Ingo Molnar 已提交
3359
}
L
Linus Torvalds 已提交
3360

I
Ingo Molnar 已提交
3361
/*
3362
 * __schedule() is the main scheduler function.
3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396
 *
 * 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
3397
 *
3398
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3399
 */
3400
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3401 3402
{
	struct task_struct *prev, *next;
3403
	unsigned long *switch_count;
3404
	struct rq_flags rf;
I
Ingo Molnar 已提交
3405
	struct rq *rq;
3406
	int cpu;
I
Ingo Molnar 已提交
3407 3408 3409 3410 3411 3412

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3413

3414
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3415
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3416

3417
	local_irq_disable();
3418
	rcu_note_context_switch(preempt);
3419

3420 3421 3422 3423
	/*
	 * 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().
3424 3425 3426
	 *
	 * The membarrier system call requires a full memory barrier
	 * after coming from user-space, before storing to rq->curr.
3427
	 */
3428
	rq_lock(rq, &rf);
3429
	smp_mb__after_spinlock();
L
Linus Torvalds 已提交
3430

I
Ingo Molnar 已提交
3431 3432
	/* Promote REQ to ACT */
	rq->clock_update_flags <<= 1;
3433
	update_rq_clock(rq);
3434

3435
	switch_count = &prev->nivcsw;
3436
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3437
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3438
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3439
		} else {
3440
			deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3441 3442
			prev->on_rq = 0;

3443 3444 3445 3446 3447
			if (prev->in_iowait) {
				atomic_inc(&rq->nr_iowait);
				delayacct_blkio_start();
			}

T
Tejun Heo 已提交
3448
			/*
3449 3450 3451
			 * 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
3452 3453 3454 3455
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3456
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3457
				if (to_wakeup)
3458
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3459 3460
			}
		}
I
Ingo Molnar 已提交
3461
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3462 3463
	}

3464
	next = pick_next_task(rq, prev, &rf);
3465
	clear_tsk_need_resched(prev);
3466
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3467 3468 3469 3470

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
3471 3472 3473
		/*
		 * The membarrier system call requires each architecture
		 * to have a full memory barrier after updating
3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
		 * 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),
3484
		 */
L
Linus Torvalds 已提交
3485 3486
		++*switch_count;

3487
		trace_sched_switch(preempt, prev, next);
I
Ingo Molnar 已提交
3488 3489 3490

		/* Also unlocks the rq: */
		rq = context_switch(rq, prev, next, &rf);
3491
	} else {
3492
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3493
		rq_unlock_irq(rq, &rf);
3494
	}
L
Linus Torvalds 已提交
3495

3496
	balance_callback(rq);
L
Linus Torvalds 已提交
3497
}
3498

3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512
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()
	 */
3513 3514
	raw_spin_lock_irq(&current->pi_lock);
	raw_spin_unlock_irq(&current->pi_lock);
3515

I
Ingo Molnar 已提交
3516
	/* Causes final put_task_struct in finish_task_switch(): */
3517
	__set_current_state(TASK_DEAD);
I
Ingo Molnar 已提交
3518 3519 3520 3521

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

3522 3523
	__schedule(false);
	BUG();
I
Ingo Molnar 已提交
3524 3525

	/* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3526
	for (;;)
I
Ingo Molnar 已提交
3527
		cpu_relax();
3528 3529
}

3530 3531
static inline void sched_submit_work(struct task_struct *tsk)
{
3532
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3533 3534 3535 3536 3537 3538 3539 3540 3541
		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);
}

3542
asmlinkage __visible void __sched schedule(void)
3543
{
3544 3545 3546
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3547
	do {
3548
		preempt_disable();
3549
		__schedule(false);
3550
		sched_preempt_enable_no_resched();
3551
	} while (need_resched());
3552
}
L
Linus Torvalds 已提交
3553 3554
EXPORT_SYMBOL(schedule);

3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579
/*
 * 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());
}

3580
#ifdef CONFIG_CONTEXT_TRACKING
3581
asmlinkage __visible void __sched schedule_user(void)
3582 3583 3584 3585 3586 3587
{
	/*
	 * 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.
3588 3589
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3590
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3591
	 * too frequently to make sense yet.
3592
	 */
3593
	enum ctx_state prev_state = exception_enter();
3594
	schedule();
3595
	exception_exit(prev_state);
3596 3597 3598
}
#endif

3599 3600 3601 3602 3603 3604 3605
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3606
	sched_preempt_enable_no_resched();
3607 3608 3609 3610
	schedule();
	preempt_disable();
}

3611
static void __sched notrace preempt_schedule_common(void)
3612 3613
{
	do {
3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626
		/*
		 * 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.
		 */
3627
		preempt_disable_notrace();
3628
		preempt_latency_start(1);
3629
		__schedule(true);
3630
		preempt_latency_stop(1);
3631
		preempt_enable_no_resched_notrace();
3632 3633 3634 3635 3636 3637 3638 3639

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

L
Linus Torvalds 已提交
3640 3641
#ifdef CONFIG_PREEMPT
/*
3642
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3643
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3644 3645
 * occur there and call schedule directly.
 */
3646
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3647 3648 3649
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3650
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3651
	 */
3652
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3653 3654
		return;

3655
	preempt_schedule_common();
L
Linus Torvalds 已提交
3656
}
3657
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3658
EXPORT_SYMBOL(preempt_schedule);
3659 3660

/**
3661
 * preempt_schedule_notrace - preempt_schedule called by tracing
3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
 *
 * 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.
 */
3674
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3675 3676 3677 3678 3679 3680 3681
{
	enum ctx_state prev_ctx;

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

	do {
3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694
		/*
		 * 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.
		 */
3695
		preempt_disable_notrace();
3696
		preempt_latency_start(1);
3697 3698 3699 3700 3701 3702
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3703
		__schedule(true);
3704 3705
		exception_exit(prev_ctx);

3706
		preempt_latency_stop(1);
3707
		preempt_enable_no_resched_notrace();
3708 3709
	} while (need_resched());
}
3710
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3711

3712
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3713 3714

/*
3715
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3716 3717 3718 3719
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3720
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3721
{
3722
	enum ctx_state prev_state;
3723

3724
	/* Catch callers which need to be fixed */
3725
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3726

3727 3728
	prev_state = exception_enter();

3729
	do {
3730
		preempt_disable();
3731
		local_irq_enable();
3732
		__schedule(true);
3733
		local_irq_disable();
3734
		sched_preempt_enable_no_resched();
3735
	} while (need_resched());
3736 3737

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3738 3739
}

3740
int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3741
			  void *key)
L
Linus Torvalds 已提交
3742
{
P
Peter Zijlstra 已提交
3743
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3744 3745 3746
}
EXPORT_SYMBOL(default_wake_function);

3747 3748
#ifdef CONFIG_RT_MUTEXES

3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763
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);
}

3764 3765
/*
 * rt_mutex_setprio - set the current priority of a task
3766 3767
 * @p: task to boost
 * @pi_task: donor task
3768 3769 3770 3771
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3772 3773
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3774
 */
3775
void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3776
{
3777
	int prio, oldprio, queued, running, queue_flag =
3778
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3779
	const struct sched_class *prev_class;
3780 3781
	struct rq_flags rf;
	struct rq *rq;
3782

3783 3784 3785 3786 3787 3788 3789 3790
	/* 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;
3791

3792
	rq = __task_rq_lock(p, &rf);
3793
	update_rq_clock(rq);
3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810
	/*
	 * 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;
3811

3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829
	/*
	 * 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;
	}

3830
	trace_sched_pi_setprio(p, pi_task);
3831
	oldprio = p->prio;
3832 3833 3834 3835

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

3836
	prev_class = p->sched_class;
3837
	queued = task_on_rq_queued(p);
3838
	running = task_current(rq, p);
3839
	if (queued)
3840
		dequeue_task(rq, p, queue_flag);
3841
	if (running)
3842
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3843

3844 3845 3846 3847 3848 3849 3850 3851 3852 3853
	/*
	 * 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)) {
3854 3855
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3856
			p->dl.dl_boosted = 1;
3857
			queue_flag |= ENQUEUE_REPLENISH;
3858 3859
		} else
			p->dl.dl_boosted = 0;
3860
		p->sched_class = &dl_sched_class;
3861 3862 3863 3864
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3865
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3866
		p->sched_class = &rt_sched_class;
3867 3868 3869
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3870 3871
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3872
		p->sched_class = &fair_sched_class;
3873
	}
I
Ingo Molnar 已提交
3874

