core.c 174.4 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 <linux/nospec.h>
11

12 13
#include <linux/kcov.h>

14
#include <asm/switch_to.h>
15
#include <asm/tlb.h>
L
Linus Torvalds 已提交
16

17
#include "../workqueue_internal.h"
18
#include "../smpboot.h"
19

20 21
#include "pelt.h"

22
#define CREATE_TRACE_POINTS
23
#include <trace/events/sched.h>
24

25
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
26

27
#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_JUMP_LABEL)
I
Ingo Molnar 已提交
28 29
/*
 * Debugging: various feature bits
30 31 32 33
 *
 * 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
34
 */
P
Peter Zijlstra 已提交
35 36
#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |
I
Ingo Molnar 已提交
37
const_debug unsigned int sysctl_sched_features =
38
#include "features.h"
P
Peter Zijlstra 已提交
39 40
	0;
#undef SCHED_FEAT
41
#endif
P
Peter Zijlstra 已提交
42

43 44 45 46 47 48
/*
 * 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;

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

55
__read_mostly int scheduler_running;
56

P
Peter Zijlstra 已提交
57 58 59 60 61
/*
 * 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
62

63 64 65
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
66
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
67 68 69 70 71 72 73 74 75 76
	__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))) {
77
			rq_pin_lock(rq, rf);
78 79 80 81 82 83 84 85 86 87 88 89
			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.
 */
90
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
91 92 93 94 95 96
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
97
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
98 99 100 101 102 103 104 105 106 107 108 109
		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)
		 *
110
		 * If we observe the old CPU in task_rq_lock(), the acquire of
111 112
		 * the old rq->lock will fully serialize against the stores.
		 *
113 114 115
		 * If we observe the new CPU in task_rq_lock(), the address
		 * dependency headed by '[L] rq = task_rq()' and the acquire
		 * will pair with the WMB to ensure we then also see migrating.
116 117
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
118
			rq_pin_lock(rq, rf);
119 120 121
			return rq;
		}
		raw_spin_unlock(&rq->lock);
122
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
123 124 125 126 127 128

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

129 130 131 132 133 134 135 136 137 138
/*
 * 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...
 */
139 140
	s64 __maybe_unused steal = 0, irq_delta = 0;

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
#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;

180
#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
181
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
182
		update_irq_load_avg(rq, irq_delta + steal);
183 184 185 186 187 188 189 190 191 192 193 194 195
#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
196 197
	if (sched_feat(WARN_DOUBLE_CLOCK))
		SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
198 199
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
200

201 202 203 204 205 206 207 208
	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
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
#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);
227
	struct rq_flags rf;
P
Peter Zijlstra 已提交
228 229 230

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

231
	rq_lock(rq, &rf);
232
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
233
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
234
	rq_unlock(rq, &rf);
P
Peter Zijlstra 已提交
235 236 237 238

	return HRTIMER_NORESTART;
}

239
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
240

241
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
242 243 244
{
	struct hrtimer *timer = &rq->hrtick_timer;

245
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
246 247
}

248 249 250 251
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
252
{
253
	struct rq *rq = arg;
254
	struct rq_flags rf;
255

256
	rq_lock(rq, &rf);
P
Peter Zijlstra 已提交
257
	__hrtick_restart(rq);
258
	rq->hrtick_csd_pending = 0;
259
	rq_unlock(rq, &rf);
260 261
}

262 263 264 265 266
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
267
void hrtick_start(struct rq *rq, u64 delay)
268
{
269
	struct hrtimer *timer = &rq->hrtick_timer;
270 271 272 273 274 275 276 277 278
	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);
279

280
	hrtimer_set_expires(timer, time);
281 282

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
283
		__hrtick_restart(rq);
284
	} else if (!rq->hrtick_csd_pending) {
285
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
286 287
		rq->hrtick_csd_pending = 1;
	}
288 289
}

290 291 292 293 294 295
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
296
void hrtick_start(struct rq *rq, u64 delay)
297
{
W
Wanpeng Li 已提交
298 299 300 301 302
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
303 304
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
305 306
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
307

308
static void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
309
{
310 311
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
312

313 314 315 316
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
317

318 319
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
320
}
A
Andrew Morton 已提交
321
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
322 323 324 325
static inline void hrtick_clear(struct rq *rq)
{
}

326
static inline void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
327 328
{
}
A
Andrew Morton 已提交
329
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
330

331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
/*
 * 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;								\
})

349
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
350 351 352 353 354 355 356 357 358 359
/*
 * 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);
}
360 361 362 363 364 365 366 367 368 369

/*
 * 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);
370
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
371 372 373 374 375 376 377 378 379 380 381 382 383 384

	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;
}

385 386 387 388 389 390
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
391 392 393 394 395 396 397

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
398 399
#endif

400 401 402 403 404 405 406 407 408
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.
	 *
409 410
	 * In order to ensure that a pending wakeup will observe our pending
	 * state, even in the failed case, an explicit smp_mb() must be used.
411
	 */
412 413
	smp_mb__before_atomic();
	if (cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))
414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433
		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
434
		/* Task can safely be re-inserted now: */
435 436 437 438
		node = node->next;
		task->wake_q.next = NULL;

		/*
439 440
		 * wake_up_process() executes a full barrier, which pairs with
		 * the queueing in wake_q_add() so as not to miss wakeups.
441 442 443 444 445 446
		 */
		wake_up_process(task);
		put_task_struct(task);
	}
}

I
Ingo Molnar 已提交
447
/*
448
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
449 450 451 452 453
 *
 * 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.
 */
454
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
455
{
456
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
457 458
	int cpu;

459
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
460

461
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
462 463
		return;

464
	cpu = cpu_of(rq);
465

466
	if (cpu == smp_processor_id()) {
467
		set_tsk_need_resched(curr);
468
		set_preempt_need_resched();
I
Ingo Molnar 已提交
469
		return;
470
	}
I
Ingo Molnar 已提交
471

472
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
473
		smp_send_reschedule(cpu);
474 475
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
476 477
}

478
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
479 480 481 482
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

483
	raw_spin_lock_irqsave(&rq->lock, flags);
484 485
	if (cpu_online(cpu) || cpu == smp_processor_id())
		resched_curr(rq);
486
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
487
}
488

489
#ifdef CONFIG_SMP
490
#ifdef CONFIG_NO_HZ_COMMON
491
/*
I
Ingo Molnar 已提交
492 493
 * In the semi idle case, use the nearest busy CPU for migrating timers
 * from an idle CPU.  This is good for power-savings.
494 495
 *
 * We don't do similar optimization for completely idle system, as
I
Ingo Molnar 已提交
496 497
 * 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).
498
 */
499
int get_nohz_timer_target(void)
500
{
501
	int i, cpu = smp_processor_id();
502 503
	struct sched_domain *sd;

504
	if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
505 506
		return cpu;

507
	rcu_read_lock();
508
	for_each_domain(cpu, sd) {
509
		for_each_cpu(i, sched_domain_span(sd)) {
510 511 512
			if (cpu == i)
				continue;

513
			if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
514 515 516 517
				cpu = i;
				goto unlock;
			}
		}
518
	}
519

520 521
	if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
		cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
522 523
unlock:
	rcu_read_unlock();
524 525
	return cpu;
}
I
Ingo Molnar 已提交
526

527 528 529 530 531 532 533 534 535 536
/*
 * 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.
 */
537
static void wake_up_idle_cpu(int cpu)
538 539 540 541 542 543
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

544
	if (set_nr_and_not_polling(rq->idle))
545
		smp_send_reschedule(cpu);
546 547
	else
		trace_sched_wake_idle_without_ipi(cpu);
548 549
}

550
static bool wake_up_full_nohz_cpu(int cpu)
551
{
552 553 554 555 556 557
	/*
	 * 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.
	 */
558 559
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
560
	if (tick_nohz_full_cpu(cpu)) {
561 562
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
563
			tick_nohz_full_kick_cpu(cpu);
564 565 566 567 568 569
		return true;
	}

	return false;
}

570 571 572 573 574
/*
 * 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.
 */
575 576
void wake_up_nohz_cpu(int cpu)
{
577
	if (!wake_up_full_nohz_cpu(cpu))
578 579 580
		wake_up_idle_cpu(cpu);
}

581
static inline bool got_nohz_idle_kick(void)
582
{
583
	int cpu = smp_processor_id();
584

P
Peter Zijlstra 已提交
585
	if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK))
586 587 588 589 590 591 592 593 594
		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
	 */
P
Peter Zijlstra 已提交
595
	atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu));
596
	return false;
597 598
}

599
#else /* CONFIG_NO_HZ_COMMON */
600

601
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
602
{
603
	return false;
P
Peter Zijlstra 已提交
604 605
}

606
#endif /* CONFIG_NO_HZ_COMMON */
607

608
#ifdef CONFIG_NO_HZ_FULL
609
bool sched_can_stop_tick(struct rq *rq)
610
{
611 612 613 614 615 616
	int fifo_nr_running;

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

617
	/*
618 619
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
620
	 */
621 622 623 624 625
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
626 627
	}

628 629 630 631 632 633 634 635 636 637 638 639 640 641
	/*
	 * 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)
642
		return false;
643

644
	return true;
645 646
}
#endif /* CONFIG_NO_HZ_FULL */
647
#endif /* CONFIG_SMP */
648

649 650
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
651
/*
652 653 654 655
 * 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.
656
 */
657
int walk_tg_tree_from(struct task_group *from,
658
			     tg_visitor down, tg_visitor up, void *data)
659 660
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
661
	int ret;
662

663 664
	parent = from;

665
down:
P
Peter Zijlstra 已提交
666 667
	ret = (*down)(parent, data);
	if (ret)
668
		goto out;
669 670 671 672 673 674 675
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
676
	ret = (*up)(parent, data);
677 678
	if (ret || parent == from)
		goto out;
679 680 681 682 683

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
684
out:
P
Peter Zijlstra 已提交
685
	return ret;
686 687
}

688
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
689
{
690
	return 0;
P
Peter Zijlstra 已提交
691
}
692 693
#endif

694
static void set_load_weight(struct task_struct *p, bool update_load)
695
{
N
Nikhil Rao 已提交
696 697 698
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
699 700 701
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
702
	if (idle_policy(p->policy)) {
703
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
704
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
705 706
		return;
	}
707

708 709 710 711 712 713 714 715 716 717
	/*
	 * 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];
	}
718 719
}

720
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
721
{
722 723 724
	if (!(flags & ENQUEUE_NOCLOCK))
		update_rq_clock(rq);

725 726
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
727

728
	p->sched_class->enqueue_task(rq, p, flags);
729 730
}

731
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
732
{
733 734 735
	if (!(flags & DEQUEUE_NOCLOCK))
		update_rq_clock(rq);

736 737
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
738

739
	p->sched_class->dequeue_task(rq, p, flags);
740 741
}

742
void activate_task(struct rq *rq, struct task_struct *p, int flags)
743 744 745 746
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

747
	enqueue_task(rq, p, flags);
748 749
}

750
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
751 752 753 754
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

755
	dequeue_task(rq, p, flags);
756 757
}

758
/*
I
Ingo Molnar 已提交
759
 * __normal_prio - return the priority that is based on the static prio
760 761 762
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
763
	return p->static_prio;
764 765
}

766 767 768 769 770 771 772
/*
 * 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.
 */
773
static inline int normal_prio(struct task_struct *p)
774 775 776
{
	int prio;

777 778 779
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
780 781 782 783 784 785 786 787 788 789 790 791 792
		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.
 */
793
static int effective_prio(struct task_struct *p)
794 795 796 797 798 799 800 801 802 803 804 805
{
	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
806 807 808
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
809 810
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
811
 */
812
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
813 814 815 816
{
	return cpu_curr(task_cpu(p)) == p;
}

817
/*
818 819 820 821 822
 * 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().
823
 */
824 825
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
826
				       int oldprio)
827 828 829
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
830
			prev_class->switched_from(rq, p);
831

P
Peter Zijlstra 已提交
832
		p->sched_class->switched_to(rq, p);
833
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
834
		p->sched_class->prio_changed(rq, p, oldprio);
835 836
}

837
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
838 839 840 841 842 843 844 845 846 847
{
	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) {
848
				resched_curr(rq);
849 850 851 852 853 854 855 856 857
				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.
	 */
858
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
859
		rq_clock_skip_update(rq);
860 861
}

L
Linus Torvalds 已提交
862
#ifdef CONFIG_SMP
863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889

static inline bool is_per_cpu_kthread(struct task_struct *p)
{
	if (!(p->flags & PF_KTHREAD))
		return false;

	if (p->nr_cpus_allowed != 1)
		return false;

	return true;
}

/*
 * Per-CPU kthreads are allowed to run on !actie && online CPUs, see
 * __set_cpus_allowed_ptr() and select_fallback_rq().
 */
static inline bool is_cpu_allowed(struct task_struct *p, int cpu)
{
	if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
		return false;

	if (is_per_cpu_kthread(p))
		return cpu_online(cpu);

	return cpu_active(cpu);
}

P
Peter Zijlstra 已提交
890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908
/*
 * 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.
 */
909 910
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
911 912 913
{
	lockdep_assert_held(&rq->lock);

914
	WRITE_ONCE(p->on_rq, TASK_ON_RQ_MIGRATING);
915
	dequeue_task(rq, p, DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
916
	set_task_cpu(p, new_cpu);
917
	rq_unlock(rq, rf);
P
Peter Zijlstra 已提交
918 919 920

	rq = cpu_rq(new_cpu);

921
	rq_lock(rq, rf);
P
Peter Zijlstra 已提交
922 923
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
924
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
925 926 927 928 929 930 931 932 933 934 935
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
I
Ingo Molnar 已提交
936
 * Move (not current) task off this CPU, onto the destination CPU. We're doing
P
Peter Zijlstra 已提交
937 938 939 940 941 942 943
 * 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.
 */
944 945
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
946 947
{
	/* Affinity changed (again). */
948
	if (!is_cpu_allowed(p, dest_cpu))
949
		return rq;
P
Peter Zijlstra 已提交
950

951
	update_rq_clock(rq);
952
	rq = move_queued_task(rq, rf, p, dest_cpu);
953 954

	return rq;
P
Peter Zijlstra 已提交
955 956 957 958 959 960 961 962 963 964
}

/*
 * 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;
965 966
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
967
	struct rq_flags rf;
P
Peter Zijlstra 已提交
968 969

	/*
I
Ingo Molnar 已提交
970 971
	 * 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
972 973 974 975 976 977 978 979
	 */
	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();
980 981

	raw_spin_lock(&p->pi_lock);
982
	rq_lock(rq, &rf);
983 984 985 986 987
	/*
	 * 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.
	 */
988 989
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
990
			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
991 992 993
		else
			p->wake_cpu = arg->dest_cpu;
	}
994
	rq_unlock(rq, &rf);
995 996
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
997 998 999 1000
	local_irq_enable();
	return 0;
}

1001 1002 1003 1004 1005
/*
 * 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
1006 1007 1008 1009 1010
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1011 1012
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1013 1014 1015
	struct rq *rq = task_rq(p);
	bool queued, running;

1016
	lockdep_assert_held(&p->pi_lock);
1017 1018 1019 1020 1021 1022 1023 1024 1025 1026

	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);
1027
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1028 1029 1030 1031
	}
	if (running)
		put_prev_task(rq, p);

1032
	p->sched_class->set_cpus_allowed(p, new_mask);
1033 1034

	if (queued)
1035
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1036
	if (running)
1037
		set_curr_task(rq, p);
1038 1039
}

P
Peter Zijlstra 已提交
1040 1041 1042 1043 1044 1045 1046 1047 1048
/*
 * 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.
 */
1049 1050
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1051
{
1052
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1053
	unsigned int dest_cpu;
1054 1055
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1056 1057
	int ret = 0;

1058
	rq = task_rq_lock(p, &rf);
1059
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
1060

