core.c 172.9 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(HAVE_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.
		 *
I
Ingo Molnar 已提交
113
		 * If we observe the new CPU in task_rq_lock, the acquire will
114 115 116
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
117
			rq_pin_lock(rq, rf);
118 119 120
			return rq;
		}
		raw_spin_unlock(&rq->lock);
121
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
122 123 124 125 126 127

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

128 129 130 131 132 133 134 135 136 137
/*
 * 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...
 */
138 139
	s64 __maybe_unused steal = 0, irq_delta = 0;

140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
#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;

179
#ifdef CONFIG_HAVE_SCHED_AVG_IRQ
180
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
181
		update_irq_load_avg(rq, irq_delta + steal);
182 183 184 185 186 187 188 189 190 191 192 193 194
#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
195 196
	if (sched_feat(WARN_DOUBLE_CLOCK))
		SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
197 198
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
199

200 201 202 203 204 205 206 207
	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
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
#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);
226
	struct rq_flags rf;
P
Peter Zijlstra 已提交
227 228 229

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

230
	rq_lock(rq, &rf);
231
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
232
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
233
	rq_unlock(rq, &rf);
P
Peter Zijlstra 已提交
234 235 236 237

	return HRTIMER_NORESTART;
}

238
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
239

240
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
241 242 243
{
	struct hrtimer *timer = &rq->hrtick_timer;

244
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
245 246
}

247 248 249 250
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
251
{
252
	struct rq *rq = arg;
253
	struct rq_flags rf;
254

255
	rq_lock(rq, &rf);
P
Peter Zijlstra 已提交
256
	__hrtick_restart(rq);
257
	rq->hrtick_csd_pending = 0;
258
	rq_unlock(rq, &rf);
259 260
}

261 262 263 264 265
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
266
void hrtick_start(struct rq *rq, u64 delay)
267
{
268
	struct hrtimer *timer = &rq->hrtick_timer;
269 270 271 272 273 274 275 276 277
	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);
278

279
	hrtimer_set_expires(timer, time);
280 281

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
282
		__hrtick_restart(rq);
283
	} else if (!rq->hrtick_csd_pending) {
284
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
285 286
		rq->hrtick_csd_pending = 1;
	}
287 288
}

289 290 291 292 293 294
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
295
void hrtick_start(struct rq *rq, u64 delay)
296
{
W
Wanpeng Li 已提交
297 298 299 300 301
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
302 303
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
304 305
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
306

307
static void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
308
{
309 310
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
311

312 313 314 315
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
316

317 318
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
319
}
A
Andrew Morton 已提交
320
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
321 322 323 324
static inline void hrtick_clear(struct rq *rq)
{
}

325
static inline void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
326 327
{
}
A
Andrew Morton 已提交
328
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
329

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
/*
 * 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;								\
})

348
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
349 350 351 352 353 354 355 356 357 358
/*
 * 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);
}
359 360 361 362 363 364 365 366 367 368

/*
 * 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);
369
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
370 371 372 373 374 375 376 377 378 379 380 381 382 383

	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;
}

384 385 386 387 388 389
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
390 391 392 393 394 395 396

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
397 398
#endif

399 400 401 402 403 404 405 406 407
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.
	 *
408 409
	 * 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.
410
	 */
411 412
	smp_mb__before_atomic();
	if (cmpxchg_relaxed(&node->next, NULL, WAKE_Q_TAIL))
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
		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
433
		/* Task can safely be re-inserted now: */
434 435 436 437
		node = node->next;
		task->wake_q.next = NULL;

		/*
438 439
		 * wake_up_process() executes a full barrier, which pairs with
		 * the queueing in wake_q_add() so as not to miss wakeups.
440 441 442 443 444 445
		 */
		wake_up_process(task);
		put_task_struct(task);
	}
}

I
Ingo Molnar 已提交
446
/*
447
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
448 449 450 451 452
 *
 * 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.
 */
453
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
454
{
455
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
456 457
	int cpu;

458
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
459

460
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
461 462
		return;

463
	cpu = cpu_of(rq);
464

465
	if (cpu == smp_processor_id()) {
466
		set_tsk_need_resched(curr);
467
		set_preempt_need_resched();
I
Ingo Molnar 已提交
468
		return;
469
	}
I
Ingo Molnar 已提交
470

471
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
472
		smp_send_reschedule(cpu);
473 474
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
475 476
}

477
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
478 479 480 481
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

482
	raw_spin_lock_irqsave(&rq->lock, flags);
483 484
	if (cpu_online(cpu) || cpu == smp_processor_id())
		resched_curr(rq);
485
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
486
}
487

488
#ifdef CONFIG_SMP
489
#ifdef CONFIG_NO_HZ_COMMON
490
/*
I
Ingo Molnar 已提交
491 492
 * In the semi idle case, use the nearest busy CPU for migrating timers
 * from an idle CPU.  This is good for power-savings.
493 494
 *
 * We don't do similar optimization for completely idle system, as
I
Ingo Molnar 已提交
495 496
 * 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).
497
 */
498
int get_nohz_timer_target(void)
499
{
500
	int i, cpu = smp_processor_id();
501 502
	struct sched_domain *sd;

503
	if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
504 505
		return cpu;

506
	rcu_read_lock();
507
	for_each_domain(cpu, sd) {
508
		for_each_cpu(i, sched_domain_span(sd)) {
509 510 511
			if (cpu == i)
				continue;

512
			if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
513 514 515 516
				cpu = i;
				goto unlock;
			}
		}
517
	}
518

519 520
	if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
		cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
521 522
unlock:
	rcu_read_unlock();
523 524
	return cpu;
}
I
Ingo Molnar 已提交
525

526 527 528 529 530 531 532 533 534 535
/*
 * 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.
 */
536
static void wake_up_idle_cpu(int cpu)
537 538 539 540 541 542
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

543
	if (set_nr_and_not_polling(rq->idle))
544
		smp_send_reschedule(cpu);
545 546
	else
		trace_sched_wake_idle_without_ipi(cpu);
547 548
}

549
static bool wake_up_full_nohz_cpu(int cpu)
550
{
551 552 553 554 555 556
	/*
	 * 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.
	 */
557 558
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
559
	if (tick_nohz_full_cpu(cpu)) {
560 561
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
562
			tick_nohz_full_kick_cpu(cpu);
563 564 565 566 567 568
		return true;
	}

	return false;
}

569 570 571 572 573
/*
 * 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.
 */
574 575
void wake_up_nohz_cpu(int cpu)
{
576
	if (!wake_up_full_nohz_cpu(cpu))
577 578 579
		wake_up_idle_cpu(cpu);
}

580
static inline bool got_nohz_idle_kick(void)
581
{
582
	int cpu = smp_processor_id();
583

P
Peter Zijlstra 已提交
584
	if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK))
585 586 587 588 589 590 591 592 593
		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
594
	atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu));
595
	return false;
596 597
}

598
#else /* CONFIG_NO_HZ_COMMON */
599

600
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
601
{
602
	return false;
P
Peter Zijlstra 已提交
603 604
}

605
#endif /* CONFIG_NO_HZ_COMMON */
606

607
#ifdef CONFIG_NO_HZ_FULL
608
bool sched_can_stop_tick(struct rq *rq)
609
{
610 611 612 613 614 615
	int fifo_nr_running;

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

616
	/*
617 618
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
619
	 */
620 621 622 623 624
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
625 626
	}

627 628 629 630 631 632 633 634 635 636 637 638 639 640
	/*
	 * 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)
641
		return false;
642

643
	return true;
644 645
}
#endif /* CONFIG_NO_HZ_FULL */
646
#endif /* CONFIG_SMP */
647

648 649
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
650
/*
651 652 653 654
 * 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.
655
 */
656
int walk_tg_tree_from(struct task_group *from,
657
			     tg_visitor down, tg_visitor up, void *data)
658 659
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
660
	int ret;
661

662 663
	parent = from;

664
down:
P
Peter Zijlstra 已提交
665 666
	ret = (*down)(parent, data);
	if (ret)
667
		goto out;
668 669 670 671 672 673 674
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
675
	ret = (*up)(parent, data);
676 677
	if (ret || parent == from)
		goto out;
678 679 680 681 682

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
683
out:
P
Peter Zijlstra 已提交
684
	return ret;
685 686
}

687
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
688
{
689
	return 0;
P
Peter Zijlstra 已提交
690
}
691 692
#endif

693
static void set_load_weight(struct task_struct *p, bool update_load)
694
{
N
Nikhil Rao 已提交
695 696 697
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
698 699 700
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
701
	if (idle_policy(p->policy)) {
702
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
703
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
704 705
		return;
	}
706

707 708 709 710 711 712 713 714 715 716
	/*
	 * 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];
	}
717 718
}

719
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
720
{
721 722 723
	if (!(flags & ENQUEUE_NOCLOCK))
		update_rq_clock(rq);

724 725
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
726

727
	p->sched_class->enqueue_task(rq, p, flags);
728 729
}

730
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
731
{
732 733 734
	if (!(flags & DEQUEUE_NOCLOCK))
		update_rq_clock(rq);

735 736
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
737

738
	p->sched_class->dequeue_task(rq, p, flags);
739 740
}

741
void activate_task(struct rq *rq, struct task_struct *p, int flags)
742 743 744 745
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

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

749
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
750 751 752 753
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

754
	dequeue_task(rq, p, flags);
755 756
}

757
/*
I
Ingo Molnar 已提交
758
 * __normal_prio - return the priority that is based on the static prio
759 760 761
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
762
	return p->static_prio;
763 764
}

765 766 767 768 769 770 771
/*
 * 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.
 */
772
static inline int normal_prio(struct task_struct *p)
773 774 775
{
	int prio;

776 777 778
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
779 780 781 782 783 784 785 786 787 788 789 790 791
		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.
 */
792
static int effective_prio(struct task_struct *p)
793 794 795 796 797 798 799 800 801 802 803 804
{
	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
805 806 807
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
808 809
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
810
 */
811
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
812 813 814 815
{
	return cpu_curr(task_cpu(p)) == p;
}

816
/*
817 818 819 820 821
 * 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().
822
 */
823 824
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
825
				       int oldprio)
826 827 828
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
829
			prev_class->switched_from(rq, p);
830

P
Peter Zijlstra 已提交
831
		p->sched_class->switched_to(rq, p);
832
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
833
		p->sched_class->prio_changed(rq, p, oldprio);
834 835
}

836
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
837 838 839 840 841 842 843 844 845 846
{
	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) {
847
				resched_curr(rq);
848 849 850 851 852 853 854 855 856
				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.
	 */
857
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
858
		rq_clock_skip_update(rq);
859 860
}

L
Linus Torvalds 已提交
861
#ifdef CONFIG_SMP
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888

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
889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907
/*
 * 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.
 */
908 909
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
910 911 912 913
{
	lockdep_assert_held(&rq->lock);

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

	rq = cpu_rq(new_cpu);

920
	rq_lock(rq, rf);
P
Peter Zijlstra 已提交
921 922
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
923
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
924 925 926 927 928 929 930 931 932 933 934
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
I
Ingo Molnar 已提交
935
 * Move (not current) task off this CPU, onto the destination CPU. We're doing
P
Peter Zijlstra 已提交
936 937 938 939 940 941 942
 * 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.
 */
943 944
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
945 946
{
	/* Affinity changed (again). */
947
	if (!is_cpu_allowed(p, dest_cpu))
948
		return rq;
P
Peter Zijlstra 已提交
949

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

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

/*
 * migration_cpu_stop - this will be executed by a highprio stopper thread
 * and performs thread migration by bumping thread off CPU then
 * 'pushing' onto another runqueue.
 */
static int migration_cpu_stop(void *data)
{
	struct migration_arg *arg = data;
964 965
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
966
	struct rq_flags rf;
P
Peter Zijlstra 已提交
967 968

	/*
I
Ingo Molnar 已提交
969 970
	 * The original target CPU might have gone down and we might
	 * be on another CPU but it doesn't matter.
P
Peter Zijlstra 已提交
971 972 973 974 975 976 977 978
	 */
	local_irq_disable();
	/*
	 * We need to explicitly wake pending tasks before running
	 * __migrate_task() such that we will not miss enforcing cpus_allowed
	 * during wakeups, see set_cpus_allowed_ptr()'s TASK_WAKING test.
	 */
	sched_ttwu_pending();
979 980

	raw_spin_lock(&p->pi_lock);
981
	rq_lock(rq, &rf);
982 983 984 985 986
	/*
	 * If task_rq(p) != rq, it cannot be migrated here, because we're
	 * holding rq->lock, if p->on_rq == 0 it cannot get enqueued because
	 * we're holding p->pi_lock.
	 */
987 988
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
989
			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
990 991 992
		else
			p->wake_cpu = arg->dest_cpu;
	}
993
	rq_unlock(rq, &rf);
994 995
	raw_spin_unlock(&p->pi_lock);

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

1000 1001 1002 1003 1004
/*
 * sched_class::set_cpus_allowed must do the below, but is not required to
 * actually call this function.
 */
void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask)
P
Peter Zijlstra 已提交
1005 1006 1007 1008 1009
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

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

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

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

	if (queued) {
		/*
		 * Because __kthread_bind() calls this on blocked tasks without
		 * holding rq->lock.
		 */
		lockdep_assert_held(&rq->lock);
1026
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1027 1028 1029 1030
	}
	if (running)
		put_prev_task(rq, p);

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

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

P
Peter Zijlstra 已提交
1039 1040 1041 1042 1043 1044 1045 1046 1047
/*
 * Change a given task's CPU affinity. Migrate the thread to a
 * proper CPU and schedule it away if the CPU it's executing on
 * is removed from the allowed bitmask.
 *
 * NOTE: the caller must have a valid reference to the task, the
 * task must not exit() & deallocate itself prematurely. The
 * call is not atomic; no spinlocks may be held.
 */
1048 1049
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1050
{
1051
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1052
	unsigned int dest_cpu;
1053 1054
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1055 1056
	int ret = 0;

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

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

1067 1068 1069 1070 1071 1072 1073 1074 1075
	/*
	 * Must re-check here, to close a race against __kthread_bind(),
	 * sched_setaffinity() is not guaranteed to observe the flag.
	 */
	if (check && (p->flags & PF_NO_SETAFFINITY)) {
		ret = -EINVAL;
		goto out;
	}

P
Peter Zijlstra 已提交
1076 1077 1078
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

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

	do_set_cpus_allowed(p, new_mask);

1086 1087 1088
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
I
Ingo Molnar 已提交
1089
		 * !active we want to ensure they are strict per-CPU threads.
1090 1091 1092 1093 1094 1095
		 */
		WARN_ON(cpumask_intersects(new_mask, cpu_online_mask) &&
			!cpumask_intersects(new_mask, cpu_active_mask) &&
			p->nr_cpus_allowed != 1);
	}

P
Peter Zijlstra 已提交
1096 1097 1098 1099
	/* Can the task run on the task's current CPU? If so, we're done */
	if (cpumask_test_cpu(task_cpu(p), new_mask))
		goto out;

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

	return ret;
}
1120 1121 1122 1123 1124

int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask)
{
	return __set_cpus_allowed_ptr(p, new_mask, false);
}
P
Peter Zijlstra 已提交
1125 1126
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1127
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1128
{
1129 1130 1131 1132 1133
#ifdef CONFIG_SCHED_DEBUG
	/*
	 * We should never call set_task_cpu() on a blocked task,
	 * ttwu() will sort out the placement.
	 */
P
Peter Zijlstra 已提交
1134
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1135
			!p->on_rq);
1136