3875 3876
	p->prio = prio;

3877
	if (queued)
3878
		enqueue_task(rq, p, queue_flag);
3879
	if (running)
3880
		set_curr_task(rq, p);
3881

P
Peter Zijlstra 已提交
3882
	check_class_changed(rq, p, prev_class, oldprio);
3883
out_unlock:
I
Ingo Molnar 已提交
3884 3885
	/* Avoid rq from going away on us: */
	preempt_disable();
3886
	__task_rq_unlock(rq, &rf);
3887 3888 3889

	balance_callback(rq);
	preempt_enable();
3890
}
3891 3892 3893 3894 3895
#else
static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	return prio;
}
3896
#endif
3897

3898
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3899
{
P
Peter Zijlstra 已提交
3900 3901
	bool queued, running;
	int old_prio, delta;
3902
	struct rq_flags rf;
3903
	struct rq *rq;
L
Linus Torvalds 已提交
3904

3905
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3906 3907 3908 3909 3910
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3911
	rq = task_rq_lock(p, &rf);
3912 3913
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3914 3915 3916 3917
	/*
	 * 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
3918
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3919
	 */
3920
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3921 3922 3923
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3924
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3925
	running = task_current(rq, p);
3926
	if (queued)
3927
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
3928 3929
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3930 3931

	p->static_prio = NICE_TO_PRIO(nice);
3932
	set_load_weight(p, true);
3933 3934 3935
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3936

3937
	if (queued) {
3938
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
L
Linus Torvalds 已提交
3939
		/*
3940 3941
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3942
		 */
3943
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3944
			resched_curr(rq);
L
Linus Torvalds 已提交
3945
	}
P
Peter Zijlstra 已提交
3946 3947
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3948
out_unlock:
3949
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3950 3951 3952
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3953 3954 3955 3956 3957
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3958
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3959
{
I
Ingo Molnar 已提交
3960
	/* Convert nice value [19,-20] to rlimit style value [1,40]: */
3961
	int nice_rlim = nice_to_rlimit(nice);
3962

3963
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3964 3965 3966
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3967 3968 3969 3970 3971 3972 3973 3974 3975
#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.
 */
3976
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3977
{
3978
	long nice, retval;
L
Linus Torvalds 已提交
3979 3980 3981 3982 3983 3984

	/*
	 * 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.
	 */
3985
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3986
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3987

3988
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3989 3990 3991
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005
	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.
 *
4006
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
4007 4008 4009
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
4010
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
4011 4012 4013 4014 4015
{
	return p->prio - MAX_RT_PRIO;
}

/**
I
Ingo Molnar 已提交
4016
 * idle_cpu - is a given CPU idle currently?
L
Linus Torvalds 已提交
4017
 * @cpu: the processor in question.
4018 4019
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
4020 4021 4022
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036
	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
4037 4038 4039
}

/**
I
Ingo Molnar 已提交
4040
 * idle_task - return the idle task for a given CPU.
L
Linus Torvalds 已提交
4041
 * @cpu: the processor in question.
4042
 *
I
Ingo Molnar 已提交
4043
 * Return: The idle task for the CPU @cpu.
L
Linus Torvalds 已提交
4044
 */
4045
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
4046 4047 4048 4049 4050 4051 4052
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
4053 4054
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
4055
 */
A
Alexey Dobriyan 已提交
4056
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
4057
{
4058
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
4059 4060
}

4061 4062 4063 4064 4065 4066
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

4067 4068
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
4069
{
4070 4071
	int policy = attr->sched_policy;

4072
	if (policy == SETPARAM_POLICY)
4073 4074
		policy = p->policy;

L
Linus Torvalds 已提交
4075
	p->policy = policy;
4076

4077 4078
	if (dl_policy(policy))
		__setparam_dl(p, attr);
4079
	else if (fair_policy(policy))
4080 4081
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

4082 4083 4084 4085 4086 4087
	/*
	 * __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;
4088
	p->normal_prio = normal_prio(p);
4089
	set_load_weight(p, true);
4090
}
4091

4092 4093
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
4094
			   const struct sched_attr *attr, bool keep_boost)
4095 4096
{
	__setscheduler_params(p, attr);
4097

4098
	/*
4099 4100
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
4101
	 */
4102
	p->prio = normal_prio(p);
4103
	if (keep_boost)
4104
		p->prio = rt_effective_prio(p, p->prio);
4105

4106 4107 4108
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
4109 4110 4111
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
4112
}
4113

4114
/*
I
Ingo Molnar 已提交
4115
 * Check the target process has a UID that matches the current process's:
4116 4117 4118 4119 4120 4121 4122 4123
 */
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);
4124 4125
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4126 4127 4128 4129
	rcu_read_unlock();
	return match;
}

4130 4131
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4132
				bool user, bool pi)
L
Linus Torvalds 已提交
4133
{
4134 4135
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4136
	int retval, oldprio, oldpolicy = -1, queued, running;
4137
	int new_effective_prio, policy = attr->sched_policy;
4138
	const struct sched_class *prev_class;
4139
	struct rq_flags rf;
4140
	int reset_on_fork;
4141
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4142
	struct rq *rq;
L
Linus Torvalds 已提交
4143

4144 4145
	/* The pi code expects interrupts enabled */
	BUG_ON(pi && in_interrupt());
L
Linus Torvalds 已提交
4146
recheck:
I
Ingo Molnar 已提交
4147
	/* Double check policy once rq lock held: */
4148 4149
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4150
		policy = oldpolicy = p->policy;
4151
	} else {
4152
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4153

4154
		if (!valid_policy(policy))
4155 4156 4157
			return -EINVAL;
	}

4158
	if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4159 4160
		return -EINVAL;

L
Linus Torvalds 已提交
4161 4162
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4163 4164
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4165
	 */
4166
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4167
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4168
		return -EINVAL;
4169 4170
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4171 4172
		return -EINVAL;

4173 4174 4175
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4176
	if (user && !capable(CAP_SYS_NICE)) {
4177
		if (fair_policy(policy)) {
4178
			if (attr->sched_nice < task_nice(p) &&
4179
			    !can_nice(p, attr->sched_nice))
4180 4181 4182
				return -EPERM;
		}

4183
		if (rt_policy(policy)) {
4184 4185
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4186

I
Ingo Molnar 已提交
4187
			/* Can't set/change the rt policy: */
4188 4189 4190
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

I
Ingo Molnar 已提交
4191
			/* Can't increase priority: */
4192 4193
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4194 4195
				return -EPERM;
		}
4196

4197 4198 4199 4200 4201 4202 4203 4204 4205
		 /*
		  * 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
4206
		/*
4207 4208
		 * 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
4209
		 */
4210
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4211
			if (!can_nice(p, task_nice(p)))
4212 4213
				return -EPERM;
		}
4214

I
Ingo Molnar 已提交
4215
		/* Can't change other user's priorities: */
4216
		if (!check_same_owner(p))
4217
			return -EPERM;
4218

I
Ingo Molnar 已提交
4219
		/* Normal users shall not reset the sched_reset_on_fork flag: */
4220 4221
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4222
	}
L
Linus Torvalds 已提交
4223

4224
	if (user) {
4225 4226 4227
		if (attr->sched_flags & SCHED_FLAG_SUGOV)
			return -EINVAL;

4228
		retval = security_task_setscheduler(p);
4229 4230 4231 4232
		if (retval)
			return retval;
	}

4233
	/*
I
Ingo Molnar 已提交
4234
	 * Make sure no PI-waiters arrive (or leave) while we are
4235
	 * changing the priority of the task:
4236
	 *
L
Lucas De Marchi 已提交
4237
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4238 4239
	 * runqueue lock must be held.
	 */
4240
	rq = task_rq_lock(p, &rf);
4241
	update_rq_clock(rq);
4242

4243
	/*
I
Ingo Molnar 已提交
4244
	 * Changing the policy of the stop threads its a very bad idea:
4245 4246
	 */
	if (p == rq->stop) {
4247
		task_rq_unlock(rq, p, &rf);
4248 4249 4250
		return -EINVAL;
	}

4251
	/*
4252 4253
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4254
	 */
4255
	if (unlikely(policy == p->policy)) {
4256
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4257 4258 4259
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4260
		if (dl_policy(policy) && dl_param_changed(p, attr))
4261
			goto change;
4262