1061 1062 1063 1064 1065 1066 1067
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1068 1069 1070 1071 1072 1073 1074 1075 1076
	/*
	 * 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
1077 1078 1079
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1080 1081
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
	if (dest_cpu >= nr_cpu_ids) {
P
Peter Zijlstra 已提交
1082 1083 1084 1085 1086 1087
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1088 1089 1090
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
I
Ingo Molnar 已提交
1091
		 * !active we want to ensure they are strict per-CPU threads.
1092 1093 1094 1095 1096 1097
		 */
		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
1098 1099 1100 1101 1102 1103 1104
	/* 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;

	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. */
1105
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1106 1107 1108
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1109 1110 1111 1112 1113
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1114
		rq = move_queued_task(rq, &rf, p, dest_cpu);
1115
	}
P
Peter Zijlstra 已提交
1116
out:
1117
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1118 1119 1120

	return ret;
}
1121 1122 1123 1124 1125

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
1126 1127
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1128
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1129
{
1130 1131 1132 1133 1134
#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
1135
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1136
			!p->on_rq);
1137

1138 1139 1140 1141 1142 1143 1144 1145 1146
	/*
	 * 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)));

1147
#ifdef CONFIG_LOCKDEP
1148 1149 1150 1151 1152
	/*
	 * 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
1153
	 * see task_group().
1154 1155 1156 1157
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1158 1159 1160
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1161 1162 1163 1164
	/*
	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
	 */
	WARN_ON_ONCE(!cpu_online(new_cpu));
1165 1166
#endif

1167
	trace_sched_migrate_task(p, new_cpu);
1168

1169
	if (task_cpu(p) != new_cpu) {
1170
		if (p->sched_class->migrate_task_rq)
1171
			p->sched_class->migrate_task_rq(p, new_cpu);
1172
		p->se.nr_migrations++;
1173
		rseq_migrate(p);
1174
		perf_event_task_migrate(p);
1175
	}
I
Ingo Molnar 已提交
1176 1177

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1178 1179
}

1180
#ifdef CONFIG_NUMA_BALANCING
1181 1182
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1183
	if (task_on_rq_queued(p)) {
1184
		struct rq *src_rq, *dst_rq;
1185
		struct rq_flags srf, drf;
1186 1187 1188 1189

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

1190 1191 1192
		rq_pin_lock(src_rq, &srf);
		rq_pin_lock(dst_rq, &drf);

1193
		p->on_rq = TASK_ON_RQ_MIGRATING;
1194 1195 1196
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1197
		p->on_rq = TASK_ON_RQ_QUEUED;
1198
		check_preempt_curr(dst_rq, p, 0);
1199 1200 1201 1202

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

1203 1204 1205 1206
	} 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
1207
		 * previous CPU our target instead of where it really is.
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223
		 */
		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;

1224 1225 1226
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1227 1228 1229
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1230 1231
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1232
	double_rq_lock(src_rq, dst_rq);
1233

1234 1235 1236 1237 1238 1239
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

1240
	if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1241 1242
		goto unlock;

1243
	if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1244 1245 1246 1247 1248 1249 1250 1251 1252
		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);
1253 1254
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1255 1256 1257 1258 1259 1260 1261

	return ret;
}

/*
 * Cross migrate two tasks
 */
1262 1263
int migrate_swap(struct task_struct *cur, struct task_struct *p,
		int target_cpu, int curr_cpu)
1264 1265 1266 1267 1268 1269
{
	struct migration_swap_arg arg;
	int ret = -EINVAL;

	arg = (struct migration_swap_arg){
		.src_task = cur,
1270
		.src_cpu = curr_cpu,
1271
		.dst_task = p,
1272
		.dst_cpu = target_cpu,
1273 1274 1275 1276 1277
	};

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

1278 1279 1280 1281
	/*
	 * 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.
	 */
1282 1283 1284
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

1285
	if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
1286 1287
		goto out;

1288
	if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1289 1290
		goto out;

1291
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1292 1293 1294 1295 1296
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}
1297
#endif /* CONFIG_NUMA_BALANCING */
1298

L
Linus Torvalds 已提交
1299 1300 1301
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1302 1303 1304 1305 1306 1307 1308
 * 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
1309 1310 1311 1312 1313 1314
 * 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
1315
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1316
{
1317
	int running, queued;
1318
	struct rq_flags rf;
R
Roland McGrath 已提交
1319
	unsigned long ncsw;
1320
	struct rq *rq;
L
Linus Torvalds 已提交
1321

1322 1323 1324 1325 1326 1327 1328 1329
	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);
1330

1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341
		/*
		 * 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
1342 1343 1344
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1345
			cpu_relax();
R
Roland McGrath 已提交
1346
		}
1347

1348 1349 1350 1351 1352
		/*
		 * 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.
		 */
1353
		rq = task_rq_lock(p, &rf);
1354
		trace_sched_wait_task(p);
1355
		running = task_running(rq, p);
1356
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1357
		ncsw = 0;
1358
		if (!match_state || p->state == match_state)
1359
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1360
		task_rq_unlock(rq, p, &rf);
1361

R
Roland McGrath 已提交
1362 1363 1364 1365 1366 1367
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
		/*
		 * 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;
		}
1378

1379 1380 1381 1382 1383
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1384
		 * So if it was still runnable (but just not actively
1385 1386 1387
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1388
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1389
			ktime_t to = NSEC_PER_SEC / HZ;
1390 1391 1392

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1393 1394
			continue;
		}
1395

1396 1397 1398 1399 1400 1401 1402
		/*
		 * 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
1403 1404

	return ncsw;
L
Linus Torvalds 已提交
1405 1406 1407 1408 1409 1410 1411 1412 1413
}

/***
 * 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
1414
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1415 1416 1417 1418 1419
 * 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.
 */
1420
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1421 1422 1423 1424 1425 1426 1427 1428 1429
{
	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
1430
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1431

1432
/*
1433
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1434 1435 1436 1437 1438
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
1439
 *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
1440
 *    see __set_cpus_allowed_ptr(). At this point the newly online
I
Ingo Molnar 已提交
1441
 *    CPU isn't yet part of the sched domains, and balancing will not
1442 1443
 *    see it.
 *
I
Ingo Molnar 已提交
1444
 *  - on CPU-down we clear cpu_active() to mask the sched domains and
1445
 *    avoid the load balancer to place new tasks on the to be removed
I
Ingo Molnar 已提交
1446
 *    CPU. Existing tasks will remain running there and will be taken
1447 1448 1449 1450 1451 1452
 *    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.
1453
 */
1454 1455
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1456 1457
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1458 1459
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1460

1461
	/*
I
Ingo Molnar 已提交
1462 1463 1464
	 * 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.
1465 1466 1467 1468 1469 1470 1471 1472
	 */
	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;
1473
			if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1474 1475
				return dest_cpu;
		}
1476
	}
1477

1478 1479
	for (;;) {
		/* Any allowed, online CPU? */
1480
		for_each_cpu(dest_cpu, &p->cpus_allowed) {
1481
			if (!is_cpu_allowed(p, dest_cpu))
1482
				continue;
1483

1484 1485
			goto out;
		}
1486

1487
		/* No more Mr. Nice Guy. */
1488 1489
		switch (state) {
		case cpuset:
1490 1491 1492 1493 1494
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
I
Ingo Molnar 已提交
1495
			/* Fall-through */
1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514
		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()) {
1515
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1516 1517
					task_pid_nr(p), p->comm, cpu);
		}
1518 1519 1520 1521 1522
	}

	return dest_cpu;
}

1523
/*
1524
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1525
 */
1526
static inline
1527
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1528
{
1529 1530
	lockdep_assert_held(&p->pi_lock);

1531
	if (p->nr_cpus_allowed > 1)
1532
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1533
	else
1534
		cpu = cpumask_any(&p->cpus_allowed);
1535 1536 1537 1538

	/*
	 * 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
1539
	 * CPU.
1540 1541 1542 1543 1544 1545
	 *
	 * 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 ]
	 */
1546
	if (unlikely(!is_cpu_allowed(p, cpu)))
1547
		cpu = select_fallback_rq(task_cpu(p), p);
1548 1549

	return cpu;
1550
}
1551 1552 1553 1554 1555 1556

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
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 1585 1586 1587
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;
	}
}

1588 1589 1590 1591 1592 1593 1594 1595
#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
1596
#endif /* CONFIG_SMP */
1597

P
Peter Zijlstra 已提交
1598
static void
1599
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1600
{
1601
	struct rq *rq;
1602

1603 1604 1605 1606
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1607

1608 1609
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1610 1611
		__schedstat_inc(rq->ttwu_local);
		__schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1612 1613 1614
	} else {
		struct sched_domain *sd;

1615
		__schedstat_inc(p->se.statistics.nr_wakeups_remote);
1616
		rcu_read_lock();
1617
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1618
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1619
				__schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1620 1621 1622
				break;
			}
		}
1623
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1624
	}
1625 1626

	if (wake_flags & WF_MIGRATED)
1627
		__schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1628 1629
#endif /* CONFIG_SMP */

1630 1631
	__schedstat_inc(rq->ttwu_count);
	__schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1632 1633

	if (wake_flags & WF_SYNC)
1634
		__schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1635 1636
}

1637
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1638
{
T
Tejun Heo 已提交
1639
	activate_task(rq, p, en_flags);
1640
	p->on_rq = TASK_ON_RQ_QUEUED;
1641

I
Ingo Molnar 已提交
1642
	/* If a worker is waking up, notify the workqueue: */
1643 1644
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1645 1646
}

1647 1648 1649
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1650
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1651
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1652 1653 1654
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1655 1656
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1657
#ifdef CONFIG_SMP
1658 1659
	if (p->sched_class->task_woken) {
		/*
1660 1661
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1662
		 */
1663
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1664
		p->sched_class->task_woken(rq, p);
1665
		rq_repin_lock(rq, rf);
1666
	}
T
Tejun Heo 已提交
1667

1668
	if (rq->idle_stamp) {
1669
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1670
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1671

1672 1673 1674
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1675
			rq->avg_idle = max;
1676

T
Tejun Heo 已提交
1677 1678 1679 1680 1681
		rq->idle_stamp = 0;
	}
#endif
}

1682
static void
1683
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1684
		 struct rq_flags *rf)
1685
{
1686
	int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1687

1688 1689
	lockdep_assert_held(&rq->lock);

1690 1691 1692
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1693 1694

	if (wake_flags & WF_MIGRATED)
1695
		en_flags |= ENQUEUE_MIGRATED;
1696 1697
#endif

1698
	ttwu_activate(rq, p, en_flags);
1699
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1700 1701 1702 1703 1704 1705 1706 1707 1708 1709
}

/*
 * 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)
{
1710
	struct rq_flags rf;
1711 1712 1713
	struct rq *rq;
	int ret = 0;

1714
	rq = __task_rq_lock(p, &rf);
1715
	if (task_on_rq_queued(p)) {
1716 1717
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1718
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1719 1720
		ret = 1;
	}
1721
	__task_rq_unlock(rq, &rf);
1722 1723 1724 1725

	return ret;
}

1726
#ifdef CONFIG_SMP
1727
void sched_ttwu_pending(void)
1728 1729
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1730
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1731
	struct task_struct *p, *t;
1732
	struct rq_flags rf;
1733

1734 1735 1736
	if (!llist)
		return;

1737
	rq_lock_irqsave(rq, &rf);
1738
	update_rq_clock(rq);
1739

1740 1741
	llist_for_each_entry_safe(p, t, llist, wake_entry)
		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1742

1743
	rq_unlock_irqrestore(rq, &rf);
1744 1745 1746 1747
}

void scheduler_ipi(void)
{
1748 1749 1750 1751 1752
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1753
	preempt_fold_need_resched();
1754

1755
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771
		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
1772
	sched_ttwu_pending();
1773 1774 1775 1776

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1777
	if (unlikely(got_nohz_idle_kick())) {
1778
		this_rq()->idle_balance = 1;
1779
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1780
	}
1781
	irq_exit();
1782 1783
}

P
Peter Zijlstra 已提交
1784
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1785
{
1786 1787
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1788 1789
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1790 1791 1792 1793 1794 1795
	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);
	}
1796
}
1797

1798 1799 1800
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
1801
	struct rq_flags rf;
1802

1803 1804 1805 1806
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1807 1808 1809 1810

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
1811
		rq_lock_irqsave(rq, &rf);
1812 1813
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
I
Ingo Molnar 已提交
1814
		/* Else CPU is not idle, do nothing here: */
1815
		rq_unlock_irqrestore(rq, &rf);
1816
	}
1817 1818 1819

out:
	rcu_read_unlock();
1820 1821
}

1822
bool cpus_share_cache(int this_cpu, int that_cpu)
1823 1824 1825
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1826
#endif /* CONFIG_SMP */
1827

1828
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1829 1830
{
	struct rq *rq = cpu_rq(cpu);
1831
	struct rq_flags rf;
1832

1833
#if defined(CONFIG_SMP)
1834
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
I
Ingo Molnar 已提交
1835
		sched_clock_cpu(cpu); /* Sync clocks across CPUs */
P
Peter Zijlstra 已提交
1836
		ttwu_queue_remote(p, cpu, wake_flags);
1837 1838 1839 1840
		return;
	}
#endif

1841
	rq_lock(rq, &rf);
1842
	update_rq_clock(rq);
1843
	ttwu_do_activate(rq, p, wake_flags, &rf);
1844
	rq_unlock(rq, &rf);
T
Tejun Heo 已提交
1845 1846
}

1847 1848 1849 1850 1851 1852
/*
 * 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
1853 1854
 * migrates, all its activity on its old CPU [c0] happens-before any subsequent
 * execution on its new CPU [c1].
1855 1856 1857 1858 1859 1860 1861 1862
 *
 * 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
 *
1863
 * Release/acquire chaining guarantees that B happens after A and C after B.
I
Ingo Molnar 已提交
1864
 * Note: the CPU doing B need not be c0 or c1
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 1894 1895
 *
 * 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)
1896
 *   2) smp_cond_load_acquire(!X->on_cpu)
1897 1898 1899 1900 1901 1902 1903 1904 1905 1906
 *
 * 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);
 *
1907
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924
 *                    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
 *
 *
1925 1926 1927
 * However, for wakeups there is a second guarantee we must provide, namely we
 * must ensure that CONDITION=1 done by the caller can not be reordered with
 * accesses to the task state; see try_to_wake_up() and set_current_state().
1928 1929
 */

T
Tejun Heo 已提交
1930
/**
L
Linus Torvalds 已提交
1931
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1932
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1933
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1934
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1935
 *
1936
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1937
 *
1938 1939 1940 1941 1942
 * 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().
 *
1943 1944 1945
 * This function executes a full memory barrier before accessing the task
 * state; see set_current_state().
 *
1946 1947
 * Return: %true if @p->state changes (an actual wakeup was done),
 *	   %false otherwise.
L
Linus Torvalds 已提交
1948
 */
1949 1950
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1951 1952
{
	unsigned long flags;
1953
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1954

1955 1956 1957 1958 1959 1960
	/*
	 * 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.
	 */
1961
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1962
	smp_mb__after_spinlock();
P
Peter Zijlstra 已提交
1963
	if (!(p->state & state))
L
Linus Torvalds 已提交
1964 1965
		goto out;

1966 1967
	trace_sched_waking(p);

I
Ingo Molnar 已提交
1968 1969
	/* We're going to change ->state: */
	success = 1;
L
Linus Torvalds 已提交
1970 1971
	cpu = task_cpu(p);

1972 1973 1974 1975 1976
	/*
	 * 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.
	 *
1977 1978 1979 1980 1981 1982 1983 1984
	 * sched_ttwu_pending()			try_to_wake_up()
	 *   STORE p->on_rq = 1			  LOAD p->state
	 *   UNLOCK rq->lock
	 *
	 * __schedule() (switch to task 'p')
	 *   LOCK rq->lock			  smp_rmb();
	 *   smp_mb__after_spinlock();
	 *   UNLOCK rq->lock
1985 1986
	 *
	 * [task p]
1987
	 *   STORE p->state = UNINTERRUPTIBLE	  LOAD p->on_rq
1988
	 *
1989 1990
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
1991 1992
	 */
	smp_rmb();
1993 1994
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
1995 1996