1137 1138 1139 1140 1141 1142 1143 1144 1145
	/*
	 * Migrating fair class task must have p->on_rq = TASK_ON_RQ_MIGRATING,
	 * because schedstat_wait_{start,end} rebase migrating task's wait_start
	 * time relying on p->on_rq.
	 */
	WARN_ON_ONCE(p->state == TASK_RUNNING &&
		     p->sched_class == &fair_sched_class &&
		     (p->on_rq && !task_on_rq_migrating(p)));

1146
#ifdef CONFIG_LOCKDEP
1147 1148 1149 1150 1151
	/*
	 * The caller should hold either p->pi_lock or rq->lock, when changing
	 * a task's CPU. ->pi_lock for waking tasks, rq->lock for runnable tasks.
	 *
	 * sched_move_task() holds both and thus holding either pins the cgroup,
P
Peter Zijlstra 已提交
1152
	 * see task_group().
1153 1154 1155 1156
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1157 1158 1159
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1160 1161 1162 1163
	/*
	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
	 */
	WARN_ON_ONCE(!cpu_online(new_cpu));
1164 1165
#endif

1166
	trace_sched_migrate_task(p, new_cpu);
1167

1168
	if (task_cpu(p) != new_cpu) {
1169
		if (p->sched_class->migrate_task_rq)
1170
			p->sched_class->migrate_task_rq(p, new_cpu);
1171
		p->se.nr_migrations++;
1172
		rseq_migrate(p);
1173
		perf_event_task_migrate(p);
1174
	}
I
Ingo Molnar 已提交
1175 1176

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1177 1178
}

1179
#ifdef CONFIG_NUMA_BALANCING
1180 1181
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1182
	if (task_on_rq_queued(p)) {
1183
		struct rq *src_rq, *dst_rq;
1184
		struct rq_flags srf, drf;
1185 1186 1187 1188

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

1189 1190 1191
		rq_pin_lock(src_rq, &srf);
		rq_pin_lock(dst_rq, &drf);

1192
		p->on_rq = TASK_ON_RQ_MIGRATING;
1193 1194 1195
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1196
		p->on_rq = TASK_ON_RQ_QUEUED;
1197
		check_preempt_curr(dst_rq, p, 0);
1198 1199 1200 1201

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

1202 1203 1204 1205
	} 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
1206
		 * previous CPU our target instead of where it really is.
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
		 */
		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;

1223 1224 1225
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1226 1227 1228
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1229 1230
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1231
	double_rq_lock(src_rq, dst_rq);
1232

1233 1234 1235 1236 1237 1238
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

1239
	if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1240 1241
		goto unlock;

1242
	if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1243 1244 1245 1246 1247 1248 1249 1250 1251
		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);
1252 1253
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1254 1255 1256 1257 1258 1259 1260

	return ret;
}

/*
 * Cross migrate two tasks
 */
1261 1262
int migrate_swap(struct task_struct *cur, struct task_struct *p,
		int target_cpu, int curr_cpu)
1263 1264 1265 1266 1267 1268
{
	struct migration_swap_arg arg;
	int ret = -EINVAL;

	arg = (struct migration_swap_arg){
		.src_task = cur,
1269
		.src_cpu = curr_cpu,
1270
		.dst_task = p,
1271
		.dst_cpu = target_cpu,
1272 1273 1274 1275 1276
	};

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

1277 1278 1279 1280
	/*
	 * 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.
	 */
1281 1282 1283
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

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

1287
	if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1288 1289
		goto out;

1290
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1291 1292 1293 1294 1295
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}
1296
#endif /* CONFIG_NUMA_BALANCING */
1297

L
Linus Torvalds 已提交
1298 1299 1300
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1301 1302 1303 1304 1305 1306 1307
 * 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
1308 1309 1310 1311 1312 1313
 * 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
1314
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1315
{
1316
	int running, queued;
1317
	struct rq_flags rf;
R
Roland McGrath 已提交
1318
	unsigned long ncsw;
1319
	struct rq *rq;
L
Linus Torvalds 已提交
1320

1321 1322 1323 1324 1325 1326 1327 1328
	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);
1329

1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
		/*
		 * 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
1341 1342 1343
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1344
			cpu_relax();
R
Roland McGrath 已提交
1345
		}
1346

1347 1348 1349 1350 1351
		/*
		 * 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.
		 */
1352
		rq = task_rq_lock(p, &rf);
1353
		trace_sched_wait_task(p);
1354
		running = task_running(rq, p);
1355
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1356
		ncsw = 0;
1357
		if (!match_state || p->state == match_state)
1358
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1359
		task_rq_unlock(rq, p, &rf);
1360

R
Roland McGrath 已提交
1361 1362 1363 1364 1365 1366
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1367 1368 1369 1370 1371 1372 1373 1374 1375 1376
		/*
		 * 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;
		}
1377

1378 1379 1380 1381 1382
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1383
		 * So if it was still runnable (but just not actively
1384 1385 1386
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1387
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1388
			ktime_t to = NSEC_PER_SEC / HZ;
1389 1390 1391

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1392 1393
			continue;
		}
1394

1395 1396 1397 1398 1399 1400 1401
		/*
		 * 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
1402 1403

	return ncsw;
L
Linus Torvalds 已提交
1404 1405 1406 1407 1408 1409 1410 1411 1412
}

/***
 * 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
1413
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1414 1415 1416 1417 1418
 * 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.
 */
1419
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1420 1421 1422 1423 1424 1425 1426 1427 1428
{
	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
1429
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1430

1431
/*
1432
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1433 1434 1435 1436 1437
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
1438
 *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
1439
 *    see __set_cpus_allowed_ptr(). At this point the newly online
I
Ingo Molnar 已提交
1440
 *    CPU isn't yet part of the sched domains, and balancing will not
1441 1442
 *    see it.
 *
I
Ingo Molnar 已提交
1443
 *  - on CPU-down we clear cpu_active() to mask the sched domains and
1444
 *    avoid the load balancer to place new tasks on the to be removed
I
Ingo Molnar 已提交
1445
 *    CPU. Existing tasks will remain running there and will be taken
1446 1447 1448 1449 1450 1451
 *    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.
1452
 */
1453 1454
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1455 1456
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1457 1458
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1459

1460
	/*
I
Ingo Molnar 已提交
1461 1462 1463
	 * 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.
1464 1465 1466 1467 1468 1469 1470 1471
	 */
	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;
1472
			if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1473 1474
				return dest_cpu;
		}
1475
	}
1476

1477 1478
	for (;;) {
		/* Any allowed, online CPU? */
1479
		for_each_cpu(dest_cpu, &p->cpus_allowed) {
1480
			if (!is_cpu_allowed(p, dest_cpu))
1481
				continue;
1482

1483 1484
			goto out;
		}
1485

1486
		/* No more Mr. Nice Guy. */
1487 1488
		switch (state) {
		case cpuset:
1489 1490 1491 1492 1493
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
I
Ingo Molnar 已提交
1494
			/* Fall-through */
1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513
		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()) {
1514
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1515 1516
					task_pid_nr(p), p->comm, cpu);
		}
1517 1518 1519 1520 1521
	}

	return dest_cpu;
}

1522
/*
1523
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1524
 */
1525
static inline
1526
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1527
{
1528 1529
	lockdep_assert_held(&p->pi_lock);

1530
	if (p->nr_cpus_allowed > 1)
1531
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1532
	else
1533
		cpu = cpumask_any(&p->cpus_allowed);
1534 1535 1536 1537

	/*
	 * 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
1538
	 * CPU.
1539 1540 1541 1542 1543 1544
	 *
	 * 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 ]
	 */
1545
	if (unlikely(!is_cpu_allowed(p, cpu)))
1546
		cpu = select_fallback_rq(task_cpu(p), p);
1547 1548

	return cpu;
1549
}
1550 1551 1552 1553 1554 1555

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1556

1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586
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;
	}
}

1587 1588 1589 1590 1591 1592 1593 1594
#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
1595
#endif /* CONFIG_SMP */
1596

P
Peter Zijlstra 已提交
1597
static void
1598
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1599
{
1600
	struct rq *rq;
1601

1602 1603 1604 1605
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1606

1607 1608
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1609 1610
		__schedstat_inc(rq->ttwu_local);
		__schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1611 1612 1613
	} else {
		struct sched_domain *sd;

1614
		__schedstat_inc(p->se.statistics.nr_wakeups_remote);
1615
		rcu_read_lock();
1616
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1617
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1618
				__schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1619 1620 1621
				break;
			}
		}
1622
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1623
	}
1624 1625

	if (wake_flags & WF_MIGRATED)
1626
		__schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1627 1628
#endif /* CONFIG_SMP */

1629 1630
	__schedstat_inc(rq->ttwu_count);
	__schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1631 1632

	if (wake_flags & WF_SYNC)
1633
		__schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1634 1635
}

1636
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1637
{
T
Tejun Heo 已提交
1638
	activate_task(rq, p, en_flags);
1639
	p->on_rq = TASK_ON_RQ_QUEUED;
1640

I
Ingo Molnar 已提交
1641
	/* If a worker is waking up, notify the workqueue: */
1642 1643
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1644 1645
}

1646 1647 1648
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1649
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1650
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1651 1652 1653
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1654 1655
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1656
#ifdef CONFIG_SMP
1657 1658
	if (p->sched_class->task_woken) {
		/*
1659 1660
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1661
		 */
1662
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1663
		p->sched_class->task_woken(rq, p);
1664
		rq_repin_lock(rq, rf);
1665
	}
T
Tejun Heo 已提交
1666

1667
	if (rq->idle_stamp) {
1668
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1669
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1670

1671 1672 1673
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1674
			rq->avg_idle = max;
1675

T
Tejun Heo 已提交
1676 1677 1678 1679 1680
		rq->idle_stamp = 0;
	}
#endif
}

1681
static void
1682
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1683
		 struct rq_flags *rf)
1684
{
1685
	int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1686

1687 1688
	lockdep_assert_held(&rq->lock);

1689 1690 1691
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1692 1693

	if (wake_flags & WF_MIGRATED)
1694
		en_flags |= ENQUEUE_MIGRATED;
1695 1696
#endif

1697
	ttwu_activate(rq, p, en_flags);
1698
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1699 1700 1701 1702 1703 1704 1705 1706 1707 1708
}

/*
 * 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)
{
1709
	struct rq_flags rf;
1710 1711 1712
	struct rq *rq;
	int ret = 0;

1713
	rq = __task_rq_lock(p, &rf);
1714
	if (task_on_rq_queued(p)) {
1715 1716
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1717
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1718 1719
		ret = 1;
	}
1720
	__task_rq_unlock(rq, &rf);
1721 1722 1723 1724

	return ret;
}

1725
#ifdef CONFIG_SMP
1726
void sched_ttwu_pending(void)
1727 1728
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1729
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1730
	struct task_struct *p, *t;
1731
	struct rq_flags rf;
1732

1733 1734 1735
	if (!llist)
		return;

1736
	rq_lock_irqsave(rq, &rf);
1737
	update_rq_clock(rq);
1738

1739 1740
	llist_for_each_entry_safe(p, t, llist, wake_entry)
		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1741

1742
	rq_unlock_irqrestore(rq, &rf);
1743 1744 1745 1746
}

void scheduler_ipi(void)
{
1747 1748 1749 1750 1751
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1752
	preempt_fold_need_resched();
1753

1754
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770
		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
1771
	sched_ttwu_pending();
1772 1773 1774 1775

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1776
	if (unlikely(got_nohz_idle_kick())) {
1777
		this_rq()->idle_balance = 1;
1778
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1779
	}
1780
	irq_exit();
1781 1782
}

P
Peter Zijlstra 已提交
1783
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1784
{
1785 1786
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1787 1788
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1789 1790 1791 1792 1793 1794
	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);
	}
1795
}
1796

1797 1798 1799
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
1800
	struct rq_flags rf;
1801

1802 1803 1804 1805
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1806 1807 1808 1809

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
1810
		rq_lock_irqsave(rq, &rf);
1811 1812
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
I
Ingo Molnar 已提交
1813
		/* Else CPU is not idle, do nothing here: */
1814
		rq_unlock_irqrestore(rq, &rf);
1815
	}
1816 1817 1818

out:
	rcu_read_unlock();
1819 1820
}

1821
bool cpus_share_cache(int this_cpu, int that_cpu)
1822 1823 1824
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1825
#endif /* CONFIG_SMP */
1826

1827
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1828 1829
{
	struct rq *rq = cpu_rq(cpu);
1830
	struct rq_flags rf;
1831

1832
#if defined(CONFIG_SMP)
1833
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
I
Ingo Molnar 已提交
1834
		sched_clock_cpu(cpu); /* Sync clocks across CPUs */
P
Peter Zijlstra 已提交
1835
		ttwu_queue_remote(p, cpu, wake_flags);
1836 1837 1838 1839
		return;
	}
#endif

1840
	rq_lock(rq, &rf);
1841
	update_rq_clock(rq);
1842
	ttwu_do_activate(rq, p, wake_flags, &rf);
1843
	rq_unlock(rq, &rf);
T
Tejun Heo 已提交
1844 1845
}

1846 1847 1848 1849 1850 1851
/*
 * 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
1852 1853
 * migrates, all its activity on its old CPU [c0] happens-before any subsequent
 * execution on its new CPU [c1].
1854 1855 1856 1857 1858 1859 1860 1861
 *
 * For migration (of runnable tasks) this is provided by the following means:
 *
 *  A) UNLOCK of the rq(c0)->lock scheduling out task t
 *  B) migration for t is required to synchronize *both* rq(c0)->lock and
 *     rq(c1)->lock (if not at the same time, then in that order).
 *  C) LOCK of the rq(c1)->lock scheduling in task
 *
1862
 * Release/acquire chaining guarantees that B happens after A and C after B.
I
Ingo Molnar 已提交
1863
 * Note: the CPU doing B need not be c0 or c1
1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894
 *
 * 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)
1895
 *   2) smp_cond_load_acquire(!X->on_cpu)
1896 1897 1898 1899 1900 1901 1902 1903 1904 1905
 *
 * 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);
 *
1906
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923
 *                    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
 *
 *
1924 1925 1926
 * 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().
1927 1928
 */

T
Tejun Heo 已提交
1929
/**
L
Linus Torvalds 已提交
1930
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1931
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1932
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1933
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1934
 *
1935
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1936
 *
1937 1938 1939 1940 1941
 * 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().
 *
1942 1943 1944
 * This function executes a full memory barrier before accessing the task
 * state; see set_current_state().
 *
1945 1946
 * Return: %true if @p->state changes (an actual wakeup was done),
 *	   %false otherwise.
L
Linus Torvalds 已提交
1947
 */
1948 1949
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1950 1951
{
	unsigned long flags;
1952
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1953

1954 1955 1956 1957 1958 1959
	/*
	 * 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.
	 */
1960
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1961
	smp_mb__after_spinlock();
P
Peter Zijlstra 已提交
1962
	if (!(p->state & state))
L
Linus Torvalds 已提交
1963 1964
		goto out;

1965 1966
	trace_sched_waking(p);

I
Ingo Molnar 已提交
1967 1968
	/* We're going to change ->state: */
	success = 1;
L
Linus Torvalds 已提交
1969 1970
	cpu = task_cpu(p);