4263
		p->sched_reset_on_fork = reset_on_fork;
4264
		task_rq_unlock(rq, p, &rf);
4265 4266
		return 0;
	}
4267
change:
4268

4269
	if (user) {
4270
#ifdef CONFIG_RT_GROUP_SCHED
4271 4272 4273 4274 4275
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4276 4277
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4278
			task_rq_unlock(rq, p, &rf);
4279 4280 4281
			return -EPERM;
		}
#endif
4282
#ifdef CONFIG_SMP
4283 4284
		if (dl_bandwidth_enabled() && dl_policy(policy) &&
				!(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4285 4286 4287 4288 4289 4290 4291
			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.
			 */
4292 4293
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4294
				task_rq_unlock(rq, p, &rf);
4295 4296 4297 4298 4299
				return -EPERM;
			}
		}
#endif
	}
4300

I
Ingo Molnar 已提交
4301
	/* Re-check policy now with rq lock held: */
L
Linus Torvalds 已提交
4302 4303
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4304
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4305 4306
		goto recheck;
	}
4307 4308 4309 4310 4311 4312

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4313
	if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4314
		task_rq_unlock(rq, p, &rf);
4315 4316 4317
		return -EBUSY;
	}

4318 4319 4320
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4321 4322 4323 4324 4325 4326 4327 4328
	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.
		 */
4329
		new_effective_prio = rt_effective_prio(p, newprio);
4330 4331
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4332 4333
	}

4334
	queued = task_on_rq_queued(p);
4335
	running = task_current(rq, p);
4336
	if (queued)
4337
		dequeue_task(rq, p, queue_flags);
4338
	if (running)
4339
		put_prev_task(rq, p);
4340

4341
	prev_class = p->sched_class;
4342
	__setscheduler(rq, p, attr, pi);
4343

4344
	if (queued) {
4345 4346 4347 4348
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4349 4350
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4351

4352
		enqueue_task(rq, p, queue_flags);
4353
	}
4354
	if (running)
4355
		set_curr_task(rq, p);
4356

P
Peter Zijlstra 已提交
4357
	check_class_changed(rq, p, prev_class, oldprio);
I
Ingo Molnar 已提交
4358 4359 4360

	/* Avoid rq from going away on us: */
	preempt_disable();
4361
	task_rq_unlock(rq, p, &rf);
4362

4363 4364
	if (pi)
		rt_mutex_adjust_pi(p);
4365

I
Ingo Molnar 已提交
4366
	/* Run balance callbacks after we've adjusted the PI chain: */
4367 4368
	balance_callback(rq);
	preempt_enable();
4369

L
Linus Torvalds 已提交
4370 4371
	return 0;
}
4372

4373 4374 4375 4376 4377 4378 4379 4380 4381
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),
	};

4382 4383
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4384 4385 4386 4387 4388
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4389
	return __sched_setscheduler(p, &attr, check, true);
4390
}
4391 4392 4393 4394 4395 4396
/**
 * 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.
 *
4397 4398
 * Return: 0 on success. An error code otherwise.
 *
4399 4400 4401
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4402
		       const struct sched_param *param)
4403
{
4404
	return _sched_setscheduler(p, policy, param, true);
4405
}
L
Linus Torvalds 已提交
4406 4407
EXPORT_SYMBOL_GPL(sched_setscheduler);

4408 4409
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4410
	return __sched_setscheduler(p, attr, true, true);
4411 4412 4413
}
EXPORT_SYMBOL_GPL(sched_setattr);

4414 4415 4416 4417 4418
int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
{
	return __sched_setscheduler(p, attr, false, true);
}

4419 4420 4421 4422 4423 4424 4425 4426 4427 4428
/**
 * 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.
4429 4430
 *
 * Return: 0 on success. An error code otherwise.
4431 4432
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4433
			       const struct sched_param *param)
4434
{
4435
	return _sched_setscheduler(p, policy, param, false);
4436
}
4437
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4438

I
Ingo Molnar 已提交
4439 4440
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4441 4442 4443
{
	struct sched_param lparam;
	struct task_struct *p;
4444
	int retval;
L
Linus Torvalds 已提交
4445 4446 4447 4448 4449

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4450 4451 4452

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4453
	p = find_process_by_pid(pid);
4454 4455 4456
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4457

L
Linus Torvalds 已提交
4458 4459 4460
	return retval;
}

4461 4462 4463
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
I
Ingo Molnar 已提交
4464
static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4465 4466 4467 4468 4469 4470 4471
{
	u32 size;
	int ret;

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

I
Ingo Molnar 已提交
4472
	/* Zero the full structure, so that a short copy will be nice: */
4473 4474 4475 4476 4477 4478
	memset(attr, 0, sizeof(*attr));

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

I
Ingo Molnar 已提交
4479 4480
	/* Bail out on silly large: */
	if (size > PAGE_SIZE)
4481 4482
		goto err_size;

I
Ingo Molnar 已提交
4483 4484
	/* ABI compatibility quirk: */
	if (!size)
4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518
		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
4519
	 * XXX: Do we want to be lenient like existing syscalls; or do we want
4520 4521
	 * to be strict and return an error on out-of-bounds values?
	 */
4522
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4523

4524
	return 0;
4525 4526 4527

err_size:
	put_user(sizeof(*attr), &uattr->size);
4528
	return -E2BIG;
4529 4530
}

L
Linus Torvalds 已提交
4531 4532 4533 4534 4535
/**
 * 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.
4536 4537
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4538
 */
I
Ingo Molnar 已提交
4539
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4540
{
4541 4542 4543
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4544 4545 4546 4547 4548 4549 4550
	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.
4551 4552
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4553
 */
4554
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4555
{
4556
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4557 4558
}

4559 4560 4561
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4562
 * @uattr: structure containing the extended parameters.
4563
 * @flags: for future extension.
4564
 */
4565 4566
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4567 4568 4569 4570 4571
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4572
	if (!uattr || pid < 0 || flags)
4573 4574
		return -EINVAL;

4575 4576 4577
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4578

4579
	if ((int)attr.sched_policy < 0)
4580
		return -EINVAL;
4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591

	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
4592 4593 4594
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4595 4596 4597
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4598
 */
4599
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4600
{
4601
	struct task_struct *p;
4602
	int retval;
L
Linus Torvalds 已提交
4603 4604

	if (pid < 0)
4605
		return -EINVAL;
L
Linus Torvalds 已提交
4606 4607

	retval = -ESRCH;
4608
	rcu_read_lock();
L
Linus Torvalds 已提交
4609 4610 4611 4612
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4613 4614
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4615
	}
4616
	rcu_read_unlock();
L
Linus Torvalds 已提交
4617 4618 4619 4620
	return retval;
}

/**
4621
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4622 4623
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4624 4625 4626
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4627
 */
4628
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4629
{
4630
	struct sched_param lp = { .sched_priority = 0 };
4631
	struct task_struct *p;
4632
	int retval;
L
Linus Torvalds 已提交
4633 4634

	if (!param || pid < 0)
4635
		return -EINVAL;
L
Linus Torvalds 已提交
4636

4637
	rcu_read_lock();
L
Linus Torvalds 已提交
4638 4639 4640 4641 4642 4643 4644 4645 4646
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4647 4648
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4649
	rcu_read_unlock();
L
Linus Torvalds 已提交
4650 4651 4652 4653 4654 4655 4656 4657 4658

	/*
	 * 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:
4659
	rcu_read_unlock();
L
Linus Torvalds 已提交
4660 4661 4662
	return retval;
}

4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685
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)
4686
				return -EFBIG;
4687 4688 4689 4690 4691
		}

		attr->size = usize;
	}

4692
	ret = copy_to_user(uattr, attr, attr->size);
4693 4694 4695
	if (ret)
		return -EFAULT;

4696
	return 0;
4697 4698 4699
}

/**
4700
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4701
 * @pid: the pid in question.
J
Juri Lelli 已提交
4702
 * @uattr: structure containing the extended parameters.
4703
 * @size: sizeof(attr) for fwd/bwd comp.
4704
 * @flags: for future extension.
4705
 */
4706 4707
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4708 4709 4710 4711 4712 4713 4714 4715
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4716
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729
		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;
4730 4731
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4732 4733 4734
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4735 4736
		attr.sched_priority = p->rt_priority;
	else
4737
		attr.sched_nice = task_nice(p);
4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4749
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4750
{
4751
	cpumask_var_t cpus_allowed, new_mask;
4752 4753
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4754