#ifdef CONFIG_SMP
1997 1998 1999 2000 2001 2002 2003
	/*
	 * 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.
	 *
2004 2005 2006 2007 2008 2009 2010 2011
	 * __schedule() (switch to task 'p')	try_to_wake_up()
	 *   STORE p->on_cpu = 1		  LOAD p->on_rq
	 *   UNLOCK rq->lock
	 *
	 * __schedule() (put 'p' to sleep)
	 *   LOCK rq->lock			  smp_rmb();
	 *   smp_mb__after_spinlock();
	 *   STORE p->on_rq = 0			  LOAD p->on_cpu
2012
	 *
2013 2014
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
2015 2016 2017
	 */
	smp_rmb();

P
Peter Zijlstra 已提交
2018
	/*
I
Ingo Molnar 已提交
2019
	 * If the owning (remote) CPU is still in the middle of schedule() with
2020
	 * this task as prev, wait until its done referencing the task.
2021
	 *
2022
	 * Pairs with the smp_store_release() in finish_task().
2023 2024 2025
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2026
	 */
2027
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2028

2029
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2030
	p->state = TASK_WAKING;
2031

2032
	if (p->in_iowait) {
2033
		delayacct_blkio_end(p);
2034 2035 2036
		atomic_dec(&task_rq(p)->nr_iowait);
	}

2037
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2038 2039
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2040
		set_task_cpu(p, cpu);
2041
	}
2042 2043 2044 2045

#else /* CONFIG_SMP */

	if (p->in_iowait) {
2046
		delayacct_blkio_end(p);
2047 2048 2049
		atomic_dec(&task_rq(p)->nr_iowait);
	}

L
Linus Torvalds 已提交
2050 2051
#endif /* CONFIG_SMP */

2052
	ttwu_queue(p, cpu, wake_flags);
2053
stat:
2054
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2055
out:
2056
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2057 2058 2059 2060

	return success;
}

T
Tejun Heo 已提交
2061 2062 2063
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2064
 * @rf: request-queue flags for pinning
T
Tejun Heo 已提交
2065
 *
2066
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2067
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2068
 * the current task.
T
Tejun Heo 已提交
2069
 */
2070
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2071 2072 2073
{
	struct rq *rq = task_rq(p);

2074 2075 2076 2077
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2078 2079
	lockdep_assert_held(&rq->lock);

2080
	if (!raw_spin_trylock(&p->pi_lock)) {
2081 2082 2083 2084 2085 2086
		/*
		 * 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.
		 */
2087
		rq_unlock(rq, rf);
2088
		raw_spin_lock(&p->pi_lock);
2089
		rq_relock(rq, rf);
2090 2091
	}

T
Tejun Heo 已提交
2092
	if (!(p->state & TASK_NORMAL))
2093
		goto out;
T
Tejun Heo 已提交
2094

2095 2096
	trace_sched_waking(p);

2097 2098
	if (!task_on_rq_queued(p)) {
		if (p->in_iowait) {
2099
			delayacct_blkio_end(p);
2100 2101
			atomic_dec(&rq->nr_iowait);
		}
2102
		ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2103
	}
P
Peter Zijlstra 已提交
2104

2105
	ttwu_do_wakeup(rq, p, 0, rf);
2106
	ttwu_stat(p, smp_processor_id(), 0);
2107 2108
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2109 2110
}

2111 2112 2113 2114 2115
/**
 * 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
2116 2117 2118
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2119
 *
2120
 * This function executes a full memory barrier before accessing the task state.
2121
 */
2122
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2123
{
2124
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2125 2126 2127
}
EXPORT_SYMBOL(wake_up_process);

2128
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2129 2130 2131 2132 2133 2134 2135
{
	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
2136 2137 2138
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2139
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2140
{
P
Peter Zijlstra 已提交
2141 2142 2143
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2144 2145
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2146
	p->se.prev_sum_exec_runtime	= 0;
2147
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2148
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2149
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2150

2151 2152 2153 2154
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2155
#ifdef CONFIG_SCHEDSTATS
2156
	/* Even if schedstat is disabled, there should not be garbage */
2157
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2158
#endif
N
Nick Piggin 已提交
2159

2160
	RB_CLEAR_NODE(&p->dl.rb_node);
2161
	init_dl_task_timer(&p->dl);
2162
	init_dl_inactive_task_timer(&p->dl);
2163
	__dl_clear_params(p);
2164

P
Peter Zijlstra 已提交
2165
	INIT_LIST_HEAD(&p->rt.run_list);
2166 2167 2168 2169
	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
2170

2171 2172 2173
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2174

2175
	init_numa_balancing(clone_flags, p);
I
Ingo Molnar 已提交
2176 2177
}

2178 2179
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2180
#ifdef CONFIG_NUMA_BALANCING
2181

2182 2183 2184
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2185
		static_branch_enable(&sched_numa_balancing);
2186
	else
2187
		static_branch_disable(&sched_numa_balancing);
2188
}
2189 2190 2191 2192 2193 2194 2195

#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;
2196
	int state = static_branch_likely(&sched_numa_balancing);
2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211

	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
2212

2213 2214
#ifdef CONFIG_SCHEDSTATS

2215
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2216
static bool __initdata __sched_schedstats = false;
2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239

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;

2240 2241 2242 2243 2244
	/*
	 * 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.
	 */
2245
	if (!strcmp(str, "enable")) {
2246
		__sched_schedstats = true;
2247 2248
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2249
		__sched_schedstats = false;
2250 2251 2252 2253 2254 2255 2256 2257 2258 2259
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2260 2261 2262 2263 2264
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284
#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;
}
2285 2286 2287 2288
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2289 2290 2291 2292

/*
 * fork()/clone()-time setup:
 */
2293
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2294
{
2295
	unsigned long flags;
I
Ingo Molnar 已提交
2296

2297
	__sched_fork(clone_flags, p);
2298
	/*
2299
	 * We mark the process as NEW here. This guarantees that
2300 2301 2302
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2303
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2304

2305 2306 2307 2308 2309
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2310 2311 2312 2313
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2314
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2315
			p->policy = SCHED_NORMAL;
2316
			p->static_prio = NICE_TO_PRIO(0);
2317 2318 2319 2320 2321
			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);
2322
		set_load_weight(p, false);
2323

2324 2325 2326 2327 2328 2329
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2330

2331
	if (dl_prio(p->prio))
2332
		return -EAGAIN;
2333
	else if (rt_prio(p->prio))
2334
		p->sched_class = &rt_sched_class;
2335
	else
H
Hiroshi Shimamoto 已提交
2336
		p->sched_class = &fair_sched_class;
2337

2338
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2339

2340 2341 2342 2343 2344 2345 2346
	/*
	 * 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.
	 */
2347
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2348
	/*
I
Ingo Molnar 已提交
2349
	 * We're setting the CPU for the first time, we don't migrate,
2350 2351
	 * so use __set_task_cpu().
	 */
2352
	__set_task_cpu(p, smp_processor_id());
2353 2354
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2355
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2356

2357
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2358
	if (likely(sched_info_on()))
2359
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2360
#endif
P
Peter Zijlstra 已提交
2361 2362
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2363
#endif
2364
	init_task_preempt_count(p);
2365
#ifdef CONFIG_SMP
2366
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2367
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2368
#endif
2369
	return 0;
L
Linus Torvalds 已提交
2370 2371
}

2372 2373 2374
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
2375
		return BW_UNIT;
2376 2377 2378 2379 2380 2381 2382 2383 2384

	/*
	 * 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;

2385
	return div64_u64(runtime << BW_SHIFT, period);
2386 2387
}

L
Linus Torvalds 已提交
2388 2389 2390 2391 2392 2393 2394
/*
 * 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.
 */
2395
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2396
{
2397
	struct rq_flags rf;
I
Ingo Molnar 已提交
2398
	struct rq *rq;
2399

2400
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2401
	p->state = TASK_RUNNING;
2402 2403 2404 2405
#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
2406
	 *  - any previously selected CPU might disappear through hotplug
2407 2408 2409
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2410
	 */
2411
	p->recent_used_cpu = task_cpu(p);
2412
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2413
#endif
2414
	rq = __task_rq_lock(p, &rf);
2415
	update_rq_clock(rq);
2416
	post_init_entity_util_avg(&p->se);
2417

2418
	activate_task(rq, p, ENQUEUE_NOCLOCK);
2419
	p->on_rq = TASK_ON_RQ_QUEUED;
2420
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2421
	check_preempt_curr(rq, p, WF_FORK);
2422
#ifdef CONFIG_SMP
2423 2424 2425 2426 2427
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2428
		rq_unpin_lock(rq, &rf);
2429
		p->sched_class->task_woken(rq, p);
2430
		rq_repin_lock(rq, &rf);
2431
	}
2432
#endif
2433
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2434 2435
}

2436 2437
#ifdef CONFIG_PREEMPT_NOTIFIERS

2438
static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key);
2439

2440 2441
void preempt_notifier_inc(void)
{
2442
	static_branch_inc(&preempt_notifier_key);
2443 2444 2445 2446 2447
}
EXPORT_SYMBOL_GPL(preempt_notifier_inc);

void preempt_notifier_dec(void)
{
2448
	static_branch_dec(&preempt_notifier_key);
2449 2450 2451
}
EXPORT_SYMBOL_GPL(preempt_notifier_dec);

2452
/**
2453
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2454
 * @notifier: notifier struct to register
2455 2456 2457
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2458
	if (!static_branch_unlikely(&preempt_notifier_key))
2459 2460
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2461 2462 2463 2464 2465 2466
	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
2467
 * @notifier: notifier struct to unregister
2468
 *
2469
 * This is *not* safe to call from within a preemption notifier.
2470 2471 2472 2473 2474 2475 2476
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2477
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2478 2479 2480
{
	struct preempt_notifier *notifier;

2481
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2482 2483 2484
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2485 2486
static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
2487
	if (static_branch_unlikely(&preempt_notifier_key))
2488 2489 2490
		__fire_sched_in_preempt_notifiers(curr);
}

2491
static void
2492 2493
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2494 2495 2496
{
	struct preempt_notifier *notifier;

2497
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2498 2499 2500
		notifier->ops->sched_out(notifier, next);
}

2501 2502 2503 2504
static __always_inline void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
2505
	if (static_branch_unlikely(&preempt_notifier_key))
2506 2507 2508
		__fire_sched_out_preempt_notifiers(curr, next);
}

2509
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2510

2511
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2512 2513 2514
{
}

2515
static inline void
2516 2517 2518 2519 2520
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2521
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2522

2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550
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
}

2551 2552
static inline void
prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
2553
{
2554 2555 2556 2557 2558 2559 2560 2561
	/*
	 * 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_);
2562 2563
#ifdef CONFIG_DEBUG_SPINLOCK
	/* this is a valid case when another task releases the spinlock */
2564
	rq->lock.owner = next;
2565
#endif
2566 2567 2568 2569
}

static inline void finish_lock_switch(struct rq *rq)
{
2570 2571 2572 2573 2574 2575 2576 2577 2578
	/*
	 * 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);
}

2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590
/*
 * 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

2591 2592 2593
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2594
 * @prev: the current task that is being switched out
2595 2596 2597 2598 2599 2600 2601 2602 2603
 * @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.
 */
2604 2605 2606
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2607
{
2608
	kcov_prepare_switch(prev);
2609
	sched_info_switch(rq, prev, next);
2610
	perf_event_task_sched_out(prev, next);
2611
	rseq_preempt(prev);
2612
	fire_sched_out_preempt_notifiers(prev, next);
2613
	prepare_task(next);
2614 2615 2616
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2617 2618 2619 2620
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2621 2622 2623 2624
 * 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
2625 2626
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2627
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2628 2629
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2630 2631 2632 2633 2634
 *
 * 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
2635
 */
2636
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2637 2638
	__releases(rq->lock)
{
2639
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2640
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2641
	long prev_state;
L
Linus Torvalds 已提交
2642

2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653
	/*
	 * 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.
	 */
2654 2655 2656 2657
	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);
2658

L
Linus Torvalds 已提交
2659 2660 2661 2662
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2663
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2664 2665
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2666 2667
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
2668
	 * finish_task), otherwise a concurrent wakeup can get prev
2669 2670
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2671
	 */
O
Oleg Nesterov 已提交
2672
	prev_state = prev->state;
2673
	vtime_task_switch(prev);
2674
	perf_event_task_sched_in(prev, current);
2675 2676
	finish_task(prev);
	finish_lock_switch(rq);
2677
	finish_arch_post_lock_switch();
2678
	kcov_finish_switch(current);
S
Steven Rostedt 已提交
2679

2680
	fire_sched_in_preempt_notifiers(current);
2681
	/*
2682 2683 2684 2685 2686 2687 2688 2689 2690 2691
	 * 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.
2692
	 */
2693 2694
	if (mm) {
		membarrier_mm_sync_core_before_usermode(mm);
L
Linus Torvalds 已提交
2695
		mmdrop(mm);
2696
	}
2697 2698 2699
	if (unlikely(prev_state == TASK_DEAD)) {
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);
2700

2701 2702 2703 2704 2705 2706 2707 2708 2709 2710
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
		 */
		kprobe_flush_task(prev);

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

		put_task_struct(prev);
2711
	}
2712

2713
	tick_nohz_task_switch();
2714
	return rq;
L
Linus Torvalds 已提交
2715 2716
}

2717 2718 2719
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2720
static void __balance_callback(struct rq *rq)
2721
{
2722 2723 2724
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2725

2726 2727 2728 2729 2730 2731 2732 2733
	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;
2734

2735
		func(rq);
2736
	}
2737 2738 2739 2740 2741 2742 2743
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2744 2745 2746
}

#else
2747

2748
static inline void balance_callback(struct rq *rq)
2749
{
L
Linus Torvalds 已提交
2750 2751
}

2752 2753
#endif

L
Linus Torvalds 已提交
2754 2755 2756 2757
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2758
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2759 2760
	__releases(rq->lock)
{
2761
	struct rq *rq;
2762

2763 2764 2765 2766 2767 2768 2769 2770 2771
	/*
	 * 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).
	 */

2772
	rq = finish_task_switch(prev);
2773
	balance_callback(rq);
2774
	preempt_enable();
2775

L
Linus Torvalds 已提交
2776
	if (current->set_child_tid)
2777
		put_user(task_pid_vnr(current), current->set_child_tid);
2778 2779

	calculate_sigpending();
L
Linus Torvalds 已提交
2780 2781 2782
}

/*
2783
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2784
 */
2785
static __always_inline struct rq *
2786
context_switch(struct rq *rq, struct task_struct *prev,
2787
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2788
{
I
Ingo Molnar 已提交
2789
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2790

2791
	prepare_task_switch(rq, prev, next);
2792

I
Ingo Molnar 已提交
2793 2794
	mm = next->mm;
	oldmm = prev->active_mm;
2795 2796 2797 2798 2799
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2800
	arch_start_context_switch(prev);
2801

2802 2803 2804 2805 2806 2807 2808
	/*
	 * 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.
	 */
2809
	if (!mm) {
L
Linus Torvalds 已提交
2810
		next->active_mm = oldmm;
V
Vegard Nossum 已提交
2811
		mmgrab(oldmm);
L
Linus Torvalds 已提交
2812 2813
		enter_lazy_tlb(oldmm, next);
	} else
2814
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2815

2816
	if (!prev->mm) {
L
Linus Torvalds 已提交
2817 2818 2819
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2820

2821
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2822

2823
	prepare_lock_switch(rq, next, rf);
L
Linus Torvalds 已提交
2824 2825 2826

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2827
	barrier();
2828 2829

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2830 2831 2832
}

/*
2833
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2834 2835
 *
 * externally visible scheduler statistics: current number of runnable
2836
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2837 2838 2839 2840 2841 2842 2843 2844 2845
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2846
}
L
Linus Torvalds 已提交
2847