1971 1972 1973 1974 1975
	/*
	 * 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.
	 *
1976 1977 1978 1979 1980 1981 1982 1983
	 * 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
1984 1985
	 *
	 * [task p]
1986
	 *   STORE p->state = UNINTERRUPTIBLE	  LOAD p->on_rq
1987
	 *
1988 1989
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
1990 1991
	 */
	smp_rmb();
1992 1993
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
1994 1995

#ifdef CONFIG_SMP
1996 1997 1998 1999 2000 2001 2002
	/*
	 * 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.
	 *
2003 2004 2005 2006 2007 2008 2009 2010
	 * __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
2011
	 *
2012 2013
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
2014 2015 2016
	 */
	smp_rmb();

P
Peter Zijlstra 已提交
2017
	/*
I
Ingo Molnar 已提交
2018
	 * If the owning (remote) CPU is still in the middle of schedule() with
2019
	 * this task as prev, wait until its done referencing the task.
2020
	 *
2021
	 * Pairs with the smp_store_release() in finish_task().
2022 2023 2024
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2025
	 */
2026
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2027

2028
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2029
	p->state = TASK_WAKING;
2030

2031
	if (p->in_iowait) {
2032
		delayacct_blkio_end(p);
2033 2034 2035
		atomic_dec(&task_rq(p)->nr_iowait);
	}

2036
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2037 2038
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2039
		set_task_cpu(p, cpu);
2040
	}
2041 2042 2043 2044

#else /* CONFIG_SMP */

	if (p->in_iowait) {
2045
		delayacct_blkio_end(p);
2046 2047 2048
		atomic_dec(&task_rq(p)->nr_iowait);
	}

L
Linus Torvalds 已提交
2049 2050
#endif /* CONFIG_SMP */

2051
	ttwu_queue(p, cpu, wake_flags);
2052
stat:
2053
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2054
out:
2055
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2056 2057 2058 2059

	return success;
}

T
Tejun Heo 已提交
2060 2061 2062
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2063
 * @rf: request-queue flags for pinning
T
Tejun Heo 已提交
2064
 *
2065
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2066
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2067
 * the current task.
T
Tejun Heo 已提交
2068
 */
2069
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2070 2071 2072
{
	struct rq *rq = task_rq(p);

2073 2074 2075 2076
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2077 2078
	lockdep_assert_held(&rq->lock);

2079
	if (!raw_spin_trylock(&p->pi_lock)) {
2080 2081 2082 2083 2084 2085
		/*
		 * 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.
		 */
2086
		rq_unlock(rq, rf);
2087
		raw_spin_lock(&p->pi_lock);
2088
		rq_relock(rq, rf);
2089 2090
	}

T
Tejun Heo 已提交
2091
	if (!(p->state & TASK_NORMAL))
2092
		goto out;
T
Tejun Heo 已提交
2093

2094 2095
	trace_sched_waking(p);

2096 2097
	if (!task_on_rq_queued(p)) {
		if (p->in_iowait) {
2098
			delayacct_blkio_end(p);
2099 2100
			atomic_dec(&rq->nr_iowait);
		}
2101
		ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2102
	}
P
Peter Zijlstra 已提交
2103

2104
	ttwu_do_wakeup(rq, p, 0, rf);
2105
	ttwu_stat(p, smp_processor_id(), 0);
2106 2107
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2108 2109
}

2110 2111 2112 2113 2114
/**
 * 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
2115 2116 2117
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2118
 *
2119
 * This function executes a full memory barrier before accessing the task state.
2120
 */
2121
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2122
{
2123
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2124 2125 2126
}
EXPORT_SYMBOL(wake_up_process);

2127
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2128 2129 2130 2131 2132 2133 2134
{
	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
2135 2136 2137
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2138
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2139
{
P
Peter Zijlstra 已提交
2140 2141 2142
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2143 2144
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2145
	p->se.prev_sum_exec_runtime	= 0;
2146
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2147
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2148
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2149

2150 2151 2152 2153
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2154
#ifdef CONFIG_SCHEDSTATS
2155
	/* Even if schedstat is disabled, there should not be garbage */
2156
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2157
#endif
N
Nick Piggin 已提交
2158

2159
	RB_CLEAR_NODE(&p->dl.rb_node);
2160
	init_dl_task_timer(&p->dl);
2161
	init_dl_inactive_task_timer(&p->dl);
2162
	__dl_clear_params(p);
2163

P
Peter Zijlstra 已提交
2164
	INIT_LIST_HEAD(&p->rt.run_list);
2165 2166 2167 2168
	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
2169

2170 2171 2172
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2173

2174
	init_numa_balancing(clone_flags, p);
I
Ingo Molnar 已提交
2175 2176
}

2177 2178
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2179
#ifdef CONFIG_NUMA_BALANCING
2180

2181 2182 2183
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2184
		static_branch_enable(&sched_numa_balancing);
2185
	else
2186
		static_branch_disable(&sched_numa_balancing);
2187
}
2188 2189 2190 2191 2192 2193 2194

#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;
2195
	int state = static_branch_likely(&sched_numa_balancing);
2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210

	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
2211

2212 2213
#ifdef CONFIG_SCHEDSTATS

2214
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2215
static bool __initdata __sched_schedstats = false;
2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238

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;

2239 2240 2241 2242 2243
	/*
	 * 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.
	 */
2244
	if (!strcmp(str, "enable")) {
2245
		__sched_schedstats = true;
2246 2247
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2248
		__sched_schedstats = false;
2249 2250 2251 2252 2253 2254 2255 2256 2257 2258
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2259 2260 2261 2262 2263
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283
#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;
}
2284 2285 2286 2287
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2288 2289 2290 2291

/*
 * fork()/clone()-time setup:
 */
2292
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2293
{
2294
	unsigned long flags;
I
Ingo Molnar 已提交
2295

2296
	__sched_fork(clone_flags, p);
2297
	/*
2298
	 * We mark the process as NEW here. This guarantees that
2299 2300 2301
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2302
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2303

2304 2305 2306 2307 2308
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2309 2310 2311 2312
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2313
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2314
			p->policy = SCHED_NORMAL;
2315
			p->static_prio = NICE_TO_PRIO(0);
2316 2317 2318 2319 2320
			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);
2321
		set_load_weight(p, false);
2322

2323 2324 2325 2326 2327 2328
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2329

2330
	if (dl_prio(p->prio))
2331
		return -EAGAIN;
2332
	else if (rt_prio(p->prio))
2333
		p->sched_class = &rt_sched_class;
2334
	else
H
Hiroshi Shimamoto 已提交
2335
		p->sched_class = &fair_sched_class;
2336

2337
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2338

2339 2340 2341 2342 2343 2344 2345
	/*
	 * 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.
	 */
2346
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2347
	/*
I
Ingo Molnar 已提交
2348
	 * We're setting the CPU for the first time, we don't migrate,
2349 2350
	 * so use __set_task_cpu().
	 */
2351
	__set_task_cpu(p, smp_processor_id());
2352 2353
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2354
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2355

2356
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2357
	if (likely(sched_info_on()))
2358
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2359
#endif
P
Peter Zijlstra 已提交
2360 2361
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2362
#endif
2363
	init_task_preempt_count(p);
2364
#ifdef CONFIG_SMP
2365
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2366
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2367
#endif
2368
	return 0;
L
Linus Torvalds 已提交
2369 2370
}

2371 2372 2373
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
2374
		return BW_UNIT;
2375 2376 2377 2378 2379 2380 2381 2382 2383

	/*
	 * 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;

2384
	return div64_u64(runtime << BW_SHIFT, period);
2385 2386
}

L
Linus Torvalds 已提交
2387 2388 2389 2390 2391 2392 2393
/*
 * 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.
 */
2394
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2395
{
2396
	struct rq_flags rf;
I
Ingo Molnar 已提交
2397
	struct rq *rq;
2398

2399
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2400
	p->state = TASK_RUNNING;
2401 2402 2403 2404
#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
2405
	 *  - any previously selected CPU might disappear through hotplug
2406 2407 2408
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2409
	 */
2410
	p->recent_used_cpu = task_cpu(p);
2411
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2412
#endif
2413
	rq = __task_rq_lock(p, &rf);
2414
	update_rq_clock(rq);
2415
	post_init_entity_util_avg(&p->se);
2416

2417
	activate_task(rq, p, ENQUEUE_NOCLOCK);
2418
	p->on_rq = TASK_ON_RQ_QUEUED;
2419
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2420
	check_preempt_curr(rq, p, WF_FORK);
2421
#ifdef CONFIG_SMP
2422 2423 2424 2425 2426
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2427
		rq_unpin_lock(rq, &rf);
2428
		p->sched_class->task_woken(rq, p);
2429
		rq_repin_lock(rq, &rf);
2430
	}
2431
#endif
2432
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2433 2434
}

2435 2436
#ifdef CONFIG_PREEMPT_NOTIFIERS

2437
static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key);
2438

2439 2440
void preempt_notifier_inc(void)
{
2441
	static_branch_inc(&preempt_notifier_key);
2442 2443 2444 2445 2446
}
EXPORT_SYMBOL_GPL(preempt_notifier_inc);

void preempt_notifier_dec(void)
{
2447
	static_branch_dec(&preempt_notifier_key);
2448 2449 2450
}
EXPORT_SYMBOL_GPL(preempt_notifier_dec);

2451
/**
2452
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2453
 * @notifier: notifier struct to register
2454 2455 2456
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2457
	if (!static_branch_unlikely(&preempt_notifier_key))
2458 2459
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2460 2461 2462 2463 2464 2465
	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
2466
 * @notifier: notifier struct to unregister
2467
 *
2468
 * This is *not* safe to call from within a preemption notifier.
2469 2470 2471 2472 2473 2474 2475
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2476
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2477 2478 2479
{
	struct preempt_notifier *notifier;

2480
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2481 2482 2483
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2484 2485
static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
2486
	if (static_branch_unlikely(&preempt_notifier_key))
2487 2488 2489
		__fire_sched_in_preempt_notifiers(curr);
}

2490
static void
2491 2492
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2493 2494 2495
{
	struct preempt_notifier *notifier;

2496
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2497 2498 2499
		notifier->ops->sched_out(notifier, next);
}

2500 2501 2502 2503
static __always_inline void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
2504
	if (static_branch_unlikely(&preempt_notifier_key))
2505 2506 2507
		__fire_sched_out_preempt_notifiers(curr, next);
}

2508
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2509

2510
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2511 2512 2513
{
}

2514
static inline void
2515 2516 2517 2518 2519
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2520
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2521

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
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
}

2550 2551
static inline void
prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
2552
{
2553 2554 2555 2556 2557 2558 2559 2560
	/*
	 * 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_);
2561 2562
#ifdef CONFIG_DEBUG_SPINLOCK
	/* this is a valid case when another task releases the spinlock */
2563
	rq->lock.owner = next;
2564
#endif
2565 2566 2567 2568
}

static inline void finish_lock_switch(struct rq *rq)
{
2569 2570 2571 2572 2573 2574 2575 2576 2577
	/*
	 * 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);
}

2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589
/*
 * 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

2590 2591 2592
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2593
 * @prev: the current task that is being switched out
2594 2595 2596 2597 2598 2599 2600 2601 2602
 * @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.
 */
2603 2604 2605
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2606
{
2607
	kcov_prepare_switch(prev);
2608
	sched_info_switch(rq, prev, next);
2609
	perf_event_task_sched_out(prev, next);
2610
	rseq_preempt(prev);
2611
	fire_sched_out_preempt_notifiers(prev, next);
2612
	prepare_task(next);
2613 2614 2615
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2616 2617 2618 2619
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2620 2621 2622 2623
 * 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
2624 2625
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2626
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2627 2628
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2629 2630 2631 2632 2633
 *
 * 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
2634
 */
2635
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2636 2637
	__releases(rq->lock)
{
2638
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2639
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2640
	long prev_state;
L
Linus Torvalds 已提交
2641

2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652
	/*
	 * 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.
	 */
2653 2654 2655 2656
	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);
2657

L
Linus Torvalds 已提交
2658 2659 2660 2661
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2662
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2663 2664
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2665 2666
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
2667
	 * finish_task), otherwise a concurrent wakeup can get prev
2668 2669
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2670
	 */
O
Oleg Nesterov 已提交
2671
	prev_state = prev->state;
2672
	vtime_task_switch(prev);
2673
	perf_event_task_sched_in(prev, current);
2674 2675
	finish_task(prev);
	finish_lock_switch(rq);
2676
	finish_arch_post_lock_switch();
2677
	kcov_finish_switch(current);
S
Steven Rostedt 已提交
2678

2679
	fire_sched_in_preempt_notifiers(current);
2680
	/*
2681 2682 2683 2684 2685 2686 2687 2688 2689 2690
	 * 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.
2691
	 */
2692 2693
	if (mm) {
		membarrier_mm_sync_core_before_usermode(mm);
L
Linus Torvalds 已提交
2694
		mmdrop(mm);
2695
	}
2696 2697 2698
	if (unlikely(prev_state == TASK_DEAD)) {
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);
2699

2700 2701 2702 2703 2704 2705 2706 2707 2708 2709
		/*
		 * 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);
2710
	}
2711

2712
	tick_nohz_task_switch();
2713
	return rq;
L
Linus Torvalds 已提交
2714 2715
}

2716 2717 2718
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2719
static void __balance_callback(struct rq *rq)
2720
{
2721 2722 2723
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2724

2725 2726 2727 2728 2729 2730 2731 2732
	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;
2733

2734
		func(rq);
2735
	}
2736 2737 2738 2739 2740 2741 2742
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2743 2744 2745
}

#else
2746

2747
static inline void balance_callback(struct rq *rq)
2748
{
L
Linus Torvalds 已提交
2749 2750
}

2751 2752
#endif

L
Linus Torvalds 已提交
2753 2754 2755 2756
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2757
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2758 2759
	__releases(rq->lock)
{
2760
	struct rq *rq;
2761

2762 2763 2764 2765 2766 2767 2768 2769 2770
	/*
	 * 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).
	 */

2771
	rq = finish_task_switch(prev);
2772
	balance_callback(rq);
2773
	preempt_enable();
2774

L
Linus Torvalds 已提交
2775
	if (current->set_child_tid)
2776
		put_user(task_pid_vnr(current), current->set_child_tid);
2777 2778

	calculate_sigpending();
L
Linus Torvalds 已提交
2779 2780 2781
}

/*
2782
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2783
 */
2784
static __always_inline struct rq *
2785
context_switch(struct rq *rq, struct task_struct *prev,
2786
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2787
{
I
Ingo Molnar 已提交
2788
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2789

2790
	prepare_task_switch(rq, prev, next);
2791

I
Ingo Molnar 已提交
2792 2793
	mm = next->mm;
	oldmm = prev->active_mm;
2794 2795 2796 2797 2798
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2799
	arch_start_context_switch(prev);
2800

2801 2802 2803 2804 2805 2806 2807
	/*
	 * 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.
	 */
2808
	if (!mm) {
L
Linus Torvalds 已提交
2809
		next->active_mm = oldmm;
V
Vegard Nossum 已提交
2810
		mmgrab(oldmm);
L
Linus Torvalds 已提交
2811 2812
		enter_lazy_tlb(oldmm, next);
	} else
2813
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2814

2815
	if (!prev->mm) {
L
Linus Torvalds 已提交
2816 2817 2818
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2819