4755
	rcu_read_lock();
L
Linus Torvalds 已提交
4756 4757 4758

	p = find_process_by_pid(pid);
	if (!p) {
4759
		rcu_read_unlock();
L
Linus Torvalds 已提交
4760 4761 4762
		return -ESRCH;
	}

4763
	/* Prevent p going away */
L
Linus Torvalds 已提交
4764
	get_task_struct(p);
4765
	rcu_read_unlock();
L
Linus Torvalds 已提交
4766

4767 4768 4769 4770
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4771 4772 4773 4774 4775 4776 4777 4778
	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
4779
	retval = -EPERM;
E
Eric W. Biederman 已提交
4780 4781 4782 4783
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4784
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4785 4786 4787
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4788

4789
	retval = security_task_setscheduler(p);
4790
	if (retval)
4791
		goto out_free_new_mask;
4792

4793 4794 4795 4796

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

4797 4798 4799 4800 4801 4802 4803
	/*
	 * 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
4804 4805 4806
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4807
			retval = -EBUSY;
4808
			rcu_read_unlock();
4809
			goto out_free_new_mask;
4810
		}
4811
		rcu_read_unlock();
4812 4813
	}
#endif
P
Peter Zijlstra 已提交
4814
again:
4815
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4816

P
Paul Menage 已提交
4817
	if (!retval) {
4818 4819
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4820 4821 4822 4823 4824
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4825
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4826 4827 4828
			goto again;
		}
	}
4829
out_free_new_mask:
4830 4831 4832 4833
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4834 4835 4836 4837 4838
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4839
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4840
{
4841 4842 4843 4844 4845
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4846 4847 4848 4849
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
I
Ingo Molnar 已提交
4850
 * sys_sched_setaffinity - set the CPU affinity of a process
L
Linus Torvalds 已提交
4851 4852
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4853
 * @user_mask_ptr: user-space pointer to the new CPU mask
4854 4855
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4856
 */
4857 4858
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4859
{
4860
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4861 4862
	int retval;

4863 4864
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4865

4866 4867 4868 4869 4870
	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
4871 4872
}

4873
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4874
{
4875
	struct task_struct *p;
4876
	unsigned long flags;
L
Linus Torvalds 已提交
4877 4878
	int retval;

4879
	rcu_read_lock();
L
Linus Torvalds 已提交
4880 4881 4882 4883 4884 4885

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

4886 4887 4888 4889
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4890
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4891
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4892
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4893 4894

out_unlock:
4895
	rcu_read_unlock();
L
Linus Torvalds 已提交
4896

4897
	return retval;
L
Linus Torvalds 已提交
4898 4899 4900
}

/**
I
Ingo Molnar 已提交
4901
 * sys_sched_getaffinity - get the CPU affinity of a process
L
Linus Torvalds 已提交
4902 4903
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4904
 * @user_mask_ptr: user-space pointer to hold the current CPU mask
4905
 *
4906 4907
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4908
 */
4909 4910
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4911 4912
{
	int ret;
4913
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4914

A
Anton Blanchard 已提交
4915
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4916 4917
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4918 4919
		return -EINVAL;

4920 4921
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4922

4923 4924
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4925
		unsigned int retlen = min(len, cpumask_size());
4926 4927

		if (copy_to_user(user_mask_ptr, mask, retlen))
4928 4929
			ret = -EFAULT;
		else
4930
			ret = retlen;
4931 4932
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4933

4934
	return ret;
L
Linus Torvalds 已提交
4935 4936 4937 4938 4939
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4940 4941
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4942 4943
 *
 * Return: 0.
L
Linus Torvalds 已提交
4944
 */
4945
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4946
{
4947 4948 4949 4950 4951 4952
	struct rq_flags rf;
	struct rq *rq;

	local_irq_disable();
	rq = this_rq();
	rq_lock(rq, &rf);
L
Linus Torvalds 已提交
4953

4954
	schedstat_inc(rq->yld_count);
4955
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4956 4957 4958 4959 4960

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
4961 4962
	preempt_disable();
	rq_unlock(rq, &rf);
4963
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4964 4965 4966 4967 4968 4969

	schedule();

	return 0;
}

4970
#ifndef CONFIG_PREEMPT
4971
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4972
{
4973
	if (should_resched(0)) {
4974
		preempt_schedule_common();
L
Linus Torvalds 已提交
4975 4976
		return 1;
	}
4977
	rcu_all_qs();
L
Linus Torvalds 已提交
4978 4979
	return 0;
}
4980
EXPORT_SYMBOL(_cond_resched);
4981
#endif
L
Linus Torvalds 已提交
4982 4983

/*
4984
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4985 4986
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4987
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4988 4989 4990
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4991
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4992
{
4993
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4994 4995
	int ret = 0;

4996 4997
	lockdep_assert_held(lock);

4998
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4999
		spin_unlock(lock);
P
Peter Zijlstra 已提交
5000
		if (resched)
5001
			preempt_schedule_common();
N
Nick Piggin 已提交
5002 5003
		else
			cpu_relax();
J
Jan Kara 已提交
5004
		ret = 1;
L
Linus Torvalds 已提交
5005 5006
		spin_lock(lock);
	}
J
Jan Kara 已提交
5007
	return ret;
L
Linus Torvalds 已提交
5008
}
5009
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
5010

5011
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
5012 5013 5014
{
	BUG_ON(!in_softirq());

5015
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
5016
		local_bh_enable();
5017
		preempt_schedule_common();
L
Linus Torvalds 已提交
5018 5019 5020 5021 5022
		local_bh_disable();
		return 1;
	}
	return 0;
}
5023
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
5024 5025 5026 5027

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
5028 5029 5030 5031 5032 5033 5034 5035 5036
 * 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
5037
 *	yield();
P
Peter Zijlstra 已提交
5038 5039 5040 5041 5042 5043 5044 5045
 *
 * 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
5046 5047 5048 5049 5050 5051 5052 5053
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

5054 5055 5056 5057
/**
 * 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
5058 5059
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5060 5061 5062 5063
 *
 * 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.
 *
5064
 * Return:
5065 5066 5067
 *	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.
5068
 */
5069
int __sched yield_to(struct task_struct *p, bool preempt)
5070 5071 5072 5073
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5074
	int yielded = 0;
5075 5076 5077 5078 5079 5080

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5081 5082 5083 5084 5085 5086 5087 5088 5089
	/*
	 * 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;
	}

5090
	double_rq_lock(rq, p_rq);
5091
	if (task_rq(p) != p_rq) {
5092 5093 5094 5095 5096
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5097
		goto out_unlock;
5098 5099

	if (curr->sched_class != p->sched_class)
5100
		goto out_unlock;
5101 5102

	if (task_running(p_rq, p) || p->state)
5103
		goto out_unlock;
5104 5105

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5106
	if (yielded) {
5107
		schedstat_inc(rq->yld_count);
5108 5109 5110 5111 5112
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5113
			resched_curr(p_rq);
5114
	}
5115

5116
out_unlock:
5117
	double_rq_unlock(rq, p_rq);
5118
out_irq:
5119 5120
	local_irq_restore(flags);

5121
	if (yielded > 0)
5122 5123 5124 5125 5126 5127
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142
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
5143
/*
I
Ingo Molnar 已提交
5144
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5145 5146 5147 5148
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5149
	int token;
L
Linus Torvalds 已提交
5150 5151
	long ret;

5152
	token = io_schedule_prepare();
L
Linus Torvalds 已提交
5153
	ret = schedule_timeout(timeout);
5154
	io_schedule_finish(token);
5155

L
Linus Torvalds 已提交
5156 5157
	return ret;
}
5158
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5159

5160 5161 5162 5163 5164 5165 5166 5167 5168 5169
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
5170 5171 5172 5173
/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5174 5175 5176
 * 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
5177
 */
5178
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5179 5180 5181 5182 5183 5184 5185 5186
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5187
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5188
	case SCHED_NORMAL:
5189
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5190
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5191 5192 5193 5194 5195 5196 5197 5198 5199 5200
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5201 5202 5203
 * 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
5204
 */
5205
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5206 5207 5208 5209 5210 5211 5212 5213
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5214
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5215
	case SCHED_NORMAL:
5216
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5217
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5218 5219 5220 5221 5222
		ret = 0;
	}
	return ret;
}