2848
/*
I
Ingo Molnar 已提交
2849
 * Check if only the current task is running on the CPU.
2850 2851 2852 2853 2854 2855 2856 2857 2858 2859
 *
 * 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)
2860 2861 2862
 */
bool single_task_running(void)
{
2863
	return raw_rq()->nr_running == 1;
2864 2865 2866
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2867
unsigned long long nr_context_switches(void)
2868
{
2869 2870
	int i;
	unsigned long long sum = 0;
2871

2872
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2873
		sum += cpu_rq(i)->nr_switches;
2874

L
Linus Torvalds 已提交
2875 2876
	return sum;
}
2877

2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907
/*
 * 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
2908 2909 2910
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2911

2912
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2913
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2914

L
Linus Torvalds 已提交
2915 2916
	return sum;
}
2917

2918 2919 2920 2921 2922 2923 2924
/*
 * 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.
 */

2925
unsigned long nr_iowait_cpu(int cpu)
2926
{
2927
	struct rq *this = cpu_rq(cpu);
2928 2929
	return atomic_read(&this->nr_iowait);
}
2930

2931 2932
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2933 2934 2935
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2936 2937
}

I
Ingo Molnar 已提交
2938
#ifdef CONFIG_SMP
2939

2940
/*
P
Peter Zijlstra 已提交
2941 2942
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2943
 */
P
Peter Zijlstra 已提交
2944
void sched_exec(void)
2945
{
P
Peter Zijlstra 已提交
2946
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2947
	unsigned long flags;
2948
	int dest_cpu;
2949

2950
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2951
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2952 2953
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2954

2955
	if (likely(cpu_active(dest_cpu))) {
2956
		struct migration_arg arg = { p, dest_cpu };
2957

2958 2959
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2960 2961
		return;
	}
2962
unlock:
2963
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2964
}
I
Ingo Molnar 已提交
2965

L
Linus Torvalds 已提交
2966 2967 2968
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2969
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2970 2971

EXPORT_PER_CPU_SYMBOL(kstat);
2972
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2973

2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990
/*
 * 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);
}

2991 2992 2993 2994 2995 2996 2997
/*
 * 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)
{
2998
	struct rq_flags rf;
2999
	struct rq *rq;
3000
	u64 ns;
3001

3002 3003
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
3004
	 * 64-bit doesn't need locks to atomically read a 64-bit value.
3005 3006 3007
	 * 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
3008 3009
	 * 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
3010
	 * indistinguishable from the read occurring a few cycles earlier.
3011 3012
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3013
	 */
3014
	if (!p->on_cpu || !task_on_rq_queued(p))
3015 3016 3017
		return p->se.sum_exec_runtime;
#endif

3018
	rq = task_rq_lock(p, &rf);
3019 3020 3021 3022 3023 3024
	/*
	 * 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)) {
3025
		prefetch_curr_exec_start(p);
3026 3027 3028 3029
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3030
	task_rq_unlock(rq, p, &rf);
3031 3032 3033

	return ns;
}
3034

3035 3036 3037 3038 3039 3040 3041 3042
/*
 * 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
3043
	struct task_struct *curr = rq->curr;
3044
	struct rq_flags rf;
3045 3046

	sched_clock_tick();
I
Ingo Molnar 已提交
3047

3048 3049
	rq_lock(rq, &rf);

3050
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3051
	curr->sched_class->task_tick(rq, curr, 0);
3052
	cpu_load_update_active(rq);
3053
	calc_global_load_tick(rq);
3054 3055

	rq_unlock(rq, &rf);
3056

3057
	perf_event_task_tick();
3058

3059
#ifdef CONFIG_SMP
3060
	rq->idle_balance = idle_cpu(cpu);
3061
	trigger_load_balance(rq);
3062
#endif
L
Linus Torvalds 已提交
3063 3064
}

3065
#ifdef CONFIG_NO_HZ_FULL
3066 3067 3068

struct tick_work {
	int			cpu;
3069
	atomic_t		state;
3070 3071
	struct delayed_work	work;
};
3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098
/* Values for ->state, see diagram below. */
#define TICK_SCHED_REMOTE_OFFLINE	0
#define TICK_SCHED_REMOTE_OFFLINING	1
#define TICK_SCHED_REMOTE_RUNNING	2

/*
 * State diagram for ->state:
 *
 *
 *          TICK_SCHED_REMOTE_OFFLINE
 *                    |   ^
 *                    |   |
 *                    |   | sched_tick_remote()
 *                    |   |
 *                    |   |
 *                    +--TICK_SCHED_REMOTE_OFFLINING
 *                    |   ^
 *                    |   |
 * sched_tick_start() |   | sched_tick_stop()
 *                    |   |
 *                    V   |
 *          TICK_SCHED_REMOTE_RUNNING
 *
 *
 * Other transitions get WARN_ON_ONCE(), except that sched_tick_remote()
 * and sched_tick_start() are happy to leave the state in RUNNING.
 */
3099 3100 3101 3102 3103 3104 3105 3106 3107

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);
3108
	struct task_struct *curr;
3109
	struct rq_flags rf;
3110
	u64 delta;
3111
	int os;
3112 3113 3114 3115 3116 3117 3118 3119

	/*
	 * 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.
	 */
3120 3121
	if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
		goto out_requeue;
3122

3123 3124
	rq_lock_irq(rq, &rf);
	curr = rq->curr;
3125
	if (is_idle_task(curr) || cpu_is_offline(cpu))
3126
		goto out_unlock;
3127

3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139
	update_rq_clock(rq);
	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);

out_unlock:
	rq_unlock_irq(rq, &rf);
3140

3141
out_requeue:
3142 3143 3144
	/*
	 * Run the remote tick once per second (1Hz). This arbitrary
	 * frequency is large enough to avoid overload but short enough
3145 3146
	 * to keep scheduler internal stats reasonably up to date.  But
	 * first update state to reflect hotplug activity if required.
3147
	 */
3148 3149 3150 3151
	os = atomic_fetch_add_unless(&twork->state, -1, TICK_SCHED_REMOTE_RUNNING);
	WARN_ON_ONCE(os == TICK_SCHED_REMOTE_OFFLINE);
	if (os == TICK_SCHED_REMOTE_RUNNING)
		queue_delayed_work(system_unbound_wq, dwork, HZ);
3152 3153 3154 3155
}

static void sched_tick_start(int cpu)
{
3156
	int os;
3157 3158 3159 3160 3161 3162 3163 3164
	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);
3165 3166 3167 3168 3169 3170 3171
	os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_RUNNING);
	WARN_ON_ONCE(os == TICK_SCHED_REMOTE_RUNNING);
	if (os == TICK_SCHED_REMOTE_OFFLINE) {
		twork->cpu = cpu;
		INIT_DELAYED_WORK(&twork->work, sched_tick_remote);
		queue_delayed_work(system_unbound_wq, &twork->work, HZ);
	}
3172 3173 3174 3175 3176 3177
}

#ifdef CONFIG_HOTPLUG_CPU
static void sched_tick_stop(int cpu)
{
	struct tick_work *twork;
3178
	int os;
3179 3180 3181 3182 3183 3184 3185

	if (housekeeping_cpu(cpu, HK_FLAG_TICK))
		return;

	WARN_ON_ONCE(!tick_work_cpu);

	twork = per_cpu_ptr(tick_work_cpu, cpu);
3186 3187 3188 3189
	/* There cannot be competing actions, but don't rely on stop-machine. */
	os = atomic_xchg(&twork->state, TICK_SCHED_REMOTE_OFFLINING);
	WARN_ON_ONCE(os != TICK_SCHED_REMOTE_RUNNING);
	/* Don't cancel, as this would mess up the state machine. */
3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202
}
#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) { }
3203
#endif
L
Linus Torvalds 已提交
3204

3205
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3206
				defined(CONFIG_TRACE_PREEMPT_TOGGLE))
3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220
/*
 * 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);
	}
}
3221

3222
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3223
{
3224
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3225 3226 3227
	/*
	 * Underflow?
	 */
3228 3229
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3230
#endif
3231
	__preempt_count_add(val);
3232
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3233 3234 3235
	/*
	 * Spinlock count overflowing soon?
	 */
3236 3237
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3238
#endif
3239
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3240
}
3241
EXPORT_SYMBOL(preempt_count_add);
3242
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3243

3244 3245 3246 3247 3248 3249 3250 3251 3252 3253
/*
 * 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());
}

3254
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3255
{
3256
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3257 3258 3259
	/*
	 * Underflow?
	 */
3260
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3261
		return;
L
Linus Torvalds 已提交
3262 3263 3264
	/*
	 * Is the spinlock portion underflowing?
	 */
3265 3266 3267
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3268
#endif
3269

3270
	preempt_latency_stop(val);
3271
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3272
}
3273
EXPORT_SYMBOL(preempt_count_sub);
3274
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3275

3276 3277 3278
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3279 3280
#endif

3281 3282 3283 3284 3285 3286 3287 3288 3289
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
3290
/*
I
Ingo Molnar 已提交
3291
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3292
 */
I
Ingo Molnar 已提交
3293
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3294
{
3295 3296 3297
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3298 3299 3300
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3301 3302
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3303

I
Ingo Molnar 已提交
3304
	debug_show_held_locks(prev);
3305
	print_modules();
I
Ingo Molnar 已提交
3306 3307
	if (irqs_disabled())
		print_irqtrace_events(prev);
3308 3309
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3310
		pr_err("Preemption disabled at:");
3311
		print_ip_sym(preempt_disable_ip);
3312 3313
		pr_cont("\n");
	}
3314 3315 3316
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3317
	dump_stack();
3318
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3319
}
L
Linus Torvalds 已提交
3320

I
Ingo Molnar 已提交
3321 3322 3323 3324 3325
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3326
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3327 3328
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3329
#endif
3330

3331
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3332
		__schedule_bug(prev);
3333 3334
		preempt_count_set(PREEMPT_DISABLED);
	}
3335
	rcu_sleep_check();
I
Ingo Molnar 已提交
3336

L
Linus Torvalds 已提交
3337 3338
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3339
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3340 3341 3342 3343 3344 3345
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3346
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3347
{
3348
	const struct sched_class *class;
I
Ingo Molnar 已提交
3349
	struct task_struct *p;
L
Linus Torvalds 已提交
3350 3351

	/*
3352 3353 3354 3355
	 * 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
3356
	 */
3357 3358 3359 3360
	if (likely((prev->sched_class == &idle_sched_class ||
		    prev->sched_class == &fair_sched_class) &&
		   rq->nr_running == rq->cfs.h_nr_running)) {

3361
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3362 3363 3364
		if (unlikely(p == RETRY_TASK))
			goto again;

I
Ingo Molnar 已提交
3365
		/* Assumes fair_sched_class->next == idle_sched_class */
3366
		if (unlikely(!p))
3367
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3368 3369

		return p;
L
Linus Torvalds 已提交
3370 3371
	}

3372
again:
3373
	for_each_class(class) {
3374
		p = class->pick_next_task(rq, prev, rf);
3375 3376 3377
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3378
			return p;
3379
		}
I
Ingo Molnar 已提交
3380
	}
3381

I
Ingo Molnar 已提交
3382 3383
	/* The idle class should always have a runnable task: */
	BUG();
I
Ingo Molnar 已提交
3384
}
L
Linus Torvalds 已提交
3385

I
Ingo Molnar 已提交
3386
/*
3387
 * __schedule() is the main scheduler function.
3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421
 *
 * 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
3422
 *
3423
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3424
 */
3425
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3426 3427
{
	struct task_struct *prev, *next;
3428
	unsigned long *switch_count;
3429
	struct rq_flags rf;
I
Ingo Molnar 已提交
3430
	struct rq *rq;
3431
	int cpu;
I
Ingo Molnar 已提交
3432 3433 3434 3435 3436 3437

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3438

3439
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3440
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3441

3442
	local_irq_disable();
3443
	rcu_note_context_switch(preempt);
3444

3445 3446 3447 3448
	/*
	 * 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().
3449 3450 3451
	 *
	 * The membarrier system call requires a full memory barrier
	 * after coming from user-space, before storing to rq->curr.
3452
	 */
3453
	rq_lock(rq, &rf);
3454
	smp_mb__after_spinlock();
L
Linus Torvalds 已提交
3455

I
Ingo Molnar 已提交
3456 3457
	/* Promote REQ to ACT */
	rq->clock_update_flags <<= 1;
3458
	update_rq_clock(rq);
3459

3460
	switch_count = &prev->nivcsw;
3461
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3462
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3463
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3464
		} else {
3465
			deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3466 3467
			prev->on_rq = 0;

3468 3469 3470 3471 3472
			if (prev->in_iowait) {
				atomic_inc(&rq->nr_iowait);
				delayacct_blkio_start();
			}

T
Tejun Heo 已提交
3473
			/*
3474 3475 3476
			 * 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
3477 3478 3479 3480
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3481
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3482
				if (to_wakeup)
3483
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3484 3485
			}
		}
I
Ingo Molnar 已提交
3486
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3487 3488
	}

3489
	next = pick_next_task(rq, prev, &rf);
3490
	clear_tsk_need_resched(prev);
3491
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3492 3493 3494 3495

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
3496 3497 3498
		/*
		 * The membarrier system call requires each architecture
		 * to have a full memory barrier after updating
3499 3500 3501 3502 3503 3504 3505 3506 3507 3508
		 * 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),
3509
		 */
L
Linus Torvalds 已提交
3510 3511
		++*switch_count;

3512
		trace_sched_switch(preempt, prev, next);
I
Ingo Molnar 已提交
3513 3514 3515

		/* Also unlocks the rq: */
		rq = context_switch(rq, prev, next, &rf);
3516
	} else {
3517
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3518
		rq_unlock_irq(rq, &rf);
3519
	}
L
Linus Torvalds 已提交
3520

3521
	balance_callback(rq);
L
Linus Torvalds 已提交
3522
}
3523

3524 3525
void __noreturn do_task_dead(void)
{
I
Ingo Molnar 已提交
3526
	/* Causes final put_task_struct in finish_task_switch(): */
3527
	set_special_state(TASK_DEAD);
I
Ingo Molnar 已提交
3528 3529 3530 3531

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

3532 3533
	__schedule(false);
	BUG();
I
Ingo Molnar 已提交
3534 3535

	/* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3536
	for (;;)
I
Ingo Molnar 已提交
3537
		cpu_relax();
3538 3539
}

3540 3541
static inline void sched_submit_work(struct task_struct *tsk)
{
3542
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3543 3544 3545 3546 3547 3548 3549 3550 3551
		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);
}

3552
asmlinkage __visible void __sched schedule(void)
3553
{
3554 3555 3556
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3557
	do {
3558
		preempt_disable();
3559
		__schedule(false);
3560
		sched_preempt_enable_no_resched();
3561
	} while (need_resched());
3562
}
L
Linus Torvalds 已提交
3563 3564
EXPORT_SYMBOL(schedule);

3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
/*
 * 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());
}

3590
#ifdef CONFIG_CONTEXT_TRACKING
3591
asmlinkage __visible void __sched schedule_user(void)
3592 3593 3594 3595 3596 3597
{
	/*
	 * 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.
3598 3599
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3600
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3601
	 * too frequently to make sense yet.
3602
	 */
3603
	enum ctx_state prev_state = exception_enter();
3604
	schedule();
3605
	exception_exit(prev_state);
3606 3607 3608
}
#endif

3609 3610 3611 3612 3613 3614 3615
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3616
	sched_preempt_enable_no_resched();
3617 3618 3619 3620
	schedule();
	preempt_disable();
}

3621
static void __sched notrace preempt_schedule_common(void)
3622 3623
{
	do {
3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636
		/*
		 * 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.
		 */
3637
		preempt_disable_notrace();
3638
		preempt_latency_start(1);
3639
		__schedule(true);
3640
		preempt_latency_stop(1);
3641
		preempt_enable_no_resched_notrace();
3642 3643 3644 3645 3646 3647 3648 3649

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

L
Linus Torvalds 已提交
3650 3651
#ifdef CONFIG_PREEMPT
/*
3652
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3653
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3654 3655
 * occur there and call schedule directly.
 */
3656
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3657 3658 3659
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3660
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3661
	 */
3662
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3663 3664
		return;