2820
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2821

2822
	prepare_lock_switch(rq, next, rf);
L
Linus Torvalds 已提交
2823 2824 2825

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2826
	barrier();
2827 2828

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2829 2830 2831
}

/*
2832
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2833 2834
 *
 * externally visible scheduler statistics: current number of runnable
2835
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2836 2837 2838 2839 2840 2841 2842 2843 2844
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2845
}
L
Linus Torvalds 已提交
2846

2847
/*
I
Ingo Molnar 已提交
2848
 * Check if only the current task is running on the CPU.
2849 2850 2851 2852 2853 2854 2855 2856 2857 2858
 *
 * 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)
2859 2860 2861
 */
bool single_task_running(void)
{
2862
	return raw_rq()->nr_running == 1;
2863 2864 2865
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2866
unsigned long long nr_context_switches(void)
2867
{
2868 2869
	int i;
	unsigned long long sum = 0;
2870

2871
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2872
		sum += cpu_rq(i)->nr_switches;
2873

L
Linus Torvalds 已提交
2874 2875
	return sum;
}
2876

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
/*
 * 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
2907 2908 2909
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2910

2911
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2912
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2913

L
Linus Torvalds 已提交
2914 2915
	return sum;
}
2916

2917 2918 2919 2920 2921 2922 2923
/*
 * 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.
 */

2924
unsigned long nr_iowait_cpu(int cpu)
2925
{
2926
	struct rq *this = cpu_rq(cpu);
2927 2928
	return atomic_read(&this->nr_iowait);
}
2929

2930 2931
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2932 2933 2934
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2935 2936
}

I
Ingo Molnar 已提交
2937
#ifdef CONFIG_SMP
2938

2939
/*
P
Peter Zijlstra 已提交
2940 2941
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2942
 */
P
Peter Zijlstra 已提交
2943
void sched_exec(void)
2944
{
P
Peter Zijlstra 已提交
2945
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2946
	unsigned long flags;
2947
	int dest_cpu;
2948

2949
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2950
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2951 2952
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2953

2954
	if (likely(cpu_active(dest_cpu))) {
2955
		struct migration_arg arg = { p, dest_cpu };
2956

2957 2958
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2959 2960
		return;
	}
2961
unlock:
2962
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2963
}
I
Ingo Molnar 已提交
2964

L
Linus Torvalds 已提交
2965 2966 2967
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2968
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2969 2970

EXPORT_PER_CPU_SYMBOL(kstat);
2971
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2972

2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989
/*
 * 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);
}

2990 2991 2992 2993 2994 2995 2996
/*
 * 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)
{
2997
	struct rq_flags rf;
2998
	struct rq *rq;
2999
	u64 ns;
3000

3001 3002
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
3003
	 * 64-bit doesn't need locks to atomically read a 64-bit value.
3004 3005 3006
	 * 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
3007 3008
	 * 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
3009
	 * indistinguishable from the read occurring a few cycles earlier.
3010 3011
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3012
	 */
3013
	if (!p->on_cpu || !task_on_rq_queued(p))
3014 3015 3016
		return p->se.sum_exec_runtime;
#endif

3017
	rq = task_rq_lock(p, &rf);
3018 3019 3020 3021 3022 3023
	/*
	 * 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)) {
3024
		prefetch_curr_exec_start(p);
3025 3026 3027 3028
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3029
	task_rq_unlock(rq, p, &rf);
3030 3031 3032

	return ns;
}
3033

3034 3035 3036 3037 3038 3039 3040 3041
/*
 * 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
3042
	struct task_struct *curr = rq->curr;
3043
	struct rq_flags rf;
3044 3045

	sched_clock_tick();
I
Ingo Molnar 已提交
3046

3047 3048
	rq_lock(rq, &rf);

3049
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3050
	curr->sched_class->task_tick(rq, curr, 0);
3051
	cpu_load_update_active(rq);
3052
	calc_global_load_tick(rq);
3053 3054

	rq_unlock(rq, &rf);
3055

3056
	perf_event_task_tick();
3057

3058
#ifdef CONFIG_SMP
3059
	rq->idle_balance = idle_cpu(cpu);
3060
	trigger_load_balance(rq);
3061
#endif
L
Linus Torvalds 已提交
3062 3063
}

3064
#ifdef CONFIG_NO_HZ_FULL
3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078

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

static struct tick_work __percpu *tick_work_cpu;

static void sched_tick_remote(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct tick_work *twork = container_of(dwork, struct tick_work, work);
	int cpu = twork->cpu;
	struct rq *rq = cpu_rq(cpu);
3079
	struct task_struct *curr;
3080
	struct rq_flags rf;
3081
	u64 delta;
3082 3083 3084 3085 3086 3087 3088 3089

	/*
	 * 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.
	 */
3090 3091
	if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
		goto out_requeue;
3092

3093 3094 3095 3096
	rq_lock_irq(rq, &rf);
	curr = rq->curr;
	if (is_idle_task(curr))
		goto out_unlock;
3097

3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109
	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);
3110

3111
out_requeue:
3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
	/*
	 * Run the remote tick once per second (1Hz). This arbitrary
	 * frequency is large enough to avoid overload but short enough
	 * to keep scheduler internal stats reasonably up to date.
	 */
	queue_delayed_work(system_unbound_wq, dwork, HZ);
}

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

	if (housekeeping_cpu(cpu, HK_FLAG_TICK))
		return;

	WARN_ON_ONCE(!tick_work_cpu);

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

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

	if (housekeeping_cpu(cpu, HK_FLAG_TICK))
		return;

	WARN_ON_ONCE(!tick_work_cpu);

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

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

	return 0;
}

#else /* !CONFIG_NO_HZ_FULL */
static inline void sched_tick_start(int cpu) { }
static inline void sched_tick_stop(int cpu) { }
3161
#endif
L
Linus Torvalds 已提交
3162

3163
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
3164
				defined(CONFIG_TRACE_PREEMPT_TOGGLE))
3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178
/*
 * 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);
	}
}
3179

3180
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3181
{
3182
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3183 3184 3185
	/*
	 * Underflow?
	 */
3186 3187
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3188
#endif
3189
	__preempt_count_add(val);
3190
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3191 3192 3193
	/*
	 * Spinlock count overflowing soon?
	 */
3194 3195
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3196
#endif
3197
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3198
}
3199
EXPORT_SYMBOL(preempt_count_add);
3200
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3201

3202 3203 3204 3205 3206 3207 3208 3209 3210 3211
/*
 * 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());
}

3212
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3213
{
3214
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3215 3216 3217
	/*
	 * Underflow?
	 */
3218
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3219
		return;
L
Linus Torvalds 已提交
3220 3221 3222
	/*
	 * Is the spinlock portion underflowing?
	 */
3223 3224 3225
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3226
#endif
3227

3228
	preempt_latency_stop(val);
3229
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3230
}
3231
EXPORT_SYMBOL(preempt_count_sub);
3232
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3233

3234 3235 3236
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3237 3238
#endif

3239 3240 3241 3242 3243 3244 3245 3246 3247
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
3248
/*
I
Ingo Molnar 已提交
3249
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3250
 */
I
Ingo Molnar 已提交
3251
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3252
{
3253 3254 3255
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3256 3257 3258
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3259 3260
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3261

I
Ingo Molnar 已提交
3262
	debug_show_held_locks(prev);
3263
	print_modules();
I
Ingo Molnar 已提交
3264 3265
	if (irqs_disabled())
		print_irqtrace_events(prev);
3266 3267
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3268
		pr_err("Preemption disabled at:");
3269
		print_ip_sym(preempt_disable_ip);
3270 3271
		pr_cont("\n");
	}
3272 3273 3274
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3275
	dump_stack();
3276
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3277
}
L
Linus Torvalds 已提交
3278

I
Ingo Molnar 已提交
3279 3280 3281 3282 3283
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3284
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3285 3286
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3287
#endif
3288

3289
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3290
		__schedule_bug(prev);
3291 3292
		preempt_count_set(PREEMPT_DISABLED);
	}
3293
	rcu_sleep_check();
I
Ingo Molnar 已提交
3294

L
Linus Torvalds 已提交
3295 3296
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3297
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3298 3299 3300 3301 3302 3303
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3304
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3305
{
3306
	const struct sched_class *class;
I
Ingo Molnar 已提交
3307
	struct task_struct *p;
L
Linus Torvalds 已提交
3308 3309

	/*
3310 3311 3312 3313
	 * 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
3314
	 */
3315 3316 3317 3318
	if (likely((prev->sched_class == &idle_sched_class ||
		    prev->sched_class == &fair_sched_class) &&
		   rq->nr_running == rq->cfs.h_nr_running)) {

3319
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3320 3321 3322
		if (unlikely(p == RETRY_TASK))
			goto again;

I
Ingo Molnar 已提交
3323
		/* Assumes fair_sched_class->next == idle_sched_class */
3324
		if (unlikely(!p))
3325
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3326 3327

		return p;
L
Linus Torvalds 已提交
3328 3329
	}

3330
again:
3331
	for_each_class(class) {
3332
		p = class->pick_next_task(rq, prev, rf);
3333 3334 3335
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3336
			return p;
3337
		}
I
Ingo Molnar 已提交
3338
	}
3339

I
Ingo Molnar 已提交
3340 3341
	/* The idle class should always have a runnable task: */
	BUG();
I
Ingo Molnar 已提交
3342
}
L
Linus Torvalds 已提交
3343

I
Ingo Molnar 已提交
3344
/*
3345
 * __schedule() is the main scheduler function.
3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379
 *
 * 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
3380
 *
3381
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3382
 */
3383
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3384 3385
{
	struct task_struct *prev, *next;
3386
	unsigned long *switch_count;
3387
	struct rq_flags rf;
I
Ingo Molnar 已提交
3388
	struct rq *rq;
3389
	int cpu;
I
Ingo Molnar 已提交
3390 3391 3392 3393 3394 3395

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3396

3397
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3398
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3399

3400
	local_irq_disable();
3401
	rcu_note_context_switch(preempt);
3402

3403 3404 3405 3406
	/*
	 * 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().
3407 3408 3409
	 *
	 * The membarrier system call requires a full memory barrier
	 * after coming from user-space, before storing to rq->curr.
3410
	 */
3411
	rq_lock(rq, &rf);
3412
	smp_mb__after_spinlock();
L
Linus Torvalds 已提交
3413

I
Ingo Molnar 已提交
3414 3415
	/* Promote REQ to ACT */
	rq->clock_update_flags <<= 1;
3416
	update_rq_clock(rq);
3417

3418
	switch_count = &prev->nivcsw;
3419
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3420
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3421
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3422
		} else {
3423
			deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3424 3425
			prev->on_rq = 0;

3426 3427 3428 3429 3430
			if (prev->in_iowait) {
				atomic_inc(&rq->nr_iowait);
				delayacct_blkio_start();
			}

T
Tejun Heo 已提交
3431
			/*
3432 3433 3434
			 * 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
3435 3436 3437 3438
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3439
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3440
				if (to_wakeup)
3441
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3442 3443
			}
		}
I
Ingo Molnar 已提交
3444
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3445 3446
	}

3447
	next = pick_next_task(rq, prev, &rf);
3448
	clear_tsk_need_resched(prev);
3449
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3450 3451 3452 3453

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
3454 3455 3456
		/*
		 * The membarrier system call requires each architecture
		 * to have a full memory barrier after updating
3457 3458 3459 3460 3461 3462 3463 3464 3465 3466
		 * 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),
3467
		 */
L
Linus Torvalds 已提交
3468 3469
		++*switch_count;

3470
		trace_sched_switch(preempt, prev, next);
I
Ingo Molnar 已提交
3471 3472 3473

		/* Also unlocks the rq: */
		rq = context_switch(rq, prev, next, &rf);
3474
	} else {
3475
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3476
		rq_unlock_irq(rq, &rf);
3477
	}
L
Linus Torvalds 已提交
3478

3479
	balance_callback(rq);
L
Linus Torvalds 已提交
3480
}
3481

3482 3483
void __noreturn do_task_dead(void)
{
I
Ingo Molnar 已提交
3484
	/* Causes final put_task_struct in finish_task_switch(): */
3485
	set_special_state(TASK_DEAD);
I
Ingo Molnar 已提交
3486 3487 3488 3489

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

3490 3491
	__schedule(false);
	BUG();
I
Ingo Molnar 已提交
3492 3493

	/* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3494
	for (;;)
I
Ingo Molnar 已提交
3495
		cpu_relax();
3496 3497
}

3498 3499
static inline void sched_submit_work(struct task_struct *tsk)
{
3500
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3501 3502 3503 3504 3505 3506 3507 3508 3509
		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);
}

3510
asmlinkage __visible void __sched schedule(void)
3511
{
3512 3513 3514
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3515
	do {
3516
		preempt_disable();
3517
		__schedule(false);
3518
		sched_preempt_enable_no_resched();
3519
	} while (need_resched());
3520
}
L
Linus Torvalds 已提交
3521 3522
EXPORT_SYMBOL(schedule);

3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547
/*
 * 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());
}

3548
#ifdef CONFIG_CONTEXT_TRACKING
3549
asmlinkage __visible void __sched schedule_user(void)
3550 3551 3552 3553 3554 3555
{
	/*
	 * 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.
3556 3557
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3558
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3559
	 * too frequently to make sense yet.
3560
	 */
3561
	enum ctx_state prev_state = exception_enter();
3562
	schedule();
3563
	exception_exit(prev_state);
3564 3565 3566
}
#endif

3567 3568 3569 3570 3571 3572 3573
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3574
	sched_preempt_enable_no_resched();
3575 3576 3577 3578
	schedule();
	preempt_disable();
}

3579
static void __sched notrace preempt_schedule_common(void)
3580 3581
{
	do {
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594
		/*
		 * 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.
		 */
3595
		preempt_disable_notrace();
3596
		preempt_latency_start(1);
3597
		__schedule(true);
3598
		preempt_latency_stop(1);
3599
		preempt_enable_no_resched_notrace();
3600 3601 3602 3603 3604 3605 3606 3607

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

L
Linus Torvalds 已提交
3608 3609
#ifdef CONFIG_PREEMPT
/*
3610
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3611
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3612 3613
 * occur there and call schedule directly.
 */
3614
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3615 3616 3617
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3618
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3619
	 */
3620
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3621 3622
		return;

3623
	preempt_schedule_common();
L
Linus Torvalds 已提交
3624
}
3625
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3626
EXPORT_SYMBOL(preempt_schedule);
3627 3628

/**
3629
 * preempt_schedule_notrace - preempt_schedule called by tracing
3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641
 *
 * 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.
 */
3642
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3643 3644 3645 3646 3647 3648 3649
{
	enum ctx_state prev_ctx;

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

	do {
3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662
		/*
		 * 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.
		 */
3663
		preempt_disable_notrace();
3664
		preempt_latency_start(1);
3665 3666 3667 3668 3669 3670
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3671
		__schedule(true);
3672 3673
		exception_exit(prev_ctx);

3674
		preempt_latency_stop(1);
3675
		preempt_enable_no_resched_notrace();
3676 3677
	} while (need_resched());
}
3678
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3679

3680
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3681 3682

/*
3683
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3684 3685 3686 3687
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3688
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3689
{
3690
	enum ctx_state prev_state;
3691

3692
	/* Catch callers which need to be fixed */
3693
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3694

3695 3696
	prev_state = exception_enter();