5223
static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
L
Linus Torvalds 已提交
5224
{
5225
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5226
	unsigned int time_slice;
5227
	struct rq_flags rf;
5228
	struct rq *rq;
5229
	int retval;
L
Linus Torvalds 已提交
5230 5231

	if (pid < 0)
5232
		return -EINVAL;
L
Linus Torvalds 已提交
5233 5234

	retval = -ESRCH;
5235
	rcu_read_lock();
L
Linus Torvalds 已提交
5236 5237 5238 5239 5240 5241 5242 5243
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5244
	rq = task_rq_lock(p, &rf);
5245 5246 5247
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5248
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5249

5250
	rcu_read_unlock();
5251 5252
	jiffies_to_timespec64(time_slice, t);
	return 0;
5253

L
Linus Torvalds 已提交
5254
out_unlock:
5255
	rcu_read_unlock();
L
Linus Torvalds 已提交
5256 5257 5258
	return retval;
}

5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269
/**
 * 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.
 */
5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295
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

5296
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5297 5298
{
	unsigned long free = 0;
5299
	int ppid;
5300

5301 5302
	if (!try_get_task_stack(p))
		return;
5303 5304 5305 5306

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

	if (p->state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5307
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5308
#ifdef CONFIG_DEBUG_STACK_USAGE
5309
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5310
#endif
5311
	ppid = 0;
5312
	rcu_read_lock();
5313 5314
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5315
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5316
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5317
		task_pid_nr(p), ppid,
5318
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5319

5320
	print_worker_info(KERN_INFO, p);
5321
	show_stack(p, NULL);
5322
	put_task_stack(p);
L
Linus Torvalds 已提交
5323
}
5324
EXPORT_SYMBOL_GPL(sched_show_task);
L
Linus Torvalds 已提交
5325

5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347
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
5348
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5349
{
5350
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5351

5352
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5353 5354
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5355
#else
P
Peter Zijlstra 已提交
5356 5357
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5358
#endif
5359
	rcu_read_lock();
5360
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5361 5362
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5363
		 * console might take a lot of time:
5364 5365 5366
		 * 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
5367 5368
		 */
		touch_nmi_watchdog();
5369
		touch_all_softlockup_watchdogs();
5370
		if (state_filter_match(state_filter, p))
5371
			sched_show_task(p);
5372
	}
L
Linus Torvalds 已提交
5373

I
Ingo Molnar 已提交
5374
#ifdef CONFIG_SCHED_DEBUG
5375 5376
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5377
#endif
5378
	rcu_read_unlock();
I
Ingo Molnar 已提交
5379 5380 5381
	/*
	 * Only show locks if all tasks are dumped:
	 */
5382
	if (!state_filter)
I
Ingo Molnar 已提交
5383
		debug_show_all_locks();
L
Linus Torvalds 已提交
5384 5385
}

5386 5387 5388
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
I
Ingo Molnar 已提交
5389
 * @cpu: CPU the idle task belongs to
5390 5391 5392 5393
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5394
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5395
{
5396
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5397 5398
	unsigned long flags;

5399 5400
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5401

5402
	__sched_fork(0, idle);
5403
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5404
	idle->se.exec_start = sched_clock();
5405
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5406

5407 5408
	kasan_unpoison_task_stack(idle);

5409 5410 5411 5412 5413 5414 5415 5416 5417
#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
5418 5419
	/*
	 * We're having a chicken and egg problem, even though we are
I
Ingo Molnar 已提交
5420
	 * holding rq->lock, the CPU isn't yet set to this CPU so the
5421 5422 5423 5424 5425 5426 5427 5428
	 * 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
5429
	__set_task_cpu(idle, cpu);
5430
	rcu_read_unlock();
L
Linus Torvalds 已提交
5431 5432

	rq->curr = rq->idle = idle;
5433
	idle->on_rq = TASK_ON_RQ_QUEUED;
5434
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5435
	idle->on_cpu = 1;
5436
#endif
5437 5438
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5439 5440

	/* Set the preempt count _outside_ the spinlocks! */
5441
	init_idle_preempt_count(idle, cpu);
5442

I
Ingo Molnar 已提交
5443 5444 5445 5446
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5447
	ftrace_graph_init_idle_task(idle, cpu);
5448
	vtime_init_idle(idle, cpu);
5449
#ifdef CONFIG_SMP
5450 5451
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5452 5453
}

5454 5455
#ifdef CONFIG_SMP

5456 5457 5458
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
5459
	int ret = 1;
5460

5461 5462 5463
	if (!cpumask_weight(cur))
		return ret;

5464
	ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5465 5466 5467 5468

	return ret;
}

5469 5470 5471 5472 5473 5474 5475
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
5476
	 * to a new cpuset; we don't want to change their CPU
5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488
	 * 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,
5489 5490
					      cs_cpus_allowed))
		ret = dl_task_can_attach(p, cs_cpus_allowed);
5491 5492 5493 5494 5495

out:
	return ret;
}

5496
bool sched_smp_initialized __read_mostly;
5497

5498 5499 5500 5501 5502 5503 5504 5505 5506 5507
#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;

5508
	if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5509 5510 5511 5512
		return -EINVAL;

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

5513
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5514 5515
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5516 5517 5518 5519 5520 5521 5522

/*
 * 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)
{
5523
	bool queued, running;
5524 5525
	struct rq_flags rf;
	struct rq *rq;
5526

5527
	rq = task_rq_lock(p, &rf);
5528
	queued = task_on_rq_queued(p);
5529 5530
	running = task_current(rq, p);

5531
	if (queued)
5532
		dequeue_task(rq, p, DEQUEUE_SAVE);
5533
	if (running)
5534
		put_prev_task(rq, p);
5535 5536 5537

	p->numa_preferred_nid = nid;

5538
	if (queued)
5539
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5540
	if (running)
5541
		set_curr_task(rq, p);
5542
	task_rq_unlock(rq, p, &rf);
5543
}
P
Peter Zijlstra 已提交
5544
#endif /* CONFIG_NUMA_BALANCING */
5545

L
Linus Torvalds 已提交
5546
#ifdef CONFIG_HOTPLUG_CPU
5547
/*
I
Ingo Molnar 已提交
5548
 * Ensure that the idle task is using init_mm right before its CPU goes
5549
 * offline.
5550
 */
5551
void idle_task_exit(void)
L
Linus Torvalds 已提交
5552
{
5553
	struct mm_struct *mm = current->active_mm;
5554

5555
	BUG_ON(cpu_online(smp_processor_id()));
5556

5557
	if (mm != &init_mm) {
5558
		switch_mm(mm, &init_mm, current);
5559 5560
		finish_arch_post_lock_switch();
	}
5561
	mmdrop(mm);
L
Linus Torvalds 已提交
5562 5563 5564
}

/*
5565 5566
 * 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
5567 5568 5569
 * 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.
5570 5571
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5572
 */
5573
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5574
{
5575
	long delta = calc_load_fold_active(rq, 1);
5576 5577
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5578 5579
}

5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595
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,
};

5596
/*
5597 5598 5599 5600 5601 5602
 * 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
5603
 */
5604
static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
L
Linus Torvalds 已提交
5605
{
5606
	struct rq *rq = dead_rq;
5607
	struct task_struct *next, *stop = rq->stop;
5608
	struct rq_flags orf = *rf;
5609
	int dest_cpu;
L
Linus Torvalds 已提交
5610 5611

	/*
5612 5613 5614 5615 5616 5617 5618
	 * 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
5619
	 */
5620
	rq->stop = NULL;
5621

5622 5623 5624 5625 5626 5627 5628
	/*
	 * 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);

5629
	for (;;) {
5630 5631
		/*
		 * There's this thread running, bail when that's the only
I
Ingo Molnar 已提交
5632
		 * remaining thread:
5633 5634
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5635
			break;
5636

5637
		/*
I
Ingo Molnar 已提交
5638
		 * pick_next_task() assumes pinned rq->lock:
5639
		 */
5640
		next = pick_next_task(rq, &fake_task, rf);
5641
		BUG_ON(!next);
V
Viresh Kumar 已提交
5642
		put_prev_task(rq, next);
5643

W
Wanpeng Li 已提交
5644 5645 5646 5647 5648 5649 5650 5651 5652
		/*
		 * 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.
		 */
5653
		rq_unlock(rq, rf);
W
Wanpeng Li 已提交
5654
		raw_spin_lock(&next->pi_lock);
5655
		rq_relock(rq, rf);
W
Wanpeng Li 已提交
5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666