3665
	preempt_schedule_common();
L
Linus Torvalds 已提交
3666
}
3667
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3668
EXPORT_SYMBOL(preempt_schedule);
3669 3670

/**
3671
 * preempt_schedule_notrace - preempt_schedule called by tracing
3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683
 *
 * 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.
 */
3684
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3685 3686 3687 3688 3689 3690 3691
{
	enum ctx_state prev_ctx;

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

	do {
3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704
		/*
		 * 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.
		 */
3705
		preempt_disable_notrace();
3706
		preempt_latency_start(1);
3707 3708 3709 3710 3711 3712
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3713
		__schedule(true);
3714 3715
		exception_exit(prev_ctx);

3716
		preempt_latency_stop(1);
3717
		preempt_enable_no_resched_notrace();
3718 3719
	} while (need_resched());
}
3720
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3721

3722
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3723 3724

/*
3725
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3726 3727 3728 3729
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3730
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3731
{
3732
	enum ctx_state prev_state;
3733

3734
	/* Catch callers which need to be fixed */
3735
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3736

3737 3738
	prev_state = exception_enter();

3739
	do {
3740
		preempt_disable();
3741
		local_irq_enable();
3742
		__schedule(true);
3743
		local_irq_disable();
3744
		sched_preempt_enable_no_resched();
3745
	} while (need_resched());
3746 3747

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3748 3749
}

3750
int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3751
			  void *key)
L
Linus Torvalds 已提交
3752
{
P
Peter Zijlstra 已提交
3753
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3754 3755 3756
}
EXPORT_SYMBOL(default_wake_function);

3757 3758
#ifdef CONFIG_RT_MUTEXES

3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773
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);
}

3774 3775
/*
 * rt_mutex_setprio - set the current priority of a task
3776 3777
 * @p: task to boost
 * @pi_task: donor task
3778 3779 3780 3781
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3782 3783
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3784
 */
3785
void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3786
{
3787
	int prio, oldprio, queued, running, queue_flag =
3788
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3789
	const struct sched_class *prev_class;
3790 3791
	struct rq_flags rf;
	struct rq *rq;
3792

3793 3794 3795 3796 3797 3798 3799 3800
	/* 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;
3801

3802
	rq = __task_rq_lock(p, &rf);
3803
	update_rq_clock(rq);
3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820
	/*
	 * 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;
3821

3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839
	/*
	 * 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;
	}

3840
	trace_sched_pi_setprio(p, pi_task);
3841
	oldprio = p->prio;
3842 3843 3844 3845

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

3846
	prev_class = p->sched_class;
3847
	queued = task_on_rq_queued(p);
3848
	running = task_current(rq, p);
3849
	if (queued)
3850
		dequeue_task(rq, p, queue_flag);
3851
	if (running)
3852
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3853

3854 3855 3856 3857 3858 3859 3860 3861 3862 3863
	/*
	 * 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)) {
3864 3865
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3866
			p->dl.dl_boosted = 1;
3867
			queue_flag |= ENQUEUE_REPLENISH;
3868 3869
		} else
			p->dl.dl_boosted = 0;
3870
		p->sched_class = &dl_sched_class;
3871 3872 3873 3874
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3875
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3876
		p->sched_class = &rt_sched_class;
3877 3878 3879
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3880 3881
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3882
		p->sched_class = &fair_sched_class;
3883
	}
I
Ingo Molnar 已提交
3884

3885 3886
	p->prio = prio;

3887
	if (queued)
3888
		enqueue_task(rq, p, queue_flag);
3889
	if (running)
3890
		set_curr_task(rq, p);
3891

P
Peter Zijlstra 已提交
3892
	check_class_changed(rq, p, prev_class, oldprio);
3893
out_unlock:
I
Ingo Molnar 已提交
3894 3895
	/* Avoid rq from going away on us: */
	preempt_disable();
3896
	__task_rq_unlock(rq, &rf);
3897 3898 3899

	balance_callback(rq);
	preempt_enable();
3900
}
3901 3902 3903 3904 3905
#else
static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	return prio;
}
3906
#endif
3907

3908
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3909
{
P
Peter Zijlstra 已提交
3910 3911
	bool queued, running;
	int old_prio, delta;
3912
	struct rq_flags rf;
3913
	struct rq *rq;
L
Linus Torvalds 已提交
3914

3915
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3916 3917 3918 3919 3920
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3921
	rq = task_rq_lock(p, &rf);
3922 3923
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3924 3925 3926 3927
	/*
	 * 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
3928
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3929
	 */
3930
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3931 3932 3933
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3934
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3935
	running = task_current(rq, p);
3936
	if (queued)
3937
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
3938 3939
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3940 3941

	p->static_prio = NICE_TO_PRIO(nice);
3942
	set_load_weight(p, true);
3943 3944 3945
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3946

3947
	if (queued) {
3948
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
L
Linus Torvalds 已提交
3949
		/*
3950 3951
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3952
		 */
3953
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3954
			resched_curr(rq);
L
Linus Torvalds 已提交
3955
	}
P
Peter Zijlstra 已提交
3956 3957
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3958
out_unlock:
3959
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3960 3961 3962
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3963 3964 3965 3966 3967
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3968
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3969
{
I
Ingo Molnar 已提交
3970
	/* Convert nice value [19,-20] to rlimit style value [1,40]: */
3971
	int nice_rlim = nice_to_rlimit(nice);
3972

3973
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3974 3975 3976
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3977 3978 3979 3980 3981 3982 3983 3984 3985
#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.
 */
3986
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3987
{
3988
	long nice, retval;
L
Linus Torvalds 已提交
3989 3990 3991 3992 3993 3994

	/*
	 * 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.
	 */
3995
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3996
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3997

3998
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3999 4000 4001
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015
	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.
 *
4016
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
4017 4018 4019
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
4020
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
4021 4022 4023 4024 4025
{
	return p->prio - MAX_RT_PRIO;
}

/**
I
Ingo Molnar 已提交
4026
 * idle_cpu - is a given CPU idle currently?
L
Linus Torvalds 已提交
4027
 * @cpu: the processor in question.
4028 4029
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
4030 4031 4032
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046
	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
4047 4048
}

4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059
/**
 * available_idle_cpu - is a given CPU idle for enqueuing work.
 * @cpu: the CPU in question.
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
 */
int available_idle_cpu(int cpu)
{
	if (!idle_cpu(cpu))
		return 0;

4060 4061 4062
	if (vcpu_is_preempted(cpu))
		return 0;

T
Thomas Gleixner 已提交
4063
	return 1;
L
Linus Torvalds 已提交
4064 4065 4066
}

/**
I
Ingo Molnar 已提交
4067
 * idle_task - return the idle task for a given CPU.
L
Linus Torvalds 已提交
4068
 * @cpu: the processor in question.
4069
 *
I
Ingo Molnar 已提交
4070
 * Return: The idle task for the CPU @cpu.
L
Linus Torvalds 已提交
4071
 */
4072
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
4073 4074 4075 4076 4077 4078 4079
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
4080 4081
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
4082
 */
A
Alexey Dobriyan 已提交
4083
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
4084
{
4085
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
4086 4087
}

4088 4089 4090 4091 4092 4093
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

4094 4095
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
4096
{
4097 4098
	int policy = attr->sched_policy;

4099
	if (policy == SETPARAM_POLICY)
4100 4101
		policy = p->policy;

L
Linus Torvalds 已提交
4102
	p->policy = policy;
4103

4104 4105
	if (dl_policy(policy))
		__setparam_dl(p, attr);
4106
	else if (fair_policy(policy))
4107 4108
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

4109 4110 4111 4112 4113 4114
	/*
	 * __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;
4115
	p->normal_prio = normal_prio(p);
4116
	set_load_weight(p, true);
4117
}
4118

4119 4120
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
4121
			   const struct sched_attr *attr, bool keep_boost)
4122 4123
{
	__setscheduler_params(p, attr);
4124

4125
	/*
4126 4127
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
4128
	 */
4129
	p->prio = normal_prio(p);
4130
	if (keep_boost)
4131
		p->prio = rt_effective_prio(p, p->prio);
4132

4133 4134 4135
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
4136 4137 4138
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
4139
}
4140

4141
/*
I
Ingo Molnar 已提交
4142
 * Check the target process has a UID that matches the current process's:
4143 4144 4145 4146 4147 4148 4149 4150
 */
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);
4151 4152
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4153 4154 4155 4156
	rcu_read_unlock();
	return match;
}

4157 4158
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4159
				bool user, bool pi)
L
Linus Torvalds 已提交
4160
{
4161 4162
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4163
	int retval, oldprio, oldpolicy = -1, queued, running;
4164
	int new_effective_prio, policy = attr->sched_policy;
4165
	const struct sched_class *prev_class;
4166
	struct rq_flags rf;
4167
	int reset_on_fork;
4168
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4169
	struct rq *rq;
L
Linus Torvalds 已提交
4170

4171 4172
	/* The pi code expects interrupts enabled */
	BUG_ON(pi && in_interrupt());
L
Linus Torvalds 已提交
4173
recheck:
I
Ingo Molnar 已提交
4174
	/* Double check policy once rq lock held: */
4175 4176
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4177
		policy = oldpolicy = p->policy;
4178
	} else {
4179
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4180

4181
		if (!valid_policy(policy))
4182 4183 4184
			return -EINVAL;
	}

4185
	if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4186 4187
		return -EINVAL;

L
Linus Torvalds 已提交
4188 4189
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4190 4191
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4192
	 */
4193
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4194
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4195
		return -EINVAL;
4196 4197
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4198 4199
		return -EINVAL;

4200 4201 4202
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4203
	if (user && !capable(CAP_SYS_NICE)) {
4204
		if (fair_policy(policy)) {
4205
			if (attr->sched_nice < task_nice(p) &&
4206
			    !can_nice(p, attr->sched_nice))
4207 4208 4209
				return -EPERM;
		}

4210
		if (rt_policy(policy)) {
4211 4212
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4213

I
Ingo Molnar 已提交
4214
			/* Can't set/change the rt policy: */
4215 4216 4217
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

I
Ingo Molnar 已提交
4218
			/* Can't increase priority: */
4219 4220
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4221 4222
				return -EPERM;
		}
4223

4224 4225 4226 4227 4228 4229 4230 4231 4232
		 /*
		  * 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
4233
		/*
4234 4235
		 * 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
4236
		 */
4237
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4238
			if (!can_nice(p, task_nice(p)))
4239 4240
				return -EPERM;
		}
4241

I
Ingo Molnar 已提交
4242
		/* Can't change other user's priorities: */
4243
		if (!check_same_owner(p))
4244
			return -EPERM;
4245

I
Ingo Molnar 已提交
4246
		/* Normal users shall not reset the sched_reset_on_fork flag: */
4247 4248
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4249
	}
L
Linus Torvalds 已提交
4250

4251
	if (user) {
4252 4253 4254
		if (attr->sched_flags & SCHED_FLAG_SUGOV)
			return -EINVAL;

4255
		retval = security_task_setscheduler(p);
4256 4257 4258 4259
		if (retval)
			return retval;
	}

4260
	/*
I
Ingo Molnar 已提交
4261
	 * Make sure no PI-waiters arrive (or leave) while we are
4262
	 * changing the priority of the task:
4263
	 *
L
Lucas De Marchi 已提交
4264
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4265 4266
	 * runqueue lock must be held.
	 */
4267
	rq = task_rq_lock(p, &rf);
4268
	update_rq_clock(rq);
4269

4270
	/*
I
Ingo Molnar 已提交
4271
	 * Changing the policy of the stop threads its a very bad idea:
4272 4273
	 */
	if (p == rq->stop) {
4274
		task_rq_unlock(rq, p, &rf);
4275 4276 4277
		return -EINVAL;
	}

4278
	/*
4279 4280
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4281
	 */
4282
	if (unlikely(policy == p->policy)) {
4283
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4284 4285 4286
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4287
		if (dl_policy(policy) && dl_param_changed(p, attr))
4288
			goto change;
4289

4290
		p->sched_reset_on_fork = reset_on_fork;
4291
		task_rq_unlock(rq, p, &rf);
4292 4293
		return 0;
	}
4294
change:
4295

4296
	if (user) {
4297
#ifdef CONFIG_RT_GROUP_SCHED
4298 4299 4300 4301 4302
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4303 4304
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4305
			task_rq_unlock(rq, p, &rf);
4306 4307 4308
			return -EPERM;
		}
#endif
4309
#ifdef CONFIG_SMP
4310 4311
		if (dl_bandwidth_enabled() && dl_policy(policy) &&
				!(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4312 4313 4314 4315 4316 4317 4318
			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.
			 */
4319 4320
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4321
				task_rq_unlock(rq, p, &rf);
4322 4323 4324 4325 4326
				return -EPERM;
			}
		}
#endif
	}
4327

I
Ingo Molnar 已提交
4328
	/* Re-check policy now with rq lock held: */
L
Linus Torvalds 已提交
4329 4330
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4331
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4332 4333
		goto recheck;
	}
4334 4335 4336 4337 4338 4339

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4340
	if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4341
		task_rq_unlock(rq, p, &rf);
4342 4343 4344
		return -EBUSY;
	}

4345 4346 4347
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4348 4349 4350 4351 4352 4353 4354 4355
	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.
		 */
4356
		new_effective_prio = rt_effective_prio(p, newprio);
4357 4358
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4359 4360
	}

4361
	queued = task_on_rq_queued(p);
4362
	running = task_current(rq, p);
4363
	if (queued)
4364
		dequeue_task(rq, p, queue_flags);
4365
	if (running)
4366
		put_prev_task(rq, p);
4367

4368
	prev_class = p->sched_class;
4369
	__setscheduler(rq, p, attr, pi);
4370

4371
	if (queued) {
4372 4373 4374 4375
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4376 4377
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4378

4379
		enqueue_task(rq, p, queue_flags);
4380
	}
4381
	if (running)
4382
		set_curr_task(rq, p);
4383

P
Peter Zijlstra 已提交
4384
	check_class_changed(rq, p, prev_class, oldprio);
I
Ingo Molnar 已提交
4385 4386 4387

	/* Avoid rq from going away on us: */
	preempt_disable();
4388
	task_rq_unlock(rq, p, &rf);
4389

4390 4391
	if (pi)
		rt_mutex_adjust_pi(p);
4392

I
Ingo Molnar 已提交
4393
	/* Run balance callbacks after we've adjusted the PI chain: */
4394 4395
	balance_callback(rq);
	preempt_enable();
4396

L
Linus Torvalds 已提交
4397 4398
	return 0;
}
4399

4400 4401 4402 4403 4404 4405 4406 4407 4408
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),
	};

4409 4410
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4411 4412 4413 4414 4415
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4416
	return __sched_setscheduler(p, &attr, check, true);
4417
}
4418 4419 4420 4421 4422 4423
/**
 * 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.
 *
4424 4425
 * Return: 0 on success. An error code otherwise.
 *
4426 4427 4428
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4429
		       const struct sched_param *param)
4430
{
4431
	return _sched_setscheduler(p, policy, param, true);
4432
}
L
Linus Torvalds 已提交
4433 4434
EXPORT_SYMBOL_GPL(sched_setscheduler);

4435 4436
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4437
	return __sched_setscheduler(p, attr, true, true);
4438 4439 4440
}
EXPORT_SYMBOL_GPL(sched_setattr);

4441 4442 4443 4444 4445
int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
{
	return __sched_setscheduler(p, attr, false, true);
}

4446 4447 4448 4449 4450 4451 4452 4453 4454 4455
/**
 * 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.
4456 4457
 *
 * Return: 0 on success. An error code otherwise.
4458 4459
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4460
			       const struct sched_param *param)
4461
{
4462
	return _sched_setscheduler(p, policy, param, false);
4463
}
4464
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4465

I
Ingo Molnar 已提交
4466 4467
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4468 4469 4470
{
	struct sched_param lparam;
	struct task_struct *p;
4471
	int retval;
L
Linus Torvalds 已提交
4472 4473 4474 4475 4476