3697
	do {
3698
		preempt_disable();
3699
		local_irq_enable();
3700
		__schedule(true);
3701
		local_irq_disable();
3702
		sched_preempt_enable_no_resched();
3703
	} while (need_resched());
3704 3705

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3706 3707
}

3708
int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3709
			  void *key)
L
Linus Torvalds 已提交
3710
{
P
Peter Zijlstra 已提交
3711
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3712 3713 3714
}
EXPORT_SYMBOL(default_wake_function);

3715 3716
#ifdef CONFIG_RT_MUTEXES

3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731
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);
}

3732 3733
/*
 * rt_mutex_setprio - set the current priority of a task
3734 3735
 * @p: task to boost
 * @pi_task: donor task
3736 3737 3738 3739
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3740 3741
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3742
 */
3743
void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3744
{
3745
	int prio, oldprio, queued, running, queue_flag =
3746
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3747
	const struct sched_class *prev_class;
3748 3749
	struct rq_flags rf;
	struct rq *rq;
3750

3751 3752 3753 3754 3755 3756 3757 3758
	/* 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;
3759

3760
	rq = __task_rq_lock(p, &rf);
3761
	update_rq_clock(rq);
3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778
	/*
	 * 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;
3779

3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797
	/*
	 * 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;
	}

3798
	trace_sched_pi_setprio(p, pi_task);
3799
	oldprio = p->prio;
3800 3801 3802 3803

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

3804
	prev_class = p->sched_class;
3805
	queued = task_on_rq_queued(p);
3806
	running = task_current(rq, p);
3807
	if (queued)
3808
		dequeue_task(rq, p, queue_flag);
3809
	if (running)
3810
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3811

3812 3813 3814 3815 3816 3817 3818 3819 3820 3821
	/*
	 * 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)) {
3822 3823
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3824
			p->dl.dl_boosted = 1;
3825
			queue_flag |= ENQUEUE_REPLENISH;
3826 3827
		} else
			p->dl.dl_boosted = 0;
3828
		p->sched_class = &dl_sched_class;
3829 3830 3831 3832
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3833
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3834
		p->sched_class = &rt_sched_class;
3835 3836 3837
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3838 3839
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3840
		p->sched_class = &fair_sched_class;
3841
	}
I
Ingo Molnar 已提交
3842

3843 3844
	p->prio = prio;

3845
	if (queued)
3846
		enqueue_task(rq, p, queue_flag);
3847
	if (running)
3848
		set_curr_task(rq, p);
3849

P
Peter Zijlstra 已提交
3850
	check_class_changed(rq, p, prev_class, oldprio);
3851
out_unlock:
I
Ingo Molnar 已提交
3852 3853
	/* Avoid rq from going away on us: */
	preempt_disable();
3854
	__task_rq_unlock(rq, &rf);
3855 3856 3857

	balance_callback(rq);
	preempt_enable();
3858
}
3859 3860 3861 3862 3863
#else
static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	return prio;
}
3864
#endif
3865

3866
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3867
{
P
Peter Zijlstra 已提交
3868 3869
	bool queued, running;
	int old_prio, delta;
3870
	struct rq_flags rf;
3871
	struct rq *rq;
L
Linus Torvalds 已提交
3872

3873
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3874 3875 3876 3877 3878
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3879
	rq = task_rq_lock(p, &rf);
3880 3881
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3882 3883 3884 3885
	/*
	 * 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
3886
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3887
	 */
3888
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3889 3890 3891
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3892
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3893
	running = task_current(rq, p);
3894
	if (queued)
3895
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
3896 3897
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3898 3899

	p->static_prio = NICE_TO_PRIO(nice);
3900
	set_load_weight(p, true);
3901 3902 3903
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3904

3905
	if (queued) {
3906
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
L
Linus Torvalds 已提交
3907
		/*
3908 3909
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3910
		 */
3911
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3912
			resched_curr(rq);
L
Linus Torvalds 已提交
3913
	}
P
Peter Zijlstra 已提交
3914 3915
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3916
out_unlock:
3917
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3918 3919 3920
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3921 3922 3923 3924 3925
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3926
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3927
{
I
Ingo Molnar 已提交
3928
	/* Convert nice value [19,-20] to rlimit style value [1,40]: */
3929
	int nice_rlim = nice_to_rlimit(nice);
3930

3931
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3932 3933 3934
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3935 3936 3937 3938 3939 3940 3941 3942 3943
#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.
 */
3944
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3945
{
3946
	long nice, retval;
L
Linus Torvalds 已提交
3947 3948 3949 3950 3951 3952

	/*
	 * 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.
	 */
3953
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3954
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3955

3956
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3957 3958 3959
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973
	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.
 *
3974
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3975 3976 3977
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3978
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3979 3980 3981 3982 3983
{
	return p->prio - MAX_RT_PRIO;
}

/**
I
Ingo Molnar 已提交
3984
 * idle_cpu - is a given CPU idle currently?
L
Linus Torvalds 已提交
3985
 * @cpu: the processor in question.
3986 3987
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3988 3989 3990
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004
	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
4005 4006
}

4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017
/**
 * 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;

4018 4019 4020
	if (vcpu_is_preempted(cpu))
		return 0;

T
Thomas Gleixner 已提交
4021
	return 1;
L
Linus Torvalds 已提交
4022 4023 4024
}

/**
I
Ingo Molnar 已提交
4025
 * idle_task - return the idle task for a given CPU.
L
Linus Torvalds 已提交
4026
 * @cpu: the processor in question.
4027
 *
I
Ingo Molnar 已提交
4028
 * Return: The idle task for the CPU @cpu.
L
Linus Torvalds 已提交
4029
 */
4030
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
4031 4032 4033 4034 4035 4036 4037
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
4038 4039
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
4040
 */
A
Alexey Dobriyan 已提交
4041
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
4042
{
4043
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
4044 4045
}

4046 4047 4048 4049 4050 4051
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

4052 4053
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
4054
{
4055 4056
	int policy = attr->sched_policy;

4057
	if (policy == SETPARAM_POLICY)
4058 4059
		policy = p->policy;

L
Linus Torvalds 已提交
4060
	p->policy = policy;
4061

4062 4063
	if (dl_policy(policy))
		__setparam_dl(p, attr);
4064
	else if (fair_policy(policy))
4065 4066
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

4067 4068 4069 4070 4071 4072
	/*
	 * __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;
4073
	p->normal_prio = normal_prio(p);
4074
	set_load_weight(p, true);
4075
}
4076

4077 4078
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
4079
			   const struct sched_attr *attr, bool keep_boost)
4080 4081
{
	__setscheduler_params(p, attr);
4082

4083
	/*
4084 4085
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
4086
	 */
4087
	p->prio = normal_prio(p);
4088
	if (keep_boost)
4089
		p->prio = rt_effective_prio(p, p->prio);
4090

4091 4092 4093
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
4094 4095 4096
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
4097
}
4098

4099
/*
I
Ingo Molnar 已提交
4100
 * Check the target process has a UID that matches the current process's:
4101 4102 4103 4104 4105 4106 4107 4108
 */
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);
4109 4110
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4111 4112 4113 4114
	rcu_read_unlock();
	return match;
}

4115 4116
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4117
				bool user, bool pi)
L
Linus Torvalds 已提交
4118
{
4119 4120
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4121
	int retval, oldprio, oldpolicy = -1, queued, running;
4122
	int new_effective_prio, policy = attr->sched_policy;
4123
	const struct sched_class *prev_class;
4124
	struct rq_flags rf;
4125
	int reset_on_fork;
4126
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4127
	struct rq *rq;
L
Linus Torvalds 已提交
4128

4129 4130
	/* The pi code expects interrupts enabled */
	BUG_ON(pi && in_interrupt());
L
Linus Torvalds 已提交
4131
recheck:
I
Ingo Molnar 已提交
4132
	/* Double check policy once rq lock held: */
4133 4134
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4135
		policy = oldpolicy = p->policy;
4136
	} else {
4137
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4138

4139
		if (!valid_policy(policy))
4140 4141 4142
			return -EINVAL;
	}

4143
	if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4144 4145
		return -EINVAL;

L
Linus Torvalds 已提交
4146 4147
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4148 4149
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4150
	 */
4151
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4152
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4153
		return -EINVAL;
4154 4155
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4156 4157
		return -EINVAL;

4158 4159 4160
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4161
	if (user && !capable(CAP_SYS_NICE)) {
4162
		if (fair_policy(policy)) {
4163
			if (attr->sched_nice < task_nice(p) &&
4164
			    !can_nice(p, attr->sched_nice))
4165 4166 4167
				return -EPERM;
		}

4168
		if (rt_policy(policy)) {
4169 4170
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4171

I
Ingo Molnar 已提交
4172
			/* Can't set/change the rt policy: */
4173 4174 4175
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

I
Ingo Molnar 已提交
4176
			/* Can't increase priority: */
4177 4178
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4179 4180
				return -EPERM;
		}
4181

4182 4183 4184 4185 4186 4187 4188 4189 4190
		 /*
		  * 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
4191
		/*
4192 4193
		 * 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
4194
		 */
4195
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4196
			if (!can_nice(p, task_nice(p)))
4197 4198
				return -EPERM;
		}
4199

I
Ingo Molnar 已提交
4200
		/* Can't change other user's priorities: */
4201
		if (!check_same_owner(p))
4202
			return -EPERM;
4203

I
Ingo Molnar 已提交
4204
		/* Normal users shall not reset the sched_reset_on_fork flag: */
4205 4206
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4207
	}
L
Linus Torvalds 已提交
4208

4209
	if (user) {
4210 4211 4212
		if (attr->sched_flags & SCHED_FLAG_SUGOV)
			return -EINVAL;

4213
		retval = security_task_setscheduler(p);
4214 4215 4216 4217
		if (retval)
			return retval;
	}

4218
	/*
I
Ingo Molnar 已提交
4219
	 * Make sure no PI-waiters arrive (or leave) while we are
4220
	 * changing the priority of the task:
4221
	 *
L
Lucas De Marchi 已提交
4222
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4223 4224
	 * runqueue lock must be held.
	 */
4225
	rq = task_rq_lock(p, &rf);
4226
	update_rq_clock(rq);
4227

4228
	/*
I
Ingo Molnar 已提交
4229
	 * Changing the policy of the stop threads its a very bad idea:
4230 4231
	 */
	if (p == rq->stop) {
4232
		task_rq_unlock(rq, p, &rf);
4233 4234 4235
		return -EINVAL;
	}

4236
	/*
4237 4238
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4239
	 */
4240
	if (unlikely(policy == p->policy)) {
4241
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4242 4243 4244
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4245
		if (dl_policy(policy) && dl_param_changed(p, attr))
4246
			goto change;
4247

4248
		p->sched_reset_on_fork = reset_on_fork;
4249
		task_rq_unlock(rq, p, &rf);
4250 4251
		return 0;
	}
4252
change:
4253

4254
	if (user) {
4255
#ifdef CONFIG_RT_GROUP_SCHED
4256 4257 4258 4259 4260
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4261 4262
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4263
			task_rq_unlock(rq, p, &rf);
4264 4265 4266
			return -EPERM;
		}
#endif
4267
#ifdef CONFIG_SMP
4268 4269
		if (dl_bandwidth_enabled() && dl_policy(policy) &&
				!(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4270 4271 4272 4273 4274 4275 4276
			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.
			 */
4277 4278
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4279
				task_rq_unlock(rq, p, &rf);
4280 4281 4282 4283 4284
				return -EPERM;
			}
		}
#endif
	}
4285

I
Ingo Molnar 已提交
4286
	/* Re-check policy now with rq lock held: */
L
Linus Torvalds 已提交
4287 4288
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4289
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4290 4291
		goto recheck;
	}
4292 4293 4294 4295 4296 4297

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4298
	if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4299
		task_rq_unlock(rq, p, &rf);
4300 4301 4302
		return -EBUSY;
	}

4303 4304 4305
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4306 4307 4308 4309 4310 4311 4312 4313
	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.
		 */
4314
		new_effective_prio = rt_effective_prio(p, newprio);
4315 4316
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4317 4318
	}

4319
	queued = task_on_rq_queued(p);
4320
	running = task_current(rq, p);
4321
	if (queued)
4322
		dequeue_task(rq, p, queue_flags);
4323
	if (running)
4324
		put_prev_task(rq, p);
4325

4326
	prev_class = p->sched_class;
4327
	__setscheduler(rq, p, attr, pi);
4328

4329
	if (queued) {
4330 4331 4332 4333
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4334 4335
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4336

4337
		enqueue_task(rq, p, queue_flags);
4338
	}
4339
	if (running)
4340
		set_curr_task(rq, p);
4341

P
Peter Zijlstra 已提交
4342
	check_class_changed(rq, p, prev_class, oldprio);
I
Ingo Molnar 已提交
4343 4344 4345

	/* Avoid rq from going away on us: */
	preempt_disable();
4346
	task_rq_unlock(rq, p, &rf);
4347

4348 4349
	if (pi)
		rt_mutex_adjust_pi(p);
4350

I
Ingo Molnar 已提交
4351
	/* Run balance callbacks after we've adjusted the PI chain: */
4352 4353
	balance_callback(rq);
	preempt_enable();
4354

L
Linus Torvalds 已提交
4355 4356
	return 0;
}
4357

4358 4359 4360 4361 4362 4363 4364 4365 4366
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),
	};

4367 4368
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4369 4370 4371 4372 4373
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4374
	return __sched_setscheduler(p, &attr, check, true);
4375
}
4376 4377 4378 4379 4380 4381
/**
 * 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.
 *
4382 4383
 * Return: 0 on success. An error code otherwise.
 *
4384 4385 4386
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4387
		       const struct sched_param *param)
4388
{
4389
	return _sched_setscheduler(p, policy, param, true);
4390
}
L
Linus Torvalds 已提交
4391 4392
EXPORT_SYMBOL_GPL(sched_setscheduler);

4393 4394
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4395
	return __sched_setscheduler(p, attr, true, true);
4396 4397 4398
}
EXPORT_SYMBOL_GPL(sched_setattr);

4399 4400 4401 4402 4403
int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
{
	return __sched_setscheduler(p, attr, false, true);
}

4404 4405 4406 4407 4408 4409 4410 4411 4412 4413
/**
 * 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.
4414 4415
 *
 * Return: 0 on success. An error code otherwise.
4416 4417
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4418
			       const struct sched_param *param)
4419
{
4420
	return _sched_setscheduler(p, policy, param, false);
4421
}
4422
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4423

I
Ingo Molnar 已提交
4424 4425
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4426 4427 4428
{
	struct sched_param lparam;
	struct task_struct *p;
4429
	int retval;
L
Linus Torvalds 已提交
4430 4431 4432 4433 4434

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4435 4436 4437

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4438
	p = find_process_by_pid(pid);
4439 4440 4441
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4442

L
Linus Torvalds 已提交
4443 4444 4445
	return retval;
}

4446 4447 4448
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
I
Ingo Molnar 已提交
4449
static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4450 4451 4452 4453 4454 4455 4456
{
	u32 size;
	int ret;

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

I
Ingo Molnar 已提交
4457
	/* Zero the full structure, so that a short copy will be nice: */
4458 4459 4460 4461 4462 4463
	memset(attr, 0, sizeof(*attr));

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

I
Ingo Molnar 已提交
4464 4465
	/* Bail out on silly large: */
	if (size > PAGE_SIZE)
4466 4467
		goto err_size;