		/*
		 * 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;
		}

5667
		/* Find suitable destination for @next, with force if needed. */
5668
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5669
		rq = __migrate_task(rq, rf, next, dest_cpu);
5670
		if (rq != dead_rq) {
5671
			rq_unlock(rq, rf);
5672
			rq = dead_rq;
5673 5674
			*rf = orf;
			rq_relock(rq, rf);
5675
		}
W
Wanpeng Li 已提交
5676
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5677
	}
5678

5679
	rq->stop = stop;
5680
}
L
Linus Torvalds 已提交
5681 5682
#endif /* CONFIG_HOTPLUG_CPU */

5683
void set_rq_online(struct rq *rq)
5684 5685 5686 5687
{
	if (!rq->online) {
		const struct sched_class *class;

5688
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5689 5690 5691 5692 5693 5694 5695 5696 5697
		rq->online = 1;

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

5698
void set_rq_offline(struct rq *rq)
5699 5700 5701 5702 5703 5704 5705 5706 5707
{
	if (rq->online) {
		const struct sched_class *class;

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

5708
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5709 5710 5711 5712
		rq->online = 0;
	}
}

5713
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5714
{
5715
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5716

5717 5718 5719
	rq->age_stamp = sched_clock_cpu(cpu);
}

I
Ingo Molnar 已提交
5720 5721 5722 5723
/*
 * used to mark begin/end of suspend/resume:
 */
static int num_cpus_frozen;
5724

L
Linus Torvalds 已提交
5725
/*
5726 5727 5728
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
5729 5730 5731
 *
 * 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
5732
 */
5733
static void cpuset_cpu_active(void)
5734
{
5735
	if (cpuhp_tasks_frozen) {
5736 5737 5738 5739 5740 5741
		/*
		 * 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.
		 */
5742 5743
		partition_sched_domains(1, NULL, NULL);
		if (--num_cpus_frozen)
5744
			return;
5745 5746 5747 5748 5749
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
5750
		cpuset_force_rebuild();
5751
	}
5752
	cpuset_update_active_cpus();
5753
}
5754

5755
static int cpuset_cpu_inactive(unsigned int cpu)
5756
{
5757
	if (!cpuhp_tasks_frozen) {
5758
		if (dl_cpu_busy(cpu))
5759
			return -EBUSY;
5760
		cpuset_update_active_cpus();
5761
	} else {
5762 5763
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
5764
	}
5765
	return 0;
5766 5767
}

5768
int sched_cpu_activate(unsigned int cpu)
5769
{
5770
	struct rq *rq = cpu_rq(cpu);
5771
	struct rq_flags rf;
5772

5773
	set_cpu_active(cpu, true);
5774

5775
	if (sched_smp_initialized) {
5776
		sched_domains_numa_masks_set(cpu);
5777
		cpuset_cpu_active();
5778
	}
5779 5780 5781 5782 5783

	/*
	 * 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
5784
	 *    after all CPUs have been brought up.
5785 5786 5787 5788
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
5789
	rq_lock_irqsave(rq, &rf);
5790 5791 5792 5793
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
5794
	rq_unlock_irqrestore(rq, &rf);
5795 5796 5797

	update_max_interval();

5798
	return 0;
5799 5800
}

5801
int sched_cpu_deactivate(unsigned int cpu)
5802 5803 5804
{
	int ret;

5805
	set_cpu_active(cpu, false);
5806 5807 5808 5809 5810 5811 5812
	/*
	 * 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.
	 */
5813
	synchronize_rcu_mult(call_rcu, call_rcu_sched);
5814 5815 5816 5817 5818 5819 5820 5821

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
5822
	}
5823 5824
	sched_domains_numa_masks_clear(cpu);
	return 0;
5825 5826
}

5827 5828 5829 5830 5831 5832 5833 5834
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();
}

5835 5836 5837
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
5838
	sched_rq_cpu_starting(cpu);
5839
	sched_tick_start(cpu);
5840
	return 0;
5841 5842
}

5843 5844 5845 5846
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5847
	struct rq_flags rf;
5848 5849 5850

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
5851
	sched_tick_stop(cpu);
5852 5853

	rq_lock_irqsave(rq, &rf);
5854 5855 5856 5857
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
5858
	migrate_tasks(rq, &rf);
5859
	BUG_ON(rq->nr_running != 1);
5860 5861
	rq_unlock_irqrestore(rq, &rf);

5862 5863
	calc_load_migrate(rq);
	update_max_interval();
5864
	nohz_balance_exit_idle(cpu);
5865
	hrtick_clear(rq);
5866 5867 5868 5869
	return 0;
}
#endif

P
Peter Zijlstra 已提交
5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885
#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
5886 5887
void __init sched_init_smp(void)
{
5888 5889
	sched_init_numa();

5890 5891
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
I
Ingo Molnar 已提交
5892
	 * CPU masks are stable and all blatant races in the below code cannot
5893 5894
	 * happen.
	 */
5895
	mutex_lock(&sched_domains_mutex);
P
Peter Zijlstra 已提交
5896
	sched_init_domains(cpu_active_mask);
5897
	mutex_unlock(&sched_domains_mutex);
5898

5899
	/* Move init over to a non-isolated CPU */
5900
	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5901
		BUG();
I
Ingo Molnar 已提交
5902
	sched_init_granularity();
5903

5904
	init_sched_rt_class();
5905
	init_sched_dl_class();
P
Peter Zijlstra 已提交
5906 5907 5908

	sched_init_smt();

5909
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
5910
}
5911 5912 5913

static int __init migration_init(void)
{
5914
	sched_rq_cpu_starting(smp_processor_id());
5915
	return 0;
L
Linus Torvalds 已提交
5916
}
5917 5918
early_initcall(migration_init);

L
Linus Torvalds 已提交
5919 5920 5921
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
5922
	sched_init_granularity();
L
Linus Torvalds 已提交
5923 5924 5925 5926 5927 5928 5929 5930 5931 5932
}
#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);
}

5933
#ifdef CONFIG_CGROUP_SCHED
5934 5935 5936 5937
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
5938
struct task_group root_task_group;
5939
LIST_HEAD(task_groups);
5940 5941 5942

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
5943
#endif
P
Peter Zijlstra 已提交
5944

5945
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5946
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
5947

L
Linus Torvalds 已提交
5948 5949
void __init sched_init(void)
{
I
Ingo Molnar 已提交
5950
	int i, j;
5951 5952
	unsigned long alloc_size = 0, ptr;

5953
	sched_clock_init();
5954
	wait_bit_init();
5955

5956 5957 5958 5959 5960 5961 5962
#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) {
5963
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5964 5965

#ifdef CONFIG_FAIR_GROUP_SCHED
5966
		root_task_group.se = (struct sched_entity **)ptr;
5967 5968
		ptr += nr_cpu_ids * sizeof(void **);

5969
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5970
		ptr += nr_cpu_ids * sizeof(void **);
5971

5972
#endif /* CONFIG_FAIR_GROUP_SCHED */
5973
#ifdef CONFIG_RT_GROUP_SCHED
5974
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5975 5976
		ptr += nr_cpu_ids * sizeof(void **);

5977
		root_task_group.rt_rq = (struct rt_rq **)ptr;
5978 5979
		ptr += nr_cpu_ids * sizeof(void **);

5980
#endif /* CONFIG_RT_GROUP_SCHED */
5981
	}
5982
#ifdef CONFIG_CPUMASK_OFFSTACK
5983 5984 5985
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5986 5987
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5988
	}
5989
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
5990

I
Ingo Molnar 已提交
5991 5992
	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
5993

G
Gregory Haskins 已提交
5994 5995 5996 5997
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

5998
#ifdef CONFIG_RT_GROUP_SCHED
5999
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
6000
			global_rt_period(), global_rt_runtime());
6001
#endif /* CONFIG_RT_GROUP_SCHED */
6002

D
Dhaval Giani 已提交
6003
#ifdef CONFIG_CGROUP_SCHED
6004 6005
	task_group_cache = KMEM_CACHE(task_group, 0);

6006 6007
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
6008
	INIT_LIST_HEAD(&root_task_group.siblings);
6009
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
6010
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
6011