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4477 4478 4479

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4480
	p = find_process_by_pid(pid);
4481 4482 4483
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4484

L
Linus Torvalds 已提交
4485 4486 4487
	return retval;
}

4488 4489 4490
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
I
Ingo Molnar 已提交
4491
static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4492 4493 4494 4495 4496 4497 4498
{
	u32 size;
	int ret;

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

I
Ingo Molnar 已提交
4499
	/* Zero the full structure, so that a short copy will be nice: */
4500 4501 4502 4503 4504 4505
	memset(attr, 0, sizeof(*attr));

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

I
Ingo Molnar 已提交
4506 4507
	/* Bail out on silly large: */
	if (size > PAGE_SIZE)
4508 4509
		goto err_size;

I
Ingo Molnar 已提交
4510 4511
	/* ABI compatibility quirk: */
	if (!size)
4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545
		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
4546
	 * XXX: Do we want to be lenient like existing syscalls; or do we want
4547 4548
	 * to be strict and return an error on out-of-bounds values?
	 */
4549
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4550

4551
	return 0;
4552 4553 4554

err_size:
	put_user(sizeof(*attr), &uattr->size);
4555
	return -E2BIG;
4556 4557
}

L
Linus Torvalds 已提交
4558 4559 4560 4561 4562
/**
 * 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.
4563 4564
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4565
 */
I
Ingo Molnar 已提交
4566
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4567
{
4568 4569 4570
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4571 4572 4573 4574 4575 4576 4577
	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.
4578 4579
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4580
 */
4581
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4582
{
4583
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4584 4585
}

4586 4587 4588
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4589
 * @uattr: structure containing the extended parameters.
4590
 * @flags: for future extension.
4591
 */
4592 4593
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4594 4595 4596 4597 4598
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4599
	if (!uattr || pid < 0 || flags)
4600 4601
		return -EINVAL;

4602 4603 4604
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4605

4606
	if ((int)attr.sched_policy < 0)
4607
		return -EINVAL;
4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618

	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
4619 4620 4621
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4622 4623 4624
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4625
 */
4626
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4627
{
4628
	struct task_struct *p;
4629
	int retval;
L
Linus Torvalds 已提交
4630 4631

	if (pid < 0)
4632
		return -EINVAL;
L
Linus Torvalds 已提交
4633 4634

	retval = -ESRCH;
4635
	rcu_read_lock();
L
Linus Torvalds 已提交
4636 4637 4638 4639
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4640 4641
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4642
	}
4643
	rcu_read_unlock();
L
Linus Torvalds 已提交
4644 4645 4646 4647
	return retval;
}

/**
4648
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4649 4650
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4651 4652 4653
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4654
 */
4655
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4656
{
4657
	struct sched_param lp = { .sched_priority = 0 };
4658
	struct task_struct *p;
4659
	int retval;
L
Linus Torvalds 已提交
4660 4661

	if (!param || pid < 0)
4662
		return -EINVAL;
L
Linus Torvalds 已提交
4663

4664
	rcu_read_lock();
L
Linus Torvalds 已提交
4665 4666 4667 4668 4669 4670 4671 4672 4673
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4674 4675
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4676
	rcu_read_unlock();
L
Linus Torvalds 已提交
4677 4678 4679 4680 4681 4682 4683 4684 4685

	/*
	 * 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:
4686
	rcu_read_unlock();
L
Linus Torvalds 已提交
4687 4688 4689
	return retval;
}

4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712
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)
4713
				return -EFBIG;
4714 4715 4716 4717 4718
		}

		attr->size = usize;
	}

4719
	ret = copy_to_user(uattr, attr, attr->size);
4720 4721 4722
	if (ret)
		return -EFAULT;

4723
	return 0;
4724 4725 4726
}

/**
4727
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4728
 * @pid: the pid in question.
J
Juri Lelli 已提交
4729
 * @uattr: structure containing the extended parameters.
4730
 * @size: sizeof(attr) for fwd/bwd comp.
4731
 * @flags: for future extension.
4732
 */
4733 4734
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4735 4736 4737 4738 4739 4740 4741 4742
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4743
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756
		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;
4757 4758
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4759 4760 4761
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4762 4763
		attr.sched_priority = p->rt_priority;
	else
4764
		attr.sched_nice = task_nice(p);
4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4776
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4777
{
4778
	cpumask_var_t cpus_allowed, new_mask;
4779 4780
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4781

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

	p = find_process_by_pid(pid);
	if (!p) {
4786
		rcu_read_unlock();
L
Linus Torvalds 已提交
4787 4788 4789
		return -ESRCH;
	}

4790
	/* Prevent p going away */
L
Linus Torvalds 已提交
4791
	get_task_struct(p);
4792
	rcu_read_unlock();
L
Linus Torvalds 已提交
4793

4794 4795 4796 4797
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4798 4799 4800 4801 4802 4803 4804 4805
	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
4806
	retval = -EPERM;
E
Eric W. Biederman 已提交
4807 4808 4809 4810
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4811
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4812 4813 4814
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4815

4816
	retval = security_task_setscheduler(p);
4817
	if (retval)
4818
		goto out_free_new_mask;
4819

4820 4821 4822 4823

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

4824 4825 4826 4827 4828 4829 4830
	/*
	 * 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
4831 4832 4833
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4834
			retval = -EBUSY;
4835
			rcu_read_unlock();
4836
			goto out_free_new_mask;
4837
		}
4838
		rcu_read_unlock();
4839 4840
	}
#endif
P
Peter Zijlstra 已提交
4841
again:
4842
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4843

P
Paul Menage 已提交
4844
	if (!retval) {
4845 4846
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4847 4848 4849 4850 4851
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4852
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4853 4854 4855
			goto again;
		}
	}
4856
out_free_new_mask:
4857 4858 4859 4860
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4861 4862 4863 4864 4865
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4866
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4867
{
4868 4869 4870 4871 4872
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4873 4874 4875 4876
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
I
Ingo Molnar 已提交
4877
 * sys_sched_setaffinity - set the CPU affinity of a process
L
Linus Torvalds 已提交
4878 4879
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4880
 * @user_mask_ptr: user-space pointer to the new CPU mask
4881 4882
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4883
 */
4884 4885
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4886
{
4887
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4888 4889
	int retval;

4890 4891
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4892

4893 4894 4895 4896 4897
	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
4898 4899
}

4900
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4901
{
4902
	struct task_struct *p;
4903
	unsigned long flags;
L
Linus Torvalds 已提交
4904 4905
	int retval;

4906
	rcu_read_lock();
L
Linus Torvalds 已提交
4907 4908 4909 4910 4911 4912

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

4913 4914 4915 4916
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4917
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4918
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4919
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4920 4921

out_unlock:
4922
	rcu_read_unlock();
L
Linus Torvalds 已提交
4923

4924
	return retval;
L
Linus Torvalds 已提交
4925 4926 4927
}

/**
I
Ingo Molnar 已提交
4928
 * sys_sched_getaffinity - get the CPU affinity of a process
L
Linus Torvalds 已提交
4929 4930
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4931
 * @user_mask_ptr: user-space pointer to hold the current CPU mask
4932
 *
4933 4934
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4935
 */
4936 4937
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4938 4939
{
	int ret;
4940
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4941

A
Anton Blanchard 已提交
4942
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4943 4944
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4945 4946
		return -EINVAL;

4947 4948
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4949

4950 4951
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4952
		unsigned int retlen = min(len, cpumask_size());
4953 4954

		if (copy_to_user(user_mask_ptr, mask, retlen))
4955 4956
			ret = -EFAULT;
		else
4957
			ret = retlen;
4958 4959
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4960

4961
	return ret;
L
Linus Torvalds 已提交
4962 4963 4964 4965 4966
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4967 4968
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4969 4970
 *
 * Return: 0.
L
Linus Torvalds 已提交
4971
 */
4972
static void do_sched_yield(void)
L
Linus Torvalds 已提交
4973
{
4974 4975 4976 4977 4978 4979
	struct rq_flags rf;
	struct rq *rq;

	local_irq_disable();
	rq = this_rq();
	rq_lock(rq, &rf);
L
Linus Torvalds 已提交
4980

4981
	schedstat_inc(rq->yld_count);
4982
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4983 4984 4985 4986 4987

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
4988 4989
	preempt_disable();
	rq_unlock(rq, &rf);
4990
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4991 4992

	schedule();
4993
}
L
Linus Torvalds 已提交
4994

4995 4996 4997
SYSCALL_DEFINE0(sched_yield)
{
	do_sched_yield();
L
Linus Torvalds 已提交
4998 4999 5000
	return 0;
}

5001
#ifndef CONFIG_PREEMPT
5002
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
5003
{
5004
	if (should_resched(0)) {
5005
		preempt_schedule_common();
L
Linus Torvalds 已提交
5006 5007
		return 1;
	}
5008
	rcu_all_qs();
L
Linus Torvalds 已提交
5009 5010
	return 0;
}
5011
EXPORT_SYMBOL(_cond_resched);
5012
#endif
L
Linus Torvalds 已提交
5013 5014

/*
5015
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
5016 5017
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
5018
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
5019 5020 5021
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
5022
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
5023
{
5024
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
5025 5026
	int ret = 0;

5027 5028
	lockdep_assert_held(lock);

5029
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
5030
		spin_unlock(lock);
P
Peter Zijlstra 已提交
5031
		if (resched)
5032
			preempt_schedule_common();
N
Nick Piggin 已提交
5033 5034
		else
			cpu_relax();
J
Jan Kara 已提交
5035
		ret = 1;
L
Linus Torvalds 已提交
5036 5037
		spin_lock(lock);
	}
J
Jan Kara 已提交
5038
	return ret;
L
Linus Torvalds 已提交
5039
}
5040
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
5041 5042 5043 5044

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
5045 5046 5047 5048 5049 5050 5051 5052 5053
 * 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
5054
 *	yield();
P
Peter Zijlstra 已提交
5055 5056 5057 5058 5059 5060 5061 5062
 *
 * 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
5063 5064 5065 5066
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
5067
	do_sched_yield();
L
Linus Torvalds 已提交
5068 5069 5070
}
EXPORT_SYMBOL(yield);

5071 5072 5073 5074
/**
 * 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
5075 5076
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5077 5078 5079 5080
 *
 * 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.
 *
5081
 * Return:
5082 5083 5084
 *	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.
5085
 */
5086
int __sched yield_to(struct task_struct *p, bool preempt)
5087 5088 5089 5090
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5091
	int yielded = 0;
5092 5093 5094 5095 5096 5097

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5098 5099 5100 5101 5102 5103 5104 5105 5106
	/*
	 * 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;
	}

5107
	double_rq_lock(rq, p_rq);
5108
	if (task_rq(p) != p_rq) {
5109 5110 5111 5112 5113
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5114
		goto out_unlock;
5115 5116

	if (curr->sched_class != p->sched_class)
5117
		goto out_unlock;
5118 5119

	if (task_running(p_rq, p) || p->state)
5120
		goto out_unlock;
5121 5122

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5123
	if (yielded) {
5124
		schedstat_inc(rq->yld_count);
5125 5126 5127 5128 5129
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5130
			resched_curr(p_rq);
5131
	}
5132

5133
out_unlock:
5134
	double_rq_unlock(rq, p_rq);
5135
out_irq:
5136 5137
	local_irq_restore(flags);

5138
	if (yielded > 0)
5139 5140 5141 5142 5143 5144
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159
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
5160
/*
I
Ingo Molnar 已提交
5161
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5162 5163 5164 5165
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5166
	int token;
L
Linus Torvalds 已提交
5167 5168
	long ret;

5169
	token = io_schedule_prepare();
L
Linus Torvalds 已提交
5170
	ret = schedule_timeout(timeout);
5171
	io_schedule_finish(token);
5172

L
Linus Torvalds 已提交
5173 5174
	return ret;
}
5175
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5176

5177
void __sched io_schedule(void)
5178 5179 5180 5181 5182 5183 5184 5185 5186
{
	int token;

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

L
Linus Torvalds 已提交
5187 5188 5189 5190
/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5191 5192 5193
 * 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
5194
 */
5195
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5196 5197 5198 5199 5200 5201 5202 5203
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5204
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5205
	case SCHED_NORMAL:
5206
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5207
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5208 5209 5210 5211 5212 5213 5214 5215 5216 5217
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5218 5219 5220
 * 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
5221
 */
5222
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5223 5224 5225 5226 5227 5228 5229 5230
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5231
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5232
	case SCHED_NORMAL:
5233
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5234
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5235 5236 5237 5238 5239
		ret = 0;
	}
	return ret;
}

5240
static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
L
Linus Torvalds 已提交
5241
{
5242
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5243
	unsigned int time_slice;
5244
	struct rq_flags rf;
5245
	struct rq *rq;
5246
	int retval;
L
Linus Torvalds 已提交
5247 5248

	if (pid < 0)
5249
		return -EINVAL;
L
Linus Torvalds 已提交
5250 5251

	retval = -ESRCH;
5252
	rcu_read_lock();
L
Linus Torvalds 已提交
5253 5254 5255 5256 5257 5258 5259 5260
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5261
	rq = task_rq_lock(p, &rf);
5262 5263 5264
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5265
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5266

5267
	rcu_read_unlock();
5268 5269
	jiffies_to_timespec64(time_slice, t);
	return 0;
5270

L
Linus Torvalds 已提交
5271
out_unlock:
5272
	rcu_read_unlock();
L
Linus Torvalds 已提交
5273 5274 5275
	return retval;
}

5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286
/**
 * 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.
 */
5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312
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

5313
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5314 5315
{
	unsigned long free = 0;
5316
	int ppid;
5317

5318 5319
	if (!try_get_task_stack(p))
		return;
5320 5321 5322 5323

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

	if (p->state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5324
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5325
#ifdef CONFIG_DEBUG_STACK_USAGE
5326
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5327
#endif
5328
	ppid = 0;
5329
	rcu_read_lock();
5330 5331
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5332
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5333
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5334
		task_pid_nr(p), ppid,
5335
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5336

5337
	print_worker_info(KERN_INFO, p);
5338
	show_stack(p, NULL);
5339
	put_task_stack(p);
L
Linus Torvalds 已提交
5340
}
5341
EXPORT_SYMBOL_GPL(sched_show_task);
L
Linus Torvalds 已提交
5342

5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364
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
5365
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5366
{
5367
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5368

5369
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5370 5371
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5372
#else
P
Peter Zijlstra 已提交
5373 5374
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5375
#endif
5376
	rcu_read_lock();
5377
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5378 5379
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5380
		 * console might take a lot of time:
5381 5382 5383
		 * 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
5384 5385
		 */
		touch_nmi_watchdog();
5386
		touch_all_softlockup_watchdogs();
5387
		if (state_filter_match(state_filter, p))
5388
			sched_show_task(p);
5389
	}
L
Linus Torvalds 已提交
5390

I
Ingo Molnar 已提交
5391
#ifdef CONFIG_SCHED_DEBUG
5392 5393
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5394
#endif
5395
	rcu_read_unlock();
I
Ingo Molnar 已提交
5396 5397 5398
	/*
	 * Only show locks if all tasks are dumped:
	 */
5399
	if (!state_filter)
I
Ingo Molnar 已提交
5400
		debug_show_all_locks();
L
Linus Torvalds 已提交
5401 5402
}

5403 5404 5405
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
I
Ingo Molnar 已提交
5406
 * @cpu: CPU the idle task belongs to
5407 5408 5409 5410
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5411
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5412
{
5413
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5414 5415
	unsigned long flags;

5416 5417
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5418

5419
	__sched_fork(0, idle);
5420
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5421
	idle->se.exec_start = sched_clock();
5422
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5423

5424 5425
	kasan_unpoison_task_stack(idle);