I
Ingo Molnar 已提交
4468 4469
	/* ABI compatibility quirk: */
	if (!size)
4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503
		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
4504
	 * XXX: Do we want to be lenient like existing syscalls; or do we want
4505 4506
	 * to be strict and return an error on out-of-bounds values?
	 */
4507
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4508

4509
	return 0;
4510 4511 4512

err_size:
	put_user(sizeof(*attr), &uattr->size);
4513
	return -E2BIG;
4514 4515
}

L
Linus Torvalds 已提交
4516 4517 4518 4519 4520
/**
 * 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.
4521 4522
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4523
 */
I
Ingo Molnar 已提交
4524
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4525
{
4526 4527 4528
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4529 4530 4531 4532 4533 4534 4535
	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.
4536 4537
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4538
 */
4539
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4540
{
4541
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4542 4543
}

4544 4545 4546
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4547
 * @uattr: structure containing the extended parameters.
4548
 * @flags: for future extension.
4549
 */
4550 4551
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4552 4553 4554 4555 4556
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4557
	if (!uattr || pid < 0 || flags)
4558 4559
		return -EINVAL;

4560 4561 4562
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4563

4564
	if ((int)attr.sched_policy < 0)
4565
		return -EINVAL;
4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576

	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
4577 4578 4579
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4580 4581 4582
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4583
 */
4584
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4585
{
4586
	struct task_struct *p;
4587
	int retval;
L
Linus Torvalds 已提交
4588 4589

	if (pid < 0)
4590
		return -EINVAL;
L
Linus Torvalds 已提交
4591 4592

	retval = -ESRCH;
4593
	rcu_read_lock();
L
Linus Torvalds 已提交
4594 4595 4596 4597
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4598 4599
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4600
	}
4601
	rcu_read_unlock();
L
Linus Torvalds 已提交
4602 4603 4604 4605
	return retval;
}

/**
4606
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4607 4608
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4609 4610 4611
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4612
 */
4613
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4614
{
4615
	struct sched_param lp = { .sched_priority = 0 };
4616
	struct task_struct *p;
4617
	int retval;
L
Linus Torvalds 已提交
4618 4619

	if (!param || pid < 0)
4620
		return -EINVAL;
L
Linus Torvalds 已提交
4621

4622
	rcu_read_lock();
L
Linus Torvalds 已提交
4623 4624 4625 4626 4627 4628 4629 4630 4631
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4632 4633
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4634
	rcu_read_unlock();
L
Linus Torvalds 已提交
4635 4636 4637 4638 4639 4640 4641 4642 4643

	/*
	 * 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:
4644
	rcu_read_unlock();
L
Linus Torvalds 已提交
4645 4646 4647
	return retval;
}

4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670
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)
4671
				return -EFBIG;
4672 4673 4674 4675 4676
		}

		attr->size = usize;
	}

4677
	ret = copy_to_user(uattr, attr, attr->size);
4678 4679 4680
	if (ret)
		return -EFAULT;

4681
	return 0;
4682 4683 4684
}

/**
4685
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4686
 * @pid: the pid in question.
J
Juri Lelli 已提交
4687
 * @uattr: structure containing the extended parameters.
4688
 * @size: sizeof(attr) for fwd/bwd comp.
4689
 * @flags: for future extension.
4690
 */
4691 4692
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4693 4694 4695 4696 4697 4698 4699 4700
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4701
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714
		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;
4715 4716
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4717 4718 4719
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4720 4721
		attr.sched_priority = p->rt_priority;
	else
4722
		attr.sched_nice = task_nice(p);
4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4734
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4735
{
4736
	cpumask_var_t cpus_allowed, new_mask;
4737 4738
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4739

4740
	rcu_read_lock();
L
Linus Torvalds 已提交
4741 4742 4743

	p = find_process_by_pid(pid);
	if (!p) {
4744
		rcu_read_unlock();
L
Linus Torvalds 已提交
4745 4746 4747
		return -ESRCH;
	}

4748
	/* Prevent p going away */
L
Linus Torvalds 已提交
4749
	get_task_struct(p);
4750
	rcu_read_unlock();
L
Linus Torvalds 已提交
4751

4752 4753 4754 4755
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4756 4757 4758 4759 4760 4761 4762 4763
	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
4764
	retval = -EPERM;
E
Eric W. Biederman 已提交
4765 4766 4767 4768
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4769
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4770 4771 4772
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4773

4774
	retval = security_task_setscheduler(p);
4775
	if (retval)
4776
		goto out_free_new_mask;
4777

4778 4779 4780 4781

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

4782 4783 4784 4785 4786 4787 4788
	/*
	 * 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
4789 4790 4791
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4792
			retval = -EBUSY;
4793
			rcu_read_unlock();
4794
			goto out_free_new_mask;
4795
		}
4796
		rcu_read_unlock();
4797 4798
	}
#endif
P
Peter Zijlstra 已提交
4799
again:
4800
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4801

P
Paul Menage 已提交
4802
	if (!retval) {
4803 4804
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4805 4806 4807 4808 4809
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4810
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4811 4812 4813
			goto again;
		}
	}
4814
out_free_new_mask:
4815 4816 4817 4818
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4819 4820 4821 4822 4823
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4824
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4825
{
4826 4827 4828 4829 4830
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4831 4832 4833 4834
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
I
Ingo Molnar 已提交
4835
 * sys_sched_setaffinity - set the CPU affinity of a process
L
Linus Torvalds 已提交
4836 4837
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4838
 * @user_mask_ptr: user-space pointer to the new CPU mask
4839 4840
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4841
 */
4842 4843
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4844
{
4845
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4846 4847
	int retval;

4848 4849
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4850

4851 4852 4853 4854 4855
	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
4856 4857
}

4858
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4859
{
4860
	struct task_struct *p;
4861
	unsigned long flags;
L
Linus Torvalds 已提交
4862 4863
	int retval;

4864
	rcu_read_lock();
L
Linus Torvalds 已提交
4865 4866 4867 4868 4869 4870

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

4871 4872 4873 4874
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4875
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4876
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4877
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4878 4879

out_unlock:
4880
	rcu_read_unlock();
L
Linus Torvalds 已提交
4881

4882
	return retval;
L
Linus Torvalds 已提交
4883 4884 4885
}

/**
I
Ingo Molnar 已提交
4886
 * sys_sched_getaffinity - get the CPU affinity of a process
L
Linus Torvalds 已提交
4887 4888
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4889
 * @user_mask_ptr: user-space pointer to hold the current CPU mask
4890
 *
4891 4892
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4893
 */
4894 4895
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4896 4897
{
	int ret;
4898
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4899

A
Anton Blanchard 已提交
4900
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4901 4902
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4903 4904
		return -EINVAL;

4905 4906
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4907

4908 4909
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4910
		unsigned int retlen = min(len, cpumask_size());
4911 4912

		if (copy_to_user(user_mask_ptr, mask, retlen))
4913 4914
			ret = -EFAULT;
		else
4915
			ret = retlen;
4916 4917
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4918

4919
	return ret;
L
Linus Torvalds 已提交
4920 4921 4922 4923 4924
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4925 4926
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4927 4928
 *
 * Return: 0.
L
Linus Torvalds 已提交
4929
 */
4930
static void do_sched_yield(void)
L
Linus Torvalds 已提交
4931
{
4932 4933 4934 4935 4936 4937
	struct rq_flags rf;
	struct rq *rq;

	local_irq_disable();
	rq = this_rq();
	rq_lock(rq, &rf);
L
Linus Torvalds 已提交
4938

4939
	schedstat_inc(rq->yld_count);
4940
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4941 4942 4943 4944 4945

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
4946 4947
	preempt_disable();
	rq_unlock(rq, &rf);
4948
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4949 4950

	schedule();
4951
}
L
Linus Torvalds 已提交
4952

4953 4954 4955
SYSCALL_DEFINE0(sched_yield)
{
	do_sched_yield();
L
Linus Torvalds 已提交
4956 4957 4958
	return 0;
}

4959
#ifndef CONFIG_PREEMPT
4960
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4961
{
4962
	if (should_resched(0)) {
4963
		preempt_schedule_common();
L
Linus Torvalds 已提交
4964 4965
		return 1;
	}
4966
	rcu_all_qs();
L
Linus Torvalds 已提交
4967 4968
	return 0;
}
4969
EXPORT_SYMBOL(_cond_resched);
4970
#endif
L
Linus Torvalds 已提交
4971 4972

/*
4973
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4974 4975
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4976
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4977 4978 4979
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4980
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4981
{
4982
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4983 4984
	int ret = 0;

4985 4986
	lockdep_assert_held(lock);

4987
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4988
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4989
		if (resched)
4990
			preempt_schedule_common();
N
Nick Piggin 已提交
4991 4992
		else
			cpu_relax();
J
Jan Kara 已提交
4993
		ret = 1;
L
Linus Torvalds 已提交
4994 4995
		spin_lock(lock);
	}
J
Jan Kara 已提交
4996
	return ret;
L
Linus Torvalds 已提交
4997
}
4998
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4999 5000 5001 5002

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
5003 5004 5005 5006 5007 5008 5009 5010 5011
 * 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
5012
 *	yield();
P
Peter Zijlstra 已提交
5013 5014 5015 5016 5017 5018 5019 5020
 *
 * 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
5021 5022 5023 5024
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
5025
	do_sched_yield();
L
Linus Torvalds 已提交
5026 5027 5028
}
EXPORT_SYMBOL(yield);

5029 5030 5031 5032
/**
 * 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
5033 5034
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5035 5036 5037 5038
 *
 * 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.
 *
5039
 * Return:
5040 5041 5042
 *	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.
5043
 */
5044
int __sched yield_to(struct task_struct *p, bool preempt)
5045 5046 5047 5048
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5049
	int yielded = 0;
5050 5051 5052 5053 5054 5055

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5056 5057 5058 5059 5060 5061 5062 5063 5064
	/*
	 * 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;
	}

5065
	double_rq_lock(rq, p_rq);
5066
	if (task_rq(p) != p_rq) {
5067 5068 5069 5070 5071
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5072
		goto out_unlock;
5073 5074

	if (curr->sched_class != p->sched_class)
5075
		goto out_unlock;
5076 5077

	if (task_running(p_rq, p) || p->state)
5078
		goto out_unlock;
5079 5080

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5081
	if (yielded) {
5082
		schedstat_inc(rq->yld_count);
5083 5084 5085 5086 5087
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5088
			resched_curr(p_rq);
5089
	}
5090

5091
out_unlock:
5092
	double_rq_unlock(rq, p_rq);
5093
out_irq:
5094 5095
	local_irq_restore(flags);

5096
	if (yielded > 0)
5097 5098 5099 5100 5101 5102
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117
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
5118
/*
I
Ingo Molnar 已提交
5119
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5120 5121 5122 5123
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5124
	int token;
L
Linus Torvalds 已提交
5125 5126
	long ret;

5127
	token = io_schedule_prepare();
L
Linus Torvalds 已提交
5128
	ret = schedule_timeout(timeout);
5129
	io_schedule_finish(token);
5130

L
Linus Torvalds 已提交
5131 5132
	return ret;
}
5133
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5134

5135 5136 5137 5138 5139 5140 5141 5142 5143 5144
void io_schedule(void)
{
	int token;

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

L
Linus Torvalds 已提交
5145 5146 5147 5148
/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5149 5150 5151
 * 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
5152
 */
5153
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5154 5155 5156 5157 5158 5159 5160 5161
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5162
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5163
	case SCHED_NORMAL:
5164
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5165
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5166 5167 5168 5169 5170 5171 5172 5173 5174 5175
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5176 5177 5178
 * 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
5179
 */
5180
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5181 5182 5183 5184 5185 5186 5187 5188
{
	int ret = -EINVAL;

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

5198
static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
L
Linus Torvalds 已提交
5199
{
5200
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5201
	unsigned int time_slice;
5202
	struct rq_flags rf;
5203
	struct rq *rq;
5204
	int retval;
L
Linus Torvalds 已提交
5205 5206

	if (pid < 0)
5207
		return -EINVAL;
L
Linus Torvalds 已提交
5208 5209

	retval = -ESRCH;
5210
	rcu_read_lock();
L
Linus Torvalds 已提交
5211 5212 5213 5214 5215 5216 5217 5218
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5219
	rq = task_rq_lock(p, &rf);
5220 5221 5222
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5223
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5224

5225
	rcu_read_unlock();
5226 5227
	jiffies_to_timespec64(time_slice, t);
	return 0;
5228

L
Linus Torvalds 已提交
5229
out_unlock:
5230
	rcu_read_unlock();
L
Linus Torvalds 已提交
5231 5232 5233
	return retval;
}

5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244
/**
 * 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.
 */
5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270
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

5271
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5272 5273
{
	unsigned long free = 0;
5274
	int ppid;
5275

5276 5277
	if (!try_get_task_stack(p))
		return;
5278 5279 5280 5281

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

	if (p->state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5282
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5283
#ifdef CONFIG_DEBUG_STACK_USAGE
5284
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5285
#endif
5286
	ppid = 0;
5287
	rcu_read_lock();
5288 5289
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5290
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5291
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5292
		task_pid_nr(p), ppid,
5293
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5294

5295
	print_worker_info(KERN_INFO, p);
5296
	show_stack(p, NULL);
5297
	put_task_stack(p);
L
Linus Torvalds 已提交
5298
}
5299
EXPORT_SYMBOL_GPL(sched_show_task);
L
Linus Torvalds 已提交
5300

5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322
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
5323
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5324
{
5325
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5326

5327
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5328 5329
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5330
#else
P
Peter Zijlstra 已提交
5331 5332
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5333
#endif
5334
	rcu_read_lock();
5335
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5336 5337
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5338
		 * console might take a lot of time:
5339 5340 5341
		 * 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
5342 5343
		 */
		touch_nmi_watchdog();
5344
		touch_all_softlockup_watchdogs();
5345
		if (state_filter_match(state_filter, p))
5346
			sched_show_task(p);
5347
	}
L
Linus Torvalds 已提交
5348

I
Ingo Molnar 已提交
5349
#ifdef CONFIG_SCHED_DEBUG
5350 5351
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5352
#endif
5353
	rcu_read_unlock();
I
Ingo Molnar 已提交
5354 5355 5356
	/*
	 * Only show locks if all tasks are dumped:
	 */
5357
	if (!state_filter)
I
Ingo Molnar 已提交
5358
		debug_show_all_locks();
L
Linus Torvalds 已提交
5359 5360
}

5361 5362 5363
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
I
Ingo Molnar 已提交
5364
 * @cpu: CPU the idle task belongs to
5365 5366 5367 5368
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5369
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5370
{
5371
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5372 5373
	unsigned long flags;

5374 5375
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5376

5377
	__sched_fork(0, idle);
5378
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5379
	idle->se.exec_start = sched_clock();
5380
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5381

5382 5383
	kasan_unpoison_task_stack(idle);

5384 5385 5386 5387 5388 5389 5390 5391 5392
#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
5393 5394
	/*
	 * We're having a chicken and egg problem, even though we are
I
Ingo Molnar 已提交
5395
	 * holding rq->lock, the CPU isn't yet set to this CPU so the
5396 5397 5398 5399 5400 5401 5402 5403
	 * 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
5404
	__set_task_cpu(idle, cpu);
5405
	rcu_read_unlock();
L
Linus Torvalds 已提交
5406 5407

	rq->curr = rq->idle = idle;
5408
	idle->on_rq = TASK_ON_RQ_QUEUED;
5409
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5410
	idle->on_cpu = 1;
5411
#endif
5412 5413
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5414 5415