6012
	for_each_possible_cpu(i) {
6013
		struct rq *rq;
L
Linus Torvalds 已提交
6014 6015

		rq = cpu_rq(i);
6016
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
6017
		rq->nr_running = 0;
6018 6019
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
6020
		init_cfs_rq(&rq->cfs);
6021 6022
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
6023
#ifdef CONFIG_FAIR_GROUP_SCHED
6024
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
6025
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
6026
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
6027
		/*
I
Ingo Molnar 已提交
6028
		 * How much CPU bandwidth does root_task_group get?
D
Dhaval Giani 已提交
6029 6030
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
I
Ingo Molnar 已提交
6031 6032
		 * gets 100% of the CPU resources in the system. This overall
		 * system CPU resource is divided among the tasks of
6033
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
6034 6035 6036
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
6037
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
6038
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
I
Ingo Molnar 已提交
6039
		 * then A0's share of the CPU resource is:
D
Dhaval Giani 已提交
6040
		 *
6041
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
6042
		 *
6043 6044
		 * 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
6045
		 */
6046
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6047
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
6048 6049 6050
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6051
#ifdef CONFIG_RT_GROUP_SCHED
6052
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
6053
#endif
L
Linus Torvalds 已提交
6054

I
Ingo Molnar 已提交
6055 6056
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
6057

L
Linus Torvalds 已提交
6058
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6059
		rq->sd = NULL;
G
Gregory Haskins 已提交
6060
		rq->rd = NULL;
6061
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6062
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
6063
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6064
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6065
		rq->push_cpu = 0;
6066
		rq->cpu = i;
6067
		rq->online = 0;
6068 6069
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6070
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6071 6072 6073

		INIT_LIST_HEAD(&rq->cfs_tasks);

6074
		rq_attach_root(rq, &def_root_domain);
6075
#ifdef CONFIG_NO_HZ_COMMON
6076
		rq->last_load_update_tick = jiffies;
6077
		atomic_set(&rq->nohz_flags, 0);
6078
#endif
6079
#endif /* CONFIG_SMP */
6080
		hrtick_rq_init(rq);
L
Linus Torvalds 已提交
6081 6082 6083
		atomic_set(&rq->nr_iowait, 0);
	}

6084
	set_load_weight(&init_task, false);
6085

L
Linus Torvalds 已提交
6086 6087 6088
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
V
Vegard Nossum 已提交
6089
	mmgrab(&init_mm);
L
Linus Torvalds 已提交
6090 6091 6092 6093 6094 6095 6096 6097 6098
	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());
6099 6100 6101

	calc_load_update = jiffies + LOAD_FREQ;

6102
#ifdef CONFIG_SMP
6103
	idle_thread_set_boot_cpu();
6104
	set_cpu_rq_start_time(smp_processor_id());
6105 6106
#endif
	init_sched_fair_class();
6107

6108 6109
	init_schedstats();

6110
	scheduler_running = 1;
L
Linus Torvalds 已提交
6111 6112
}

6113
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6114 6115
static inline int preempt_count_equals(int preempt_offset)
{
6116
	int nested = preempt_count() + rcu_preempt_depth();
6117

A
Arnd Bergmann 已提交
6118
	return (nested == preempt_offset);
6119 6120
}

6121
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6122
{
P
Peter Zijlstra 已提交
6123 6124 6125 6126 6127
	/*
	 * 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.
	 */
6128
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
6129 6130 6131 6132
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
6133
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
6134

6135 6136 6137 6138 6139
	___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
6140
{
I
Ingo Molnar 已提交
6141 6142 6143
	/* Ratelimiting timestamp: */
	static unsigned long prev_jiffy;

6144
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
6145

I
Ingo Molnar 已提交
6146 6147 6148
	/* WARN_ON_ONCE() by default, no rate limit required: */
	rcu_sleep_check();

6149 6150
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
6151 6152
	    system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
	    oops_in_progress)
I
Ingo Molnar 已提交
6153
		return;
6154

I
Ingo Molnar 已提交
6155 6156 6157 6158
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

I
Ingo Molnar 已提交
6159
	/* Save this before calling printk(), since that will clobber it: */
6160 6161
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
6162 6163 6164 6165 6166 6167 6168
	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
6169

6170 6171 6172
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
6173 6174 6175
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
6176 6177
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
6178
		pr_err("Preemption disabled at:");
6179
		print_ip_sym(preempt_disable_ip);
6180 6181
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
6182
	dump_stack();
6183
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
6184
}
6185
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
6186 6187 6188
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6189
void normalize_rt_tasks(void)
6190
{
6191
	struct task_struct *g, *p;
6192 6193 6194
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
6195

6196
	read_lock(&tasklist_lock);
6197
	for_each_process_thread(g, p) {
6198 6199 6200
		/*
		 * Only normalize user tasks:
		 */
6201
		if (p->flags & PF_KTHREAD)
6202 6203
			continue;

6204 6205 6206 6207
		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
6208

6209
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
6210 6211 6212 6213
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
6214
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
6215
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6216
			continue;
I
Ingo Molnar 已提交
6217
		}
L
Linus Torvalds 已提交
6218

6219
		__sched_setscheduler(p, &attr, false, false);
6220
	}
6221
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
6222 6223 6224
}

#endif /* CONFIG_MAGIC_SYSRQ */
6225

6226
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6227
/*
6228
 * These functions are only useful for the IA64 MCA handling, or kdb.
6229 6230 6231 6232 6233 6234 6235 6236 6237
 *
 * 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
6238
 * curr_task - return the current task for a given CPU.
6239 6240 6241
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6242 6243
 *
 * Return: The current task for @cpu.
6244
 */
6245
struct task_struct *curr_task(int cpu)
6246 6247 6248 6249
{
	return cpu_curr(cpu);
}

6250 6251 6252
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6253
/**
I
Ingo Molnar 已提交
6254
 * set_curr_task - set the current task for a given CPU.
6255 6256 6257 6258
 * @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
6259
 * are serviced on a separate stack. It allows the architecture to switch the
I
Ingo Molnar 已提交
6260
 * notion of the current task on a CPU in a non-blocking manner. This function
6261 6262 6263 6264 6265 6266 6267
 * 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!
 */
6268
void ia64_set_curr_task(int cpu, struct task_struct *p)
6269 6270 6271 6272 6273
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6274

D
Dhaval Giani 已提交
6275
#ifdef CONFIG_CGROUP_SCHED
6276 6277 6278
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6279
static void sched_free_group(struct task_group *tg)
6280 6281 6282
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6283
	autogroup_free(tg);
6284
	kmem_cache_free(task_group_cache, tg);
6285 6286 6287
}

/* allocate runqueue etc for a new task group */
6288
struct task_group *sched_create_group(struct task_group *parent)
6289 6290 6291
{
	struct task_group *tg;

6292
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6293 6294 6295
	if (!tg)
		return ERR_PTR(-ENOMEM);

6296
	if (!alloc_fair_sched_group(tg, parent))
6297 6298
		goto err;

6299
	if (!alloc_rt_sched_group(tg, parent))
6300 6301
		goto err;

6302 6303 6304
	return tg;

err:
6305
	sched_free_group(tg);
6306 6307 6308 6309 6310 6311 6312
	return ERR_PTR(-ENOMEM);
}

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

6313
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6314
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6315

I
Ingo Molnar 已提交
6316 6317
	/* Root should already exist: */
	WARN_ON(!parent);
P
Peter Zijlstra 已提交
6318 6319 6320

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6321
	list_add_rcu(&tg->siblings, &parent->children);
6322
	spin_unlock_irqrestore(&task_group_lock, flags);
6323 6324

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
6325 6326
}

6327
/* rcu callback to free various structures associated with a task group */
6328
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6329
{
I
Ingo Molnar 已提交
6330
	/* Now it should be safe to free those cfs_rqs: */
6331
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6332 6333
}

6334
void sched_destroy_group(struct task_group *tg)
6335
{
I
Ingo Molnar 已提交
6336
	/* Wait for possible concurrent references to cfs_rqs complete: */
6337
	call_rcu(&tg->rcu, sched_free_group_rcu);
6338 6339 6340
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6341
{
6342
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
6343

I
Ingo Molnar 已提交
6344
	/* End participation in shares distribution: */
6345
	unregister_fair_sched_group(tg);
6346 6347

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6348
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6349
	list_del_rcu(&tg->siblings);
6350
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6351 6352
}

6353
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
6354
{
P
Peter Zijlstra 已提交
6355
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6356

6357 6358 6359 6360 6361 6362
	/*
	 * 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
6363 6364 6365 6366
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6367
#ifdef CONFIG_FAIR_GROUP_SCHED
6368 6369
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
6370
	else
P
Peter Zijlstra 已提交
6371
#endif
6372
		set_task_rq(tsk, task_cpu(tsk));
6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383
}