5426 5427 5428 5429 5430 5431 5432 5433 5434
#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
5435 5436
	/*
	 * We're having a chicken and egg problem, even though we are
I
Ingo Molnar 已提交
5437
	 * holding rq->lock, the CPU isn't yet set to this CPU so the
5438 5439 5440 5441 5442 5443 5444 5445
	 * 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
5446
	__set_task_cpu(idle, cpu);
5447
	rcu_read_unlock();
L
Linus Torvalds 已提交
5448 5449

	rq->curr = rq->idle = idle;
5450
	idle->on_rq = TASK_ON_RQ_QUEUED;
5451
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5452
	idle->on_cpu = 1;
5453
#endif
5454 5455
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5456 5457

	/* Set the preempt count _outside_ the spinlocks! */
5458
	init_idle_preempt_count(idle, cpu);
5459

I
Ingo Molnar 已提交
5460 5461 5462 5463
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5464
	ftrace_graph_init_idle_task(idle, cpu);
5465
	vtime_init_idle(idle, cpu);
5466
#ifdef CONFIG_SMP
5467 5468
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5469 5470
}

5471 5472
#ifdef CONFIG_SMP

5473 5474 5475
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
5476
	int ret = 1;
5477

5478 5479 5480
	if (!cpumask_weight(cur))
		return ret;

5481
	ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5482 5483 5484 5485

	return ret;
}

5486 5487 5488 5489 5490 5491 5492
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
5493
	 * to a new cpuset; we don't want to change their CPU
5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505
	 * 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,
5506 5507
					      cs_cpus_allowed))
		ret = dl_task_can_attach(p, cs_cpus_allowed);
5508 5509 5510 5511 5512

out:
	return ret;
}

5513
bool sched_smp_initialized __read_mostly;
5514

5515 5516 5517 5518 5519 5520 5521 5522 5523 5524
#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;

5525
	if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5526 5527 5528 5529
		return -EINVAL;

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

5530
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5531 5532
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5533 5534 5535 5536 5537 5538 5539

/*
 * 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)
{
5540
	bool queued, running;
5541 5542
	struct rq_flags rf;
	struct rq *rq;
5543

5544
	rq = task_rq_lock(p, &rf);
5545
	queued = task_on_rq_queued(p);
5546 5547
	running = task_current(rq, p);

5548
	if (queued)
5549
		dequeue_task(rq, p, DEQUEUE_SAVE);
5550
	if (running)
5551
		put_prev_task(rq, p);
5552 5553 5554

	p->numa_preferred_nid = nid;

5555
	if (queued)
5556
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5557
	if (running)
5558
		set_curr_task(rq, p);
5559
	task_rq_unlock(rq, p, &rf);
5560
}
P
Peter Zijlstra 已提交
5561
#endif /* CONFIG_NUMA_BALANCING */
5562

L
Linus Torvalds 已提交
5563
#ifdef CONFIG_HOTPLUG_CPU
5564
/*
I
Ingo Molnar 已提交
5565
 * Ensure that the idle task is using init_mm right before its CPU goes
5566
 * offline.
5567
 */
5568
void idle_task_exit(void)
L
Linus Torvalds 已提交
5569
{
5570
	struct mm_struct *mm = current->active_mm;
5571

5572
	BUG_ON(cpu_online(smp_processor_id()));
5573

5574
	if (mm != &init_mm) {
5575
		switch_mm(mm, &init_mm, current);
5576
		current->active_mm = &init_mm;
5577 5578
		finish_arch_post_lock_switch();
	}
5579
	mmdrop(mm);
L
Linus Torvalds 已提交
5580 5581 5582
}

/*
5583 5584
 * 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
5585 5586 5587
 * 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.
5588 5589
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5590
 */
5591
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5592
{
5593
	long delta = calc_load_fold_active(rq, 1);
5594 5595
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5596 5597
}

5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613
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,
};

5614
/*
5615 5616 5617 5618 5619 5620
 * 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
5621
 */
5622
static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
L
Linus Torvalds 已提交
5623
{
5624
	struct rq *rq = dead_rq;
5625
	struct task_struct *next, *stop = rq->stop;
5626
	struct rq_flags orf = *rf;
5627
	int dest_cpu;
L
Linus Torvalds 已提交
5628 5629

	/*
5630 5631 5632 5633 5634 5635 5636
	 * 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
5637
	 */
5638
	rq->stop = NULL;
5639

5640 5641 5642 5643 5644 5645 5646
	/*
	 * 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);

5647
	for (;;) {
5648 5649
		/*
		 * There's this thread running, bail when that's the only
I
Ingo Molnar 已提交
5650
		 * remaining thread:
5651 5652
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5653
			break;
5654

5655
		/*
I
Ingo Molnar 已提交
5656
		 * pick_next_task() assumes pinned rq->lock:
5657
		 */
5658
		next = pick_next_task(rq, &fake_task, rf);
5659
		BUG_ON(!next);
V
Viresh Kumar 已提交
5660
		put_prev_task(rq, next);
5661

W
Wanpeng Li 已提交
5662 5663 5664 5665 5666 5667 5668 5669 5670
		/*
		 * 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.
		 */
5671
		rq_unlock(rq, rf);
W
Wanpeng Li 已提交
5672
		raw_spin_lock(&next->pi_lock);
5673
		rq_relock(rq, rf);
W
Wanpeng Li 已提交
5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684

		/*
		 * 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;
		}

5685
		/* Find suitable destination for @next, with force if needed. */
5686
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5687
		rq = __migrate_task(rq, rf, next, dest_cpu);
5688
		if (rq != dead_rq) {
5689
			rq_unlock(rq, rf);
5690
			rq = dead_rq;
5691 5692
			*rf = orf;
			rq_relock(rq, rf);
5693
		}
W
Wanpeng Li 已提交
5694
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5695
	}
5696

5697
	rq->stop = stop;
5698
}
L
Linus Torvalds 已提交
5699 5700
#endif /* CONFIG_HOTPLUG_CPU */

5701
void set_rq_online(struct rq *rq)
5702 5703 5704 5705
{
	if (!rq->online) {
		const struct sched_class *class;

5706
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5707 5708 5709 5710 5711 5712 5713 5714 5715
		rq->online = 1;

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

5716
void set_rq_offline(struct rq *rq)
5717 5718 5719 5720 5721 5722 5723 5724 5725
{
	if (rq->online) {
		const struct sched_class *class;

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

5726
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5727 5728 5729 5730
		rq->online = 0;
	}
}

I
Ingo Molnar 已提交
5731 5732 5733 5734
/*
 * used to mark begin/end of suspend/resume:
 */
static int num_cpus_frozen;
5735

L
Linus Torvalds 已提交
5736
/*
5737 5738 5739
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
5740 5741 5742
 *
 * 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
5743
 */
5744
static void cpuset_cpu_active(void)
5745
{
5746
	if (cpuhp_tasks_frozen) {
5747 5748 5749 5750 5751 5752
		/*
		 * 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.
		 */
5753 5754
		partition_sched_domains(1, NULL, NULL);
		if (--num_cpus_frozen)
5755
			return;
5756 5757 5758 5759 5760
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
5761
		cpuset_force_rebuild();
5762
	}
5763
	cpuset_update_active_cpus();
5764
}
5765

5766
static int cpuset_cpu_inactive(unsigned int cpu)
5767
{
5768
	if (!cpuhp_tasks_frozen) {
5769
		if (dl_cpu_busy(cpu))
5770
			return -EBUSY;
5771
		cpuset_update_active_cpus();
5772
	} else {
5773 5774
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
5775
	}
5776
	return 0;
5777 5778
}

5779
int sched_cpu_activate(unsigned int cpu)
5780
{
5781
	struct rq *rq = cpu_rq(cpu);
5782
	struct rq_flags rf;
5783

5784 5785
#ifdef CONFIG_SCHED_SMT
	/*
5786
	 * When going up, increment the number of cores with SMT present.
5787
	 */
5788 5789
	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
		static_branch_inc_cpuslocked(&sched_smt_present);
5790
#endif
5791
	set_cpu_active(cpu, true);
5792

5793
	if (sched_smp_initialized) {
5794
		sched_domains_numa_masks_set(cpu);
5795
		cpuset_cpu_active();
5796
	}
5797 5798 5799 5800 5801

	/*
	 * 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
5802
	 *    after all CPUs have been brought up.
5803 5804 5805 5806
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
5807
	rq_lock_irqsave(rq, &rf);
5808 5809 5810 5811
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
5812
	rq_unlock_irqrestore(rq, &rf);
5813 5814 5815

	update_max_interval();

5816
	return 0;
5817 5818
}

5819
int sched_cpu_deactivate(unsigned int cpu)
5820 5821 5822
{
	int ret;

5823
	set_cpu_active(cpu, false);
5824 5825 5826 5827 5828 5829 5830
	/*
	 * 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.
	 */
5831
	synchronize_rcu_mult(call_rcu, call_rcu_sched);
5832

5833 5834 5835 5836 5837 5838 5839 5840
#ifdef CONFIG_SCHED_SMT
	/*
	 * When going down, decrement the number of cores with SMT present.
	 */
	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
		static_branch_dec_cpuslocked(&sched_smt_present);
#endif

5841 5842 5843 5844 5845 5846 5847
	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
5848
	}
5849 5850
	sched_domains_numa_masks_clear(cpu);
	return 0;
5851 5852
}

5853 5854 5855 5856 5857 5858 5859 5860
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();
}

5861 5862
int sched_cpu_starting(unsigned int cpu)
{
5863
	sched_rq_cpu_starting(cpu);
5864
	sched_tick_start(cpu);
5865
	return 0;
5866 5867
}

5868 5869 5870 5871
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5872
	struct rq_flags rf;
5873 5874 5875

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
5876
	sched_tick_stop(cpu);
5877 5878

	rq_lock_irqsave(rq, &rf);
5879 5880 5881 5882
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
5883
	migrate_tasks(rq, &rf);
5884
	BUG_ON(rq->nr_running != 1);
5885 5886
	rq_unlock_irqrestore(rq, &rf);

5887 5888
	calc_load_migrate(rq);
	update_max_interval();
5889
	nohz_balance_exit_idle(rq);
5890
	hrtick_clear(rq);
5891 5892 5893 5894
	return 0;
}
#endif

L
Linus Torvalds 已提交
5895 5896
void __init sched_init_smp(void)
{
5897 5898
	sched_init_numa();

5899 5900
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
I
Ingo Molnar 已提交
5901
	 * CPU masks are stable and all blatant races in the below code cannot
5902 5903
	 * happen. The hotplug lock is nevertheless taken to satisfy lockdep,
	 * but there won't be any contention on it.
5904
	 */
5905
	cpus_read_lock();
5906
	mutex_lock(&sched_domains_mutex);
P
Peter Zijlstra 已提交
5907
	sched_init_domains(cpu_active_mask);
5908
	mutex_unlock(&sched_domains_mutex);
5909
	cpus_read_unlock();
5910

5911
	/* Move init over to a non-isolated CPU */
5912
	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5913
		BUG();
I
Ingo Molnar 已提交
5914
	sched_init_granularity();
5915

5916
	init_sched_rt_class();
5917
	init_sched_dl_class();
P
Peter Zijlstra 已提交
5918

5919
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
5920
}
5921 5922 5923

static int __init migration_init(void)
{
5924
	sched_rq_cpu_starting(smp_processor_id());
5925
	return 0;
L
Linus Torvalds 已提交
5926
}
5927 5928
early_initcall(migration_init);

L
Linus Torvalds 已提交
5929 5930 5931
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
5932
	sched_init_granularity();
L
Linus Torvalds 已提交
5933 5934 5935 5936 5937 5938 5939 5940 5941 5942
}
#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);
}

5943
#ifdef CONFIG_CGROUP_SCHED
5944 5945 5946 5947
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
5948
struct task_group root_task_group;
5949
LIST_HEAD(task_groups);
5950 5951 5952

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
5953
#endif
P
Peter Zijlstra 已提交
5954

5955
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5956
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
5957

L
Linus Torvalds 已提交
5958 5959
void __init sched_init(void)
{
I
Ingo Molnar 已提交
5960
	int i, j;
5961 5962
	unsigned long alloc_size = 0, ptr;

5963
	wait_bit_init();
5964

5965 5966 5967 5968 5969 5970 5971
#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) {
5972
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5973 5974

#ifdef CONFIG_FAIR_GROUP_SCHED
5975
		root_task_group.se = (struct sched_entity **)ptr;
5976 5977
		ptr += nr_cpu_ids * sizeof(void **);

5978
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5979
		ptr += nr_cpu_ids * sizeof(void **);
5980

5981
#endif /* CONFIG_FAIR_GROUP_SCHED */
5982
#ifdef CONFIG_RT_GROUP_SCHED
5983
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5984 5985
		ptr += nr_cpu_ids * sizeof(void **);

5986
		root_task_group.rt_rq = (struct rt_rq **)ptr;
5987 5988
		ptr += nr_cpu_ids * sizeof(void **);

5989
#endif /* CONFIG_RT_GROUP_SCHED */
5990
	}
5991
#ifdef CONFIG_CPUMASK_OFFSTACK
5992 5993 5994
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5995 5996
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5997
	}
5998
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
5999

I
Ingo Molnar 已提交
6000 6001
	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
6002

G
Gregory Haskins 已提交
6003 6004 6005 6006
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

6007
#ifdef CONFIG_RT_GROUP_SCHED
6008
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
6009
			global_rt_period(), global_rt_runtime());
6010
#endif /* CONFIG_RT_GROUP_SCHED */
6011

D
Dhaval Giani 已提交
6012
#ifdef CONFIG_CGROUP_SCHED
6013 6014
	task_group_cache = KMEM_CACHE(task_group, 0);

6015 6016
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
6017
	INIT_LIST_HEAD(&root_task_group.siblings);
6018
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
6019
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
6020

6021
	for_each_possible_cpu(i) {
6022
		struct rq *rq;
L
Linus Torvalds 已提交
6023 6024

		rq = cpu_rq(i);
6025
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
6026
		rq->nr_running = 0;
6027 6028
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
6029
		init_cfs_rq(&rq->cfs);
6030 6031
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
6032
#ifdef CONFIG_FAIR_GROUP_SCHED
6033
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
6034
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
6035
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
6036
		/*
I
Ingo Molnar 已提交
6037
		 * How much CPU bandwidth does root_task_group get?
D
Dhaval Giani 已提交
6038 6039
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
I
Ingo Molnar 已提交
6040 6041
		 * gets 100% of the CPU resources in the system. This overall
		 * system CPU resource is divided among the tasks of
6042
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
6043 6044 6045
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
6046
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
6047
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
I
Ingo Molnar 已提交
6048
		 * then A0's share of the CPU resource is:
D
Dhaval Giani 已提交
6049
		 *
6050
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
6051
		 *
6052 6053
		 * 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
6054
		 */
6055
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6056
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
6057 6058 6059
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6060
#ifdef CONFIG_RT_GROUP_SCHED
6061
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
6062
#endif
L
Linus Torvalds 已提交
6063

I
Ingo Molnar 已提交
6064 6065
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
6066

L
Linus Torvalds 已提交
6067
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6068
		rq->sd = NULL;
G
Gregory Haskins 已提交
6069
		rq->rd = NULL;
6070
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6071
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
6072
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6073
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6074
		rq->push_cpu = 0;
6075
		rq->cpu = i;
6076
		rq->online = 0;
6077 6078
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6079
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6080 6081 6082

		INIT_LIST_HEAD(&rq->cfs_tasks);

6083
		rq_attach_root(rq, &def_root_domain);
6084
#ifdef CONFIG_NO_HZ_COMMON
6085
		rq->last_load_update_tick = jiffies;
6086
		rq->last_blocked_load_update_tick = jiffies;
6087
		atomic_set(&rq->nohz_flags, 0);
6088
#endif
6089
#endif /* CONFIG_SMP */
6090
		hrtick_rq_init(rq);
L
Linus Torvalds 已提交
6091 6092 6093
		atomic_set(&rq->nr_iowait, 0);
	}

6094
	set_load_weight(&init_task, false);
6095

L
Linus Torvalds 已提交
6096 6097 6098
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
V
Vegard Nossum 已提交
6099
	mmgrab(&init_mm);
L
Linus Torvalds 已提交
6100 6101 6102 6103 6104 6105 6106 6107 6108
	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());
6109 6110 6111

	calc_load_update = jiffies + LOAD_FREQ;