	/* Set the preempt count _outside_ the spinlocks! */
5416
	init_idle_preempt_count(idle, cpu);
5417

I
Ingo Molnar 已提交
5418 5419 5420 5421
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5422
	ftrace_graph_init_idle_task(idle, cpu);
5423
	vtime_init_idle(idle, cpu);
5424
#ifdef CONFIG_SMP
5425 5426
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5427 5428
}

5429 5430
#ifdef CONFIG_SMP

5431 5432 5433
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
5434
	int ret = 1;
5435

5436 5437 5438
	if (!cpumask_weight(cur))
		return ret;

5439
	ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5440 5441 5442 5443

	return ret;
}

5444 5445 5446 5447 5448 5449 5450
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
5451
	 * to a new cpuset; we don't want to change their CPU
5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463
	 * 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,
5464 5465
					      cs_cpus_allowed))
		ret = dl_task_can_attach(p, cs_cpus_allowed);
5466 5467 5468 5469 5470

out:
	return ret;
}

5471
bool sched_smp_initialized __read_mostly;
5472

5473 5474 5475 5476 5477 5478 5479 5480 5481 5482
#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;

5483
	if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5484 5485 5486 5487
		return -EINVAL;

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

5488
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5489 5490
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5491 5492 5493 5494 5495 5496 5497

/*
 * 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)
{
5498
	bool queued, running;
5499 5500
	struct rq_flags rf;
	struct rq *rq;
5501

5502
	rq = task_rq_lock(p, &rf);
5503
	queued = task_on_rq_queued(p);
5504 5505
	running = task_current(rq, p);

5506
	if (queued)
5507
		dequeue_task(rq, p, DEQUEUE_SAVE);
5508
	if (running)
5509
		put_prev_task(rq, p);
5510 5511 5512

	p->numa_preferred_nid = nid;

5513
	if (queued)
5514
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5515
	if (running)
5516
		set_curr_task(rq, p);
5517
	task_rq_unlock(rq, p, &rf);
5518
}
P
Peter Zijlstra 已提交
5519
#endif /* CONFIG_NUMA_BALANCING */
5520

L
Linus Torvalds 已提交
5521
#ifdef CONFIG_HOTPLUG_CPU
5522
/*
I
Ingo Molnar 已提交
5523
 * Ensure that the idle task is using init_mm right before its CPU goes
5524
 * offline.
5525
 */
5526
void idle_task_exit(void)
L
Linus Torvalds 已提交
5527
{
5528
	struct mm_struct *mm = current->active_mm;
5529

5530
	BUG_ON(cpu_online(smp_processor_id()));
5531

5532
	if (mm != &init_mm) {
5533
		switch_mm(mm, &init_mm, current);
5534
		current->active_mm = &init_mm;
5535 5536
		finish_arch_post_lock_switch();
	}
5537
	mmdrop(mm);
L
Linus Torvalds 已提交
5538 5539 5540
}

/*
5541 5542
 * 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
5543 5544 5545
 * 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.
5546 5547
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5548
 */
5549
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5550
{
5551
	long delta = calc_load_fold_active(rq, 1);
5552 5553
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5554 5555
}

5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571
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,
};

5572
/*
5573 5574 5575 5576 5577 5578
 * 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
5579
 */
5580
static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
L
Linus Torvalds 已提交
5581
{
5582
	struct rq *rq = dead_rq;
5583
	struct task_struct *next, *stop = rq->stop;
5584
	struct rq_flags orf = *rf;
5585
	int dest_cpu;
L
Linus Torvalds 已提交
5586 5587

	/*
5588 5589 5590 5591 5592 5593 5594
	 * 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
5595
	 */
5596
	rq->stop = NULL;
5597

5598 5599 5600 5601 5602 5603 5604
	/*
	 * 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);

5605
	for (;;) {
5606 5607
		/*
		 * There's this thread running, bail when that's the only
I
Ingo Molnar 已提交
5608
		 * remaining thread:
5609 5610
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5611
			break;
5612

5613
		/*
I
Ingo Molnar 已提交
5614
		 * pick_next_task() assumes pinned rq->lock:
5615
		 */
5616
		next = pick_next_task(rq, &fake_task, rf);
5617
		BUG_ON(!next);
V
Viresh Kumar 已提交
5618
		put_prev_task(rq, next);
5619

W
Wanpeng Li 已提交
5620 5621 5622 5623 5624 5625 5626 5627 5628
		/*
		 * 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.
		 */
5629
		rq_unlock(rq, rf);
W
Wanpeng Li 已提交
5630
		raw_spin_lock(&next->pi_lock);
5631
		rq_relock(rq, rf);
W
Wanpeng Li 已提交
5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642

		/*
		 * 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;
		}

5643
		/* Find suitable destination for @next, with force if needed. */
5644
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5645
		rq = __migrate_task(rq, rf, next, dest_cpu);
5646
		if (rq != dead_rq) {
5647
			rq_unlock(rq, rf);
5648
			rq = dead_rq;
5649 5650
			*rf = orf;
			rq_relock(rq, rf);
5651
		}
W
Wanpeng Li 已提交
5652
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5653
	}
5654

5655
	rq->stop = stop;
5656
}
L
Linus Torvalds 已提交
5657 5658
#endif /* CONFIG_HOTPLUG_CPU */

5659
void set_rq_online(struct rq *rq)
5660 5661 5662 5663
{
	if (!rq->online) {
		const struct sched_class *class;

5664
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5665 5666 5667 5668 5669 5670 5671 5672 5673
		rq->online = 1;

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

5674
void set_rq_offline(struct rq *rq)
5675 5676 5677 5678 5679 5680 5681 5682 5683
{
	if (rq->online) {
		const struct sched_class *class;

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

5684
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5685 5686 5687 5688
		rq->online = 0;
	}
}

I
Ingo Molnar 已提交
5689 5690 5691 5692
/*
 * used to mark begin/end of suspend/resume:
 */
static int num_cpus_frozen;
5693

L
Linus Torvalds 已提交
5694
/*
5695 5696 5697
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
5698 5699 5700
 *
 * 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
5701
 */
5702
static void cpuset_cpu_active(void)
5703
{
5704
	if (cpuhp_tasks_frozen) {
5705 5706 5707 5708 5709 5710
		/*
		 * 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.
		 */
5711 5712
		partition_sched_domains(1, NULL, NULL);
		if (--num_cpus_frozen)
5713
			return;
5714 5715 5716 5717 5718
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
5719
		cpuset_force_rebuild();
5720
	}
5721
	cpuset_update_active_cpus();
5722
}
5723

5724
static int cpuset_cpu_inactive(unsigned int cpu)
5725
{
5726
	if (!cpuhp_tasks_frozen) {
5727
		if (dl_cpu_busy(cpu))
5728
			return -EBUSY;
5729
		cpuset_update_active_cpus();
5730
	} else {
5731 5732
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
5733
	}
5734
	return 0;
5735 5736
}

5737
int sched_cpu_activate(unsigned int cpu)
5738
{
5739
	struct rq *rq = cpu_rq(cpu);
5740
	struct rq_flags rf;
5741

5742 5743
#ifdef CONFIG_SCHED_SMT
	/*
5744
	 * When going up, increment the number of cores with SMT present.
5745
	 */
5746 5747
	if (cpumask_weight(cpu_smt_mask(cpu)) == 2)
		static_branch_inc_cpuslocked(&sched_smt_present);
5748
#endif
5749
	set_cpu_active(cpu, true);
5750

5751
	if (sched_smp_initialized) {
5752
		sched_domains_numa_masks_set(cpu);
5753
		cpuset_cpu_active();
5754
	}
5755 5756 5757 5758 5759

	/*
	 * 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
5760
	 *    after all CPUs have been brought up.
5761 5762 5763 5764
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
5765
	rq_lock_irqsave(rq, &rf);
5766 5767 5768 5769
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
5770
	rq_unlock_irqrestore(rq, &rf);
5771 5772 5773

	update_max_interval();

5774
	return 0;
5775 5776
}

5777
int sched_cpu_deactivate(unsigned int cpu)
5778 5779 5780
{
	int ret;

5781
	set_cpu_active(cpu, false);
5782 5783 5784 5785 5786 5787 5788
	/*
	 * 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.
	 */
5789
	synchronize_rcu_mult(call_rcu, call_rcu_sched);
5790

5791 5792 5793 5794 5795 5796 5797 5798
#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

5799 5800 5801 5802 5803 5804 5805
	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
5806
	}
5807 5808
	sched_domains_numa_masks_clear(cpu);
	return 0;
5809 5810
}

5811 5812 5813 5814 5815 5816 5817 5818
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();
}

5819 5820
int sched_cpu_starting(unsigned int cpu)
{
5821
	sched_rq_cpu_starting(cpu);
5822
	sched_tick_start(cpu);
5823
	return 0;
5824 5825
}

5826 5827 5828 5829
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5830
	struct rq_flags rf;
5831 5832 5833

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
5834
	sched_tick_stop(cpu);
5835 5836

	rq_lock_irqsave(rq, &rf);
5837 5838 5839 5840
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
5841
	migrate_tasks(rq, &rf);
5842
	BUG_ON(rq->nr_running != 1);
5843 5844
	rq_unlock_irqrestore(rq, &rf);

5845 5846
	calc_load_migrate(rq);
	update_max_interval();
5847
	nohz_balance_exit_idle(rq);
5848
	hrtick_clear(rq);
5849 5850 5851 5852
	return 0;
}
#endif

L
Linus Torvalds 已提交
5853 5854
void __init sched_init_smp(void)
{
5855 5856
	sched_init_numa();

5857 5858
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
I
Ingo Molnar 已提交
5859
	 * CPU masks are stable and all blatant races in the below code cannot
5860 5861
	 * happen. The hotplug lock is nevertheless taken to satisfy lockdep,
	 * but there won't be any contention on it.
5862
	 */
5863
	cpus_read_lock();
5864
	mutex_lock(&sched_domains_mutex);
P
Peter Zijlstra 已提交
5865
	sched_init_domains(cpu_active_mask);
5866
	mutex_unlock(&sched_domains_mutex);
5867
	cpus_read_unlock();
5868

5869
	/* Move init over to a non-isolated CPU */
5870
	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5871
		BUG();
I
Ingo Molnar 已提交
5872
	sched_init_granularity();
5873

5874
	init_sched_rt_class();
5875
	init_sched_dl_class();
P
Peter Zijlstra 已提交
5876

5877
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
5878
}
5879 5880 5881

static int __init migration_init(void)
{
5882
	sched_rq_cpu_starting(smp_processor_id());
5883
	return 0;
L
Linus Torvalds 已提交
5884
}
5885 5886
early_initcall(migration_init);

L
Linus Torvalds 已提交
5887 5888 5889
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
5890
	sched_init_granularity();
L
Linus Torvalds 已提交
5891 5892 5893 5894 5895 5896 5897 5898 5899 5900
}
#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);
}

5901
#ifdef CONFIG_CGROUP_SCHED
5902 5903 5904 5905
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
5906
struct task_group root_task_group;
5907
LIST_HEAD(task_groups);
5908 5909 5910

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
5911
#endif
P
Peter Zijlstra 已提交
5912

5913
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5914
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
5915

L
Linus Torvalds 已提交
5916 5917
void __init sched_init(void)
{
I
Ingo Molnar 已提交
5918
	int i, j;
5919 5920
	unsigned long alloc_size = 0, ptr;

5921
	wait_bit_init();
5922

5923 5924 5925 5926 5927 5928 5929
#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) {
5930
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5931 5932

#ifdef CONFIG_FAIR_GROUP_SCHED
5933
		root_task_group.se = (struct sched_entity **)ptr;
5934 5935
		ptr += nr_cpu_ids * sizeof(void **);

5936
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5937
		ptr += nr_cpu_ids * sizeof(void **);
5938

5939
#endif /* CONFIG_FAIR_GROUP_SCHED */
5940
#ifdef CONFIG_RT_GROUP_SCHED
5941
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5942 5943
		ptr += nr_cpu_ids * sizeof(void **);

5944
		root_task_group.rt_rq = (struct rt_rq **)ptr;
5945 5946
		ptr += nr_cpu_ids * sizeof(void **);

5947
#endif /* CONFIG_RT_GROUP_SCHED */
5948
	}
5949
#ifdef CONFIG_CPUMASK_OFFSTACK
5950 5951 5952
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5953 5954
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5955
	}
5956
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
5957

I
Ingo Molnar 已提交
5958 5959
	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
5960

G
Gregory Haskins 已提交
5961 5962 5963 5964
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

5965
#ifdef CONFIG_RT_GROUP_SCHED
5966
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
5967
			global_rt_period(), global_rt_runtime());
5968
#endif /* CONFIG_RT_GROUP_SCHED */
5969

D
Dhaval Giani 已提交
5970
#ifdef CONFIG_CGROUP_SCHED
5971 5972
	task_group_cache = KMEM_CACHE(task_group, 0);

5973 5974
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
5975
	INIT_LIST_HEAD(&root_task_group.siblings);
5976
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
5977
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
5978

5979
	for_each_possible_cpu(i) {
5980
		struct rq *rq;
L
Linus Torvalds 已提交
5981 5982

		rq = cpu_rq(i);
5983
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
5984
		rq->nr_running = 0;
5985 5986
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
5987
		init_cfs_rq(&rq->cfs);
5988 5989
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
5990
#ifdef CONFIG_FAIR_GROUP_SCHED
5991
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
5992
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
5993
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
5994
		/*
I
Ingo Molnar 已提交
5995
		 * How much CPU bandwidth does root_task_group get?
D
Dhaval Giani 已提交
5996 5997
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
I
Ingo Molnar 已提交
5998 5999
		 * gets 100% of the CPU resources in the system. This overall
		 * system CPU resource is divided among the tasks of
6000
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
6001 6002 6003
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
6004
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
6005
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
I
Ingo Molnar 已提交
6006
		 * then A0's share of the CPU resource is:
D
Dhaval Giani 已提交
6007
		 *
6008
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
6009
		 *
6010 6011
		 * 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
6012
		 */
6013
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6014
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
6015 6016 6017
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6018
#ifdef CONFIG_RT_GROUP_SCHED
6019
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
6020
#endif
L
Linus Torvalds 已提交
6021

I
Ingo Molnar 已提交
6022 6023
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
6024

L
Linus Torvalds 已提交
6025
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6026
		rq->sd = NULL;
G
Gregory Haskins 已提交
6027
		rq->rd = NULL;
6028
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6029
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
6030
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6031
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6032
		rq->push_cpu = 0;
6033
		rq->cpu = i;
6034
		rq->online = 0;
6035 6036
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6037
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6038 6039 6040

		INIT_LIST_HEAD(&rq->cfs_tasks);

6041
		rq_attach_root(rq, &def_root_domain);
6042
#ifdef CONFIG_NO_HZ_COMMON
6043
		rq->last_load_update_tick = jiffies;
6044
		rq->last_blocked_load_update_tick = jiffies;
6045
		atomic_set(&rq->nohz_flags, 0);
6046
#endif
6047
#endif /* CONFIG_SMP */
6048
		hrtick_rq_init(rq);
L
Linus Torvalds 已提交
6049 6050 6051
		atomic_set(&rq->nr_iowait, 0);
	}