/*
 * 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)
{
6384 6385
	int queued, running, queue_flags =
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6386 6387 6388 6389
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);
6390
	update_rq_clock(rq);
6391 6392 6393 6394 6395

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

	if (queued)
6396
		dequeue_task(rq, tsk, queue_flags);
6397
	if (running)
6398 6399 6400
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
6401

6402
	if (queued)
6403
		enqueue_task(rq, tsk, queue_flags);
6404
	if (running)
6405
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6406

6407
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
6408
}
6409

6410
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6411
{
6412
	return css ? container_of(css, struct task_group, css) : NULL;
6413 6414
}

6415 6416
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6417
{
6418 6419
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
6420

6421
	if (!parent) {
6422
		/* This is early initialization for the top cgroup */
6423
		return &root_task_group.css;
6424 6425
	}

6426
	tg = sched_create_group(parent);
6427 6428 6429 6430 6431 6432
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443
/* 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;
}

6444
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6445
{
6446
	struct task_group *tg = css_tg(css);
6447

6448
	sched_offline_group(tg);
6449 6450
}

6451
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6452
{
6453
	struct task_group *tg = css_tg(css);
6454

6455 6456 6457 6458
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
6459 6460
}

6461 6462 6463 6464
/*
 * 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.
 */
6465
static void cpu_cgroup_fork(struct task_struct *task)
6466
{
6467 6468 6469 6470 6471
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

6472
	update_rq_clock(rq);
6473 6474 6475
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
6476 6477
}

6478
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6479
{
6480
	struct task_struct *task;
6481
	struct cgroup_subsys_state *css;
6482
	int ret = 0;
6483

6484
	cgroup_taskset_for_each(task, css, tset) {
6485
#ifdef CONFIG_RT_GROUP_SCHED
6486
		if (!sched_rt_can_attach(css_tg(css), task))
6487
			return -EINVAL;
6488
#else
6489 6490 6491
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
6492
#endif
6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508
		/*
		 * 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;
6509
	}
6510
	return ret;
6511
}
6512

6513
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6514
{
6515
	struct task_struct *task;
6516
	struct cgroup_subsys_state *css;
6517

6518
	cgroup_taskset_for_each(task, css, tset)
6519
		sched_move_task(task);
6520 6521
}

6522
#ifdef CONFIG_FAIR_GROUP_SCHED
6523 6524
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
6525
{
6526
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
6527 6528
}

6529 6530
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
6531
{
6532
	struct task_group *tg = css_tg(css);
6533

6534
	return (u64) scale_load_down(tg->shares);
6535
}
6536 6537

#ifdef CONFIG_CFS_BANDWIDTH
6538 6539
static DEFINE_MUTEX(cfs_constraints_mutex);

6540 6541 6542
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

6543 6544
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

6545 6546
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
6547
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
6548
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568

	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;

6569 6570 6571 6572 6573
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
6574 6575 6576 6577 6578
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

6579
	runtime_enabled = quota != RUNTIME_INF;
6580
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6581 6582 6583 6584 6585 6586
	/*
	 * 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();
6587 6588 6589
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
6590

P
Paul Turner 已提交
6591
	__refill_cfs_bandwidth_runtime(cfs_b);
I
Ingo Molnar 已提交
6592 6593

	/* Restart the period timer (if active) to handle new period expiry: */
P
Peter Zijlstra 已提交
6594 6595
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
I
Ingo Molnar 已提交
6596

6597 6598
	raw_spin_unlock_irq(&cfs_b->lock);

6599
	for_each_online_cpu(i) {
6600
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6601
		struct rq *rq = cfs_rq->rq;
6602
		struct rq_flags rf;
6603

6604
		rq_lock_irq(rq, &rf);
6605
		cfs_rq->runtime_enabled = runtime_enabled;
6606
		cfs_rq->runtime_remaining = 0;
6607

6608
		if (cfs_rq->throttled)
6609
			unthrottle_cfs_rq(cfs_rq);
6610
		rq_unlock_irq(rq, &rf);
6611
	}
6612 6613
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
6614 6615
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
6616
	put_online_cpus();
6617

6618
	return ret;
6619 6620 6621 6622 6623 6624
}

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

6625
	period = ktime_to_ns(tg->cfs_bandwidth.period);
6626 6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637
	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;

6638
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6639 6640
		return -1;

6641
	quota_us = tg->cfs_bandwidth.quota;
6642 6643 6644 6645 6646 6647 6648 6649 6650 6651
	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;
6652
	quota = tg->cfs_bandwidth.quota;
6653 6654 6655 6656 6657 6658 6659 6660

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

6661
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6662 6663 6664 6665 6666
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

6667 6668
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
6669
{
6670
	return tg_get_cfs_quota(css_tg(css));
6671 6672
}

6673 6674
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
6675
{
6676
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6677 6678
}

6679 6680
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6681
{
6682
	return tg_get_cfs_period(css_tg(css));
6683 6684
}

6685 6686
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
6687
{
6688
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
6689 6690
}

6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722
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;
6723
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6724 6725 6726 6727 6728
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
6729
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6730 6731

		quota = normalize_cfs_quota(tg, d);
6732
		parent_quota = parent_b->hierarchical_quota;
6733 6734

		/*
I
Ingo Molnar 已提交
6735 6736
		 * Ensure max(child_quota) <= parent_quota, inherit when no
		 * limit is set:
6737 6738 6739 6740 6741 6742
		 */
		if (quota == RUNTIME_INF)
			quota = parent_quota;
		else if (parent_quota != RUNTIME_INF && quota > parent_quota)
			return -EINVAL;
	}
6743
	cfs_b->hierarchical_quota = quota;
6744 6745 6746 6747 6748 6749

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
6750
	int ret;
6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761
	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);
	}

6762 6763 6764 6765 6766
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6767
}
6768

6769
static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6770
{
6771
	struct task_group *tg = css_tg(seq_css(sf));
6772
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6773

6774 6775 6776
	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);
6777 6778 6779

	return 0;
}
6780
#endif /* CONFIG_CFS_BANDWIDTH */
6781
#endif /* CONFIG_FAIR_GROUP_SCHED */
6782

6783
#ifdef CONFIG_RT_GROUP_SCHED
6784 6785
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
6786
{
6787
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
6788 6789
}

6790 6791
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
6792
{
6793
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
6794
}
6795

6796 6797
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
6798
{
6799
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
6800 6801
}

6802 6803
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6804
{
6805
	return sched_group_rt_period(css_tg(css));
6806
}
6807
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
6808

6809
static struct cftype cpu_legacy_files[] = {
6810
#ifdef CONFIG_FAIR_GROUP_SCHED
6811 6812
	{
		.name = "shares",
6813 6814
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
6815
	},
6816
#endif
6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827
#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,
	},
6828 6829
	{
		.name = "stat",
6830
		.seq_show = cpu_cfs_stat_show,
6831
	},
6832
#endif
6833
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6834
	{
P
Peter Zijlstra 已提交
6835
		.name = "rt_runtime_us",
6836 6837
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
6838
	},
6839 6840
	{
		.name = "rt_period_us",
6841 6842
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
6843
	},
6844
#endif
I
Ingo Molnar 已提交
6845
	{ }	/* Terminate */
6846 6847
};

6848 6849
static int cpu_extra_stat_show(struct seq_file *sf,
			       struct cgroup_subsys_state *css)
6850 6851 6852
{
#ifdef CONFIG_CFS_BANDWIDTH
	{
6853
		struct task_group *tg = css_tg(css);
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 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 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
		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 */
};

7011
struct cgroup_subsys cpu_cgrp_subsys = {
7012
	.css_alloc	= cpu_cgroup_css_alloc,
7013
	.css_online	= cpu_cgroup_css_online,
7014
	.css_released	= cpu_cgroup_css_released,
7015
	.css_free	= cpu_cgroup_css_free,
7016
	.css_extra_stat_show = cpu_extra_stat_show,
7017
	.fork		= cpu_cgroup_fork,
7018 7019
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
7020
	.legacy_cftypes	= cpu_legacy_files,
7021
	.dfl_cftypes	= cpu_files,
7022
	.early_init	= true,
7023
	.threaded	= true,
7024 7025
};

7026
#endif	/* CONFIG_CGROUP_SCHED */
7027

7028 7029 7030 7031 7032
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073

/*
 * 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,
};
7074 7075

#undef CREATE_TRACE_POINTS