6112
#ifdef CONFIG_SMP
6113
	idle_thread_set_boot_cpu();
6114 6115
#endif
	init_sched_fair_class();
6116

6117 6118
	init_schedstats();

6119
	scheduler_running = 1;
L
Linus Torvalds 已提交
6120 6121
}

6122
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6123 6124
static inline int preempt_count_equals(int preempt_offset)
{
6125
	int nested = preempt_count() + rcu_preempt_depth();
6126

A
Arnd Bergmann 已提交
6127
	return (nested == preempt_offset);
6128 6129
}

6130
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6131
{
P
Peter Zijlstra 已提交
6132 6133 6134 6135 6136
	/*
	 * 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.
	 */
6137
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
6138 6139 6140 6141
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
6142
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
6143

6144 6145 6146 6147 6148
	___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
6149
{
I
Ingo Molnar 已提交
6150 6151 6152
	/* Ratelimiting timestamp: */
	static unsigned long prev_jiffy;

6153
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
6154

I
Ingo Molnar 已提交
6155 6156 6157
	/* WARN_ON_ONCE() by default, no rate limit required: */
	rcu_sleep_check();

6158 6159
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
6160 6161
	    system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
	    oops_in_progress)
I
Ingo Molnar 已提交
6162
		return;
6163

I
Ingo Molnar 已提交
6164 6165 6166 6167
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

I
Ingo Molnar 已提交
6168
	/* Save this before calling printk(), since that will clobber it: */
6169 6170
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
6171 6172 6173 6174 6175 6176 6177
	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
6178

6179 6180 6181
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
6182 6183 6184
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
6185 6186
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
6187
		pr_err("Preemption disabled at:");
6188
		print_ip_sym(preempt_disable_ip);
6189 6190
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
6191
	dump_stack();
6192
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
6193
}
6194
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
6195 6196 6197
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6198
void normalize_rt_tasks(void)
6199
{
6200
	struct task_struct *g, *p;
6201 6202 6203
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
6204

6205
	read_lock(&tasklist_lock);
6206
	for_each_process_thread(g, p) {
6207 6208 6209
		/*
		 * Only normalize user tasks:
		 */
6210
		if (p->flags & PF_KTHREAD)
6211 6212
			continue;

6213 6214 6215 6216
		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
6217

6218
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
6219 6220 6221 6222
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
6223
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
6224
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6225
			continue;
I
Ingo Molnar 已提交
6226
		}
L
Linus Torvalds 已提交
6227

6228
		__sched_setscheduler(p, &attr, false, false);
6229
	}
6230
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
6231 6232 6233
}

#endif /* CONFIG_MAGIC_SYSRQ */
6234

6235
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6236
/*
6237
 * These functions are only useful for the IA64 MCA handling, or kdb.
6238 6239 6240 6241 6242 6243 6244 6245 6246
 *
 * 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
6247
 * curr_task - return the current task for a given CPU.
6248 6249 6250
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6251 6252
 *
 * Return: The current task for @cpu.
6253
 */
6254
struct task_struct *curr_task(int cpu)
6255 6256 6257 6258
{
	return cpu_curr(cpu);
}

6259 6260 6261
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6262
/**
I
Ingo Molnar 已提交
6263
 * set_curr_task - set the current task for a given CPU.
6264 6265 6266 6267
 * @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
6268
 * are serviced on a separate stack. It allows the architecture to switch the
I
Ingo Molnar 已提交
6269
 * notion of the current task on a CPU in a non-blocking manner. This function
6270 6271 6272 6273 6274 6275 6276
 * 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!
 */
6277
void ia64_set_curr_task(int cpu, struct task_struct *p)
6278 6279 6280 6281 6282
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6283

D
Dhaval Giani 已提交
6284
#ifdef CONFIG_CGROUP_SCHED
6285 6286 6287
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6288
static void sched_free_group(struct task_group *tg)
6289 6290 6291
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6292
	autogroup_free(tg);
6293
	kmem_cache_free(task_group_cache, tg);
6294 6295 6296
}

/* allocate runqueue etc for a new task group */
6297
struct task_group *sched_create_group(struct task_group *parent)
6298 6299 6300
{
	struct task_group *tg;

6301
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6302 6303 6304
	if (!tg)
		return ERR_PTR(-ENOMEM);

6305
	if (!alloc_fair_sched_group(tg, parent))
6306 6307
		goto err;

6308
	if (!alloc_rt_sched_group(tg, parent))
6309 6310
		goto err;

6311 6312 6313
	return tg;

err:
6314
	sched_free_group(tg);
6315 6316 6317 6318 6319 6320 6321
	return ERR_PTR(-ENOMEM);
}

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

6322
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6323
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6324

I
Ingo Molnar 已提交
6325 6326
	/* Root should already exist: */
	WARN_ON(!parent);
P
Peter Zijlstra 已提交
6327 6328 6329

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6330
	list_add_rcu(&tg->siblings, &parent->children);
6331
	spin_unlock_irqrestore(&task_group_lock, flags);
6332 6333

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
6334 6335
}

6336
/* rcu callback to free various structures associated with a task group */
6337
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6338
{
I
Ingo Molnar 已提交
6339
	/* Now it should be safe to free those cfs_rqs: */
6340
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6341 6342
}

6343
void sched_destroy_group(struct task_group *tg)
6344
{
I
Ingo Molnar 已提交
6345
	/* Wait for possible concurrent references to cfs_rqs complete: */
6346
	call_rcu(&tg->rcu, sched_free_group_rcu);
6347 6348 6349
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6350
{
6351
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
6352

I
Ingo Molnar 已提交
6353
	/* End participation in shares distribution: */
6354
	unregister_fair_sched_group(tg);
6355 6356

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6357
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6358
	list_del_rcu(&tg->siblings);
6359
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6360 6361
}

6362
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
6363
{
P
Peter Zijlstra 已提交
6364
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6365

6366 6367 6368 6369 6370 6371
	/*
	 * 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
6372 6373 6374 6375
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6376
#ifdef CONFIG_FAIR_GROUP_SCHED
6377 6378
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
6379
	else
P
Peter Zijlstra 已提交
6380
#endif
6381
		set_task_rq(tsk, task_cpu(tsk));
6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392
}

/*
 * 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)
{
6393 6394
	int queued, running, queue_flags =
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6395 6396 6397 6398
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);
6399
	update_rq_clock(rq);
6400 6401 6402 6403 6404

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

	if (queued)
6405
		dequeue_task(rq, tsk, queue_flags);
6406
	if (running)
6407 6408 6409
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
6410

6411
	if (queued)
6412
		enqueue_task(rq, tsk, queue_flags);
6413
	if (running)
6414
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6415

6416
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
6417
}
6418

6419
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6420
{
6421
	return css ? container_of(css, struct task_group, css) : NULL;
6422 6423
}

6424 6425
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6426
{
6427 6428
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
6429

6430
	if (!parent) {
6431
		/* This is early initialization for the top cgroup */
6432
		return &root_task_group.css;
6433 6434
	}

6435
	tg = sched_create_group(parent);
6436 6437 6438 6439 6440 6441
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452
/* 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;
}

6453
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6454
{
6455
	struct task_group *tg = css_tg(css);
6456

6457
	sched_offline_group(tg);
6458 6459
}

6460
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6461
{
6462
	struct task_group *tg = css_tg(css);
6463

6464 6465 6466 6467
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
6468 6469
}

6470 6471 6472 6473
/*
 * 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.
 */
6474
static void cpu_cgroup_fork(struct task_struct *task)
6475
{
6476 6477 6478 6479 6480
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

6481
	update_rq_clock(rq);
6482 6483 6484
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
6485 6486
}

6487
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6488
{
6489
	struct task_struct *task;
6490
	struct cgroup_subsys_state *css;
6491
	int ret = 0;
6492

6493
	cgroup_taskset_for_each(task, css, tset) {
6494
#ifdef CONFIG_RT_GROUP_SCHED
6495
		if (!sched_rt_can_attach(css_tg(css), task))
6496
			return -EINVAL;
6497
#endif
6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513
		/*
		 * 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;
6514
	}
6515
	return ret;
6516
}
6517

6518
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6519
{
6520
	struct task_struct *task;
6521
	struct cgroup_subsys_state *css;
6522

6523
	cgroup_taskset_for_each(task, css, tset)
6524
		sched_move_task(task);
6525 6526
}

6527
#ifdef CONFIG_FAIR_GROUP_SCHED
6528 6529
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
6530
{
6531 6532
	if (shareval > scale_load_down(ULONG_MAX))
		shareval = MAX_SHARES;
6533
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
6534 6535
}

6536 6537
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
6538
{
6539
	struct task_group *tg = css_tg(css);
6540

6541
	return (u64) scale_load_down(tg->shares);
6542
}
6543 6544

#ifdef CONFIG_CFS_BANDWIDTH
6545 6546
static DEFINE_MUTEX(cfs_constraints_mutex);

6547 6548 6549
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

6550 6551
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

6552 6553
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
6554
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
6555
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575

	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;

6576 6577 6578 6579 6580
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
6581 6582 6583 6584 6585
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

6586
	runtime_enabled = quota != RUNTIME_INF;
6587
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6588 6589 6590 6591 6592 6593
	/*
	 * 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();
6594 6595 6596
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
6597

P
Paul Turner 已提交
6598
	__refill_cfs_bandwidth_runtime(cfs_b);
I
Ingo Molnar 已提交
6599 6600

	/* Restart the period timer (if active) to handle new period expiry: */
P
Peter Zijlstra 已提交
6601 6602
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
I
Ingo Molnar 已提交
6603

6604 6605
	raw_spin_unlock_irq(&cfs_b->lock);

6606
	for_each_online_cpu(i) {
6607
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6608
		struct rq *rq = cfs_rq->rq;
6609
		struct rq_flags rf;
6610

6611
		rq_lock_irq(rq, &rf);
6612
		cfs_rq->runtime_enabled = runtime_enabled;
6613
		cfs_rq->runtime_remaining = 0;
6614

6615
		if (cfs_rq->throttled)
6616
			unthrottle_cfs_rq(cfs_rq);
6617
		rq_unlock_irq(rq, &rf);
6618
	}
6619 6620
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
6621 6622
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
6623
	put_online_cpus();
6624

6625
	return ret;
6626 6627 6628 6629 6630 6631
}

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

6632
	period = ktime_to_ns(tg->cfs_bandwidth.period);
6633 6634
	if (cfs_quota_us < 0)
		quota = RUNTIME_INF;
6635
	else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC)
6636
		quota = (u64)cfs_quota_us * NSEC_PER_USEC;
6637 6638
	else
		return -EINVAL;
6639 6640 6641 6642 6643 6644 6645 6646

	return tg_set_cfs_bandwidth(tg, period, quota);
}

long tg_get_cfs_quota(struct task_group *tg)
{
	u64 quota_us;

6647
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6648 6649
		return -1;

6650
	quota_us = tg->cfs_bandwidth.quota;
6651 6652 6653 6654 6655 6656 6657 6658 6659
	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;

6660 6661 6662
	if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC)
		return -EINVAL;

6663
	period = (u64)cfs_period_us * NSEC_PER_USEC;
6664
	quota = tg->cfs_bandwidth.quota;
6665 6666 6667 6668 6669 6670 6671 6672

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

6673
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6674 6675 6676 6677 6678
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

6679 6680
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
6681
{
6682
	return tg_get_cfs_quota(css_tg(css));
6683 6684
}

6685 6686
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
6687
{
6688
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6689 6690
}

6691 6692
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6693
{
6694
	return tg_get_cfs_period(css_tg(css));
6695 6696
}

6697 6698
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
6699
{
6700
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
6701 6702
}

6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734
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;
6735
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6736 6737 6738 6739 6740
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
6741
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6742 6743

		quota = normalize_cfs_quota(tg, d);
6744
		parent_quota = parent_b->hierarchical_quota;
6745 6746

		/*
6747 6748
		 * Ensure max(child_quota) <= parent_quota.  On cgroup2,
		 * always take the min.  On cgroup1, only inherit when no
I
Ingo Molnar 已提交
6749
		 * limit is set:
6750
		 */
6751 6752 6753 6754 6755 6756 6757 6758
		if (cgroup_subsys_on_dfl(cpu_cgrp_subsys)) {
			quota = min(quota, parent_quota);
		} else {
			if (quota == RUNTIME_INF)
				quota = parent_quota;
			else if (parent_quota != RUNTIME_INF && quota > parent_quota)
				return -EINVAL;
		}
6759
	}
6760
	cfs_b->hierarchical_quota = quota;
6761 6762 6763 6764 6765 6766

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
6767
	int ret;
6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778
	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);
	}

6779 6780 6781 6782 6783
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6784
}
6785

6786
static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6787
{
6788
	struct task_group *tg = css_tg(seq_css(sf));
6789
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6790

6791 6792 6793
	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);
6794

6795 6796 6797 6798 6799 6800 6801 6802 6803 6804
	if (schedstat_enabled() && tg != &root_task_group) {
		u64 ws = 0;
		int i;

		for_each_possible_cpu(i)
			ws += schedstat_val(tg->se[i]->statistics.wait_sum);

		seq_printf(sf, "wait_sum %llu\n", ws);
	}

6805 6806
	return 0;
}
6807
#endif /* CONFIG_CFS_BANDWIDTH */
6808
#endif /* CONFIG_FAIR_GROUP_SCHED */
6809

6810
#ifdef CONFIG_RT_GROUP_SCHED
6811 6812
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
6813
{
6814
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
6815 6816
}

6817 6818
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
6819
{
6820
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
6821
}
6822

6823 6824
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
6825
{
6826
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
6827 6828
}

6829 6830
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6831
{
6832
	return sched_group_rt_period(css_tg(css));
6833
}
6834
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
6835

6836
static struct cftype cpu_legacy_files[] = {
6837
#ifdef CONFIG_FAIR_GROUP_SCHED
6838 6839
	{
		.name = "shares",
6840 6841
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
6842
	},
6843
#endif
6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854
#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,
	},
6855 6856
	{
		.name = "stat",
6857
		.seq_show = cpu_cfs_stat_show,
6858
	},
6859
#endif
6860
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6861
	{
P
Peter Zijlstra 已提交
6862
		.name = "rt_runtime_us",
6863 6864
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
6865
	},
6866 6867
	{
		.name = "rt_period_us",
6868 6869
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
6870
	},
6871
#endif
I
Ingo Molnar 已提交
6872
	{ }	/* Terminate */
6873 6874
};

6875 6876
static int cpu_extra_stat_show(struct seq_file *sf,
			       struct cgroup_subsys_state *css)
6877 6878 6879
{
#ifdef CONFIG_CFS_BANDWIDTH
	{
6880
		struct task_group *tg = css_tg(css);
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
		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;
6947
	int idx;
6948 6949 6950 6951

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

6952 6953 6954 6955
	idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO;
	idx = array_index_nospec(idx, 40);
	weight = sched_prio_to_weight[idx];

6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976
	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 */

6977
	if (sscanf(buf, "%20s %llu", tok, periodp) < 1)
6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041
		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 */
};

7042
struct cgroup_subsys cpu_cgrp_subsys = {
7043
	.css_alloc	= cpu_cgroup_css_alloc,
7044
	.css_online	= cpu_cgroup_css_online,
7045
	.css_released	= cpu_cgroup_css_released,
7046
	.css_free	= cpu_cgroup_css_free,
7047
	.css_extra_stat_show = cpu_extra_stat_show,
7048
	.fork		= cpu_cgroup_fork,
7049 7050
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
7051
	.legacy_cftypes	= cpu_legacy_files,
7052
	.dfl_cftypes	= cpu_files,
7053
	.early_init	= true,
7054
	.threaded	= true,
7055 7056
};

7057
#endif	/* CONFIG_CGROUP_SCHED */
7058

7059 7060 7061 7062 7063
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100 7101 7102 7103 7104

/*
 * 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,
};
7105 7106

#undef CREATE_TRACE_POINTS