6052
	set_load_weight(&init_task, false);
6053

L
Linus Torvalds 已提交
6054 6055 6056
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
V
Vegard Nossum 已提交
6057
	mmgrab(&init_mm);
L
Linus Torvalds 已提交
6058 6059 6060 6061 6062 6063 6064 6065 6066
	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());
6067 6068 6069

	calc_load_update = jiffies + LOAD_FREQ;

6070
#ifdef CONFIG_SMP
6071
	idle_thread_set_boot_cpu();
6072 6073
#endif
	init_sched_fair_class();
6074

6075 6076
	init_schedstats();

6077
	scheduler_running = 1;
L
Linus Torvalds 已提交
6078 6079
}

6080
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6081 6082
static inline int preempt_count_equals(int preempt_offset)
{
6083
	int nested = preempt_count() + rcu_preempt_depth();
6084

A
Arnd Bergmann 已提交
6085
	return (nested == preempt_offset);
6086 6087
}

6088
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6089
{
P
Peter Zijlstra 已提交
6090 6091 6092 6093 6094
	/*
	 * 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.
	 */
6095
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
6096 6097 6098 6099
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
6100
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
6101

6102 6103 6104 6105 6106
	___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
6107
{
I
Ingo Molnar 已提交
6108 6109 6110
	/* Ratelimiting timestamp: */
	static unsigned long prev_jiffy;

6111
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
6112

I
Ingo Molnar 已提交
6113 6114 6115
	/* WARN_ON_ONCE() by default, no rate limit required: */
	rcu_sleep_check();

6116 6117
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
6118 6119
	    system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
	    oops_in_progress)
I
Ingo Molnar 已提交
6120
		return;
6121

I
Ingo Molnar 已提交
6122 6123 6124 6125
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

I
Ingo Molnar 已提交
6126
	/* Save this before calling printk(), since that will clobber it: */
6127 6128
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
6129 6130 6131 6132 6133 6134 6135
	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
6136

6137 6138 6139
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
6140 6141 6142
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
6143 6144
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
6145
		pr_err("Preemption disabled at:");
6146
		print_ip_sym(preempt_disable_ip);
6147 6148
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
6149
	dump_stack();
6150
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
6151
}
6152
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
6153 6154 6155
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6156
void normalize_rt_tasks(void)
6157
{
6158
	struct task_struct *g, *p;
6159 6160 6161
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
6162

6163
	read_lock(&tasklist_lock);
6164
	for_each_process_thread(g, p) {
6165 6166 6167
		/*
		 * Only normalize user tasks:
		 */
6168
		if (p->flags & PF_KTHREAD)
6169 6170
			continue;

6171 6172 6173 6174
		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
6175

6176
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
6177 6178 6179 6180
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
6181
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
6182
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6183
			continue;
I
Ingo Molnar 已提交
6184
		}
L
Linus Torvalds 已提交
6185

6186
		__sched_setscheduler(p, &attr, false, false);
6187
	}
6188
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
6189 6190 6191
}

#endif /* CONFIG_MAGIC_SYSRQ */
6192

6193
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6194
/*
6195
 * These functions are only useful for the IA64 MCA handling, or kdb.
6196 6197 6198 6199 6200 6201 6202 6203 6204
 *
 * 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
6205
 * curr_task - return the current task for a given CPU.
6206 6207 6208
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6209 6210
 *
 * Return: The current task for @cpu.
6211
 */
6212
struct task_struct *curr_task(int cpu)
6213 6214 6215 6216
{
	return cpu_curr(cpu);
}

6217 6218 6219
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6220
/**
I
Ingo Molnar 已提交
6221
 * set_curr_task - set the current task for a given CPU.
6222 6223 6224 6225
 * @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
6226
 * are serviced on a separate stack. It allows the architecture to switch the
I
Ingo Molnar 已提交
6227
 * notion of the current task on a CPU in a non-blocking manner. This function
6228 6229 6230 6231 6232 6233 6234
 * 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!
 */
6235
void ia64_set_curr_task(int cpu, struct task_struct *p)
6236 6237 6238 6239 6240
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6241

D
Dhaval Giani 已提交
6242
#ifdef CONFIG_CGROUP_SCHED
6243 6244 6245
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6246
static void sched_free_group(struct task_group *tg)
6247 6248 6249
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6250
	autogroup_free(tg);
6251
	kmem_cache_free(task_group_cache, tg);
6252 6253 6254
}

/* allocate runqueue etc for a new task group */
6255
struct task_group *sched_create_group(struct task_group *parent)
6256 6257 6258
{
	struct task_group *tg;

6259
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6260 6261 6262
	if (!tg)
		return ERR_PTR(-ENOMEM);

6263
	if (!alloc_fair_sched_group(tg, parent))
6264 6265
		goto err;

6266
	if (!alloc_rt_sched_group(tg, parent))
6267 6268
		goto err;

6269 6270 6271
	return tg;

err:
6272
	sched_free_group(tg);
6273 6274 6275 6276 6277 6278 6279
	return ERR_PTR(-ENOMEM);
}

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

6280
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6281
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6282

I
Ingo Molnar 已提交
6283 6284
	/* Root should already exist: */
	WARN_ON(!parent);
P
Peter Zijlstra 已提交
6285 6286 6287

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6288
	list_add_rcu(&tg->siblings, &parent->children);
6289
	spin_unlock_irqrestore(&task_group_lock, flags);
6290 6291

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
6292 6293
}

6294
/* rcu callback to free various structures associated with a task group */
6295
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6296
{
I
Ingo Molnar 已提交
6297
	/* Now it should be safe to free those cfs_rqs: */
6298
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6299 6300
}

6301
void sched_destroy_group(struct task_group *tg)
6302
{
I
Ingo Molnar 已提交
6303
	/* Wait for possible concurrent references to cfs_rqs complete: */
6304
	call_rcu(&tg->rcu, sched_free_group_rcu);
6305 6306 6307
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6308
{
6309
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
6310

I
Ingo Molnar 已提交
6311
	/* End participation in shares distribution: */
6312
	unregister_fair_sched_group(tg);
6313 6314

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6315
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6316
	list_del_rcu(&tg->siblings);
6317
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6318 6319
}

6320
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
6321
{
P
Peter Zijlstra 已提交
6322
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6323

6324 6325 6326 6327 6328 6329
	/*
	 * 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
6330 6331 6332 6333
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6334
#ifdef CONFIG_FAIR_GROUP_SCHED
6335 6336
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
6337
	else
P
Peter Zijlstra 已提交
6338
#endif
6339
		set_task_rq(tsk, task_cpu(tsk));
6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350
}

/*
 * 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)
{
6351 6352
	int queued, running, queue_flags =
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6353 6354 6355 6356
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);
6357
	update_rq_clock(rq);
6358 6359 6360 6361 6362

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

	if (queued)
6363
		dequeue_task(rq, tsk, queue_flags);
6364
	if (running)
6365 6366 6367
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
6368

6369
	if (queued)
6370
		enqueue_task(rq, tsk, queue_flags);
6371
	if (running)
6372
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6373

6374
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
6375
}
6376

6377
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6378
{
6379
	return css ? container_of(css, struct task_group, css) : NULL;
6380 6381
}

6382 6383
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6384
{
6385 6386
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
6387

6388
	if (!parent) {
6389
		/* This is early initialization for the top cgroup */
6390
		return &root_task_group.css;
6391 6392
	}

6393
	tg = sched_create_group(parent);
6394 6395 6396 6397 6398 6399
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410
/* 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;
}

6411
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6412
{
6413
	struct task_group *tg = css_tg(css);
6414

6415
	sched_offline_group(tg);
6416 6417
}

6418
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6419
{
6420
	struct task_group *tg = css_tg(css);
6421

6422 6423 6424 6425
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
6426 6427
}

6428 6429 6430 6431
/*
 * 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.
 */
6432
static void cpu_cgroup_fork(struct task_struct *task)
6433
{
6434 6435 6436 6437 6438
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

6439
	update_rq_clock(rq);
6440 6441 6442
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
6443 6444
}

6445
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6446
{
6447
	struct task_struct *task;
6448
	struct cgroup_subsys_state *css;
6449
	int ret = 0;
6450

6451
	cgroup_taskset_for_each(task, css, tset) {
6452
#ifdef CONFIG_RT_GROUP_SCHED
6453
		if (!sched_rt_can_attach(css_tg(css), task))
6454
			return -EINVAL;
6455
#else
6456 6457 6458
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
6459
#endif
6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475
		/*
		 * 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;
6476
	}
6477
	return ret;
6478
}
6479

6480
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6481
{
6482
	struct task_struct *task;
6483
	struct cgroup_subsys_state *css;
6484

6485
	cgroup_taskset_for_each(task, css, tset)
6486
		sched_move_task(task);
6487 6488
}

6489
#ifdef CONFIG_FAIR_GROUP_SCHED
6490 6491
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
6492
{
6493
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
6494 6495
}

6496 6497
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
6498
{
6499
	struct task_group *tg = css_tg(css);
6500

6501
	return (u64) scale_load_down(tg->shares);
6502
}
6503 6504

#ifdef CONFIG_CFS_BANDWIDTH
6505 6506
static DEFINE_MUTEX(cfs_constraints_mutex);

6507 6508 6509
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

6510 6511
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

6512 6513
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
6514
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
6515
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535

	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;

6536 6537 6538 6539 6540
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
6541 6542 6543 6544 6545
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

6546
	runtime_enabled = quota != RUNTIME_INF;
6547
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6548 6549 6550 6551 6552 6553
	/*
	 * 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();
6554 6555 6556
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
6557

P
Paul Turner 已提交
6558
	__refill_cfs_bandwidth_runtime(cfs_b);
I
Ingo Molnar 已提交
6559 6560

	/* Restart the period timer (if active) to handle new period expiry: */
P
Peter Zijlstra 已提交
6561 6562
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
I
Ingo Molnar 已提交
6563

6564 6565
	raw_spin_unlock_irq(&cfs_b->lock);

6566
	for_each_online_cpu(i) {
6567
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6568
		struct rq *rq = cfs_rq->rq;
6569
		struct rq_flags rf;
6570

6571
		rq_lock_irq(rq, &rf);
6572
		cfs_rq->runtime_enabled = runtime_enabled;
6573
		cfs_rq->runtime_remaining = 0;
6574

6575
		if (cfs_rq->throttled)
6576
			unthrottle_cfs_rq(cfs_rq);
6577
		rq_unlock_irq(rq, &rf);
6578
	}
6579 6580
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
6581 6582
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
6583
	put_online_cpus();
6584

6585
	return ret;
6586 6587 6588 6589 6590 6591
}

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

6592
	period = ktime_to_ns(tg->cfs_bandwidth.period);
6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604
	if (cfs_quota_us < 0)
		quota = RUNTIME_INF;
	else
		quota = (u64)cfs_quota_us * NSEC_PER_USEC;

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

6605
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6606 6607
		return -1;

6608
	quota_us = tg->cfs_bandwidth.quota;
6609 6610 6611 6612 6613 6614 6615 6616 6617 6618
	do_div(quota_us, NSEC_PER_USEC);

	return quota_us;
}

int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
{
	u64 quota, period;

	period = (u64)cfs_period_us * NSEC_PER_USEC;
6619
	quota = tg->cfs_bandwidth.quota;
6620 6621 6622 6623 6624 6625 6626 6627

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

6628
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6629 6630 6631 6632 6633
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

6634 6635
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
6636
{
6637
	return tg_get_cfs_quota(css_tg(css));
6638 6639
}

6640 6641
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
6642
{
6643
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6644 6645
}

6646 6647
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6648
{
6649
	return tg_get_cfs_period(css_tg(css));
6650 6651
}

6652 6653
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
6654
{
6655
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
6656 6657
}

6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689
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;
6690
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6691 6692 6693 6694 6695
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
6696
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6697 6698

		quota = normalize_cfs_quota(tg, d);
6699
		parent_quota = parent_b->hierarchical_quota;
6700 6701

		/*
6702 6703
		 * Ensure max(child_quota) <= parent_quota.  On cgroup2,
		 * always take the min.  On cgroup1, only inherit when no
I
Ingo Molnar 已提交
6704
		 * limit is set:
6705
		 */
6706 6707 6708 6709 6710 6711 6712 6713
		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;
		}
6714
	}
6715
	cfs_b->hierarchical_quota = quota;
6716 6717 6718 6719 6720 6721

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
6722
	int ret;
6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733
	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);
	}

6734 6735 6736 6737 6738
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6739
}
6740

6741
static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6742
{
6743
	struct task_group *tg = css_tg(seq_css(sf));
6744
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6745

6746 6747 6748
	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);
6749

6750 6751 6752 6753 6754 6755 6756 6757 6758 6759
	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);
	}

6760 6761
	return 0;
}
6762
#endif /* CONFIG_CFS_BANDWIDTH */
6763
#endif /* CONFIG_FAIR_GROUP_SCHED */
6764

6765
#ifdef CONFIG_RT_GROUP_SCHED
6766 6767
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
6768
{
6769
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
6770 6771
}

6772 6773
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
6774
{
6775
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
6776
}
6777

6778 6779
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
6780
{
6781
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
6782 6783
}

6784 6785
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6786
{
6787
	return sched_group_rt_period(css_tg(css));
6788
}
6789
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
6790

6791
static struct cftype cpu_legacy_files[] = {
6792
#ifdef CONFIG_FAIR_GROUP_SCHED
6793 6794
	{
		.name = "shares",
6795 6796
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
6797
	},
6798
#endif
6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809
#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,
	},
6810 6811
	{
		.name = "stat",
6812
		.seq_show = cpu_cfs_stat_show,
6813
	},
6814
#endif
6815
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6816
	{
P
Peter Zijlstra 已提交
6817
		.name = "rt_runtime_us",
6818 6819
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
6820
	},
6821 6822
	{
		.name = "rt_period_us",
6823 6824
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
6825
	},
6826
#endif
I
Ingo Molnar 已提交
6827
	{ }	/* Terminate */
6828 6829
};

6830 6831
static int cpu_extra_stat_show(struct seq_file *sf,
			       struct cgroup_subsys_state *css)
6832 6833 6834
{
#ifdef CONFIG_CFS_BANDWIDTH
	{
6835
		struct task_group *tg = css_tg(css);
6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901
		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;
6902
	int idx;
6903 6904 6905 6906

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

6907 6908 6909 6910
	idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO;
	idx = array_index_nospec(idx, 40);
	weight = sched_prio_to_weight[idx];

6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996
	return sched_group_set_shares(css_tg(css), scale_load(weight));
}
#endif

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

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

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

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

	*periodp *= NSEC_PER_USEC;

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

	return 0;
}

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

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

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

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

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

6997
struct cgroup_subsys cpu_cgrp_subsys = {
6998
	.css_alloc	= cpu_cgroup_css_alloc,
6999
	.css_online	= cpu_cgroup_css_online,
7000
	.css_released	= cpu_cgroup_css_released,
7001
	.css_free	= cpu_cgroup_css_free,
7002
	.css_extra_stat_show = cpu_extra_stat_show,
7003
	.fork		= cpu_cgroup_fork,
7004 7005
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
7006
	.legacy_cftypes	= cpu_legacy_files,
7007
	.dfl_cftypes	= cpu_files,
7008
	.early_init	= true,
7009
	.threaded	= true,
7010 7011
};

7012
#endif	/* CONFIG_CGROUP_SCHED */
7013

7014 7015 7016 7017 7018
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039 7040 7041 7042 7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059

/*
 * 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,
};
7060 7061

#undef CREATE_TRACE_POINTS