core.c 173.5 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 *  kernel/sched/core.c
L
Linus Torvalds 已提交
3
 *
I
Ingo Molnar 已提交
4
 *  Core kernel scheduler code and related syscalls
L
Linus Torvalds 已提交
5 6 7
 *
 *  Copyright (C) 1991-2002  Linus Torvalds
 */
8
#include "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
#define CREATE_TRACE_POINTS
21
#include <trace/events/sched.h>
22

23
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
24

25
#if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL)
I
Ingo Molnar 已提交
26 27
/*
 * Debugging: various feature bits
28 29 30 31
 *
 * 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
32
 */
P
Peter Zijlstra 已提交
33 34
#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |
I
Ingo Molnar 已提交
35
const_debug unsigned int sysctl_sched_features =
36
#include "features.h"
P
Peter Zijlstra 已提交
37 38
	0;
#undef SCHED_FEAT
39
#endif
P
Peter Zijlstra 已提交
40

41 42 43 44 45 46
/*
 * 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;

47 48 49 50 51 52 53 54
/*
 * period over which we average the RT time consumption, measured
 * in ms.
 *
 * default: 1s
 */
const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC;

P
Peter Zijlstra 已提交
55
/*
I
Ingo Molnar 已提交
56
 * period over which we measure -rt task CPU usage in us.
P
Peter Zijlstra 已提交
57 58
 * default: 1s
 */
P
Peter Zijlstra 已提交
59
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
60

61
__read_mostly int scheduler_running;
62

P
Peter Zijlstra 已提交
63 64 65 66 67
/*
 * 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
68

69 70 71
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
72
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
73 74 75 76 77 78 79 80 81 82
	__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))) {
83
			rq_pin_lock(rq, rf);
84 85 86 87 88 89 90 91 92 93 94 95
			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.
 */
96
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
97 98 99 100 101 102
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
103
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
104 105 106 107 108 109 110 111 112 113 114 115
		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)
		 *
116
		 * If we observe the old CPU in task_rq_lock, the acquire of
117 118
		 * the old rq->lock will fully serialize against the stores.
		 *
I
Ingo Molnar 已提交
119
		 * If we observe the new CPU in task_rq_lock, the acquire will
120 121 122
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
123
			rq_pin_lock(rq, rf);
124 125 126
			return rq;
		}
		raw_spin_unlock(&rq->lock);
127
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
128 129 130 131 132 133

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

134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201
/*
 * RQ-clock updating methods:
 */

static void update_rq_clock_task(struct rq *rq, s64 delta)
{
/*
 * In theory, the compile should just see 0 here, and optimize out the call
 * to sched_rt_avg_update. But I don't trust it...
 */
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
	s64 steal = 0, irq_delta = 0;
#endif
#ifdef CONFIG_IRQ_TIME_ACCOUNTING
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;

	/*
	 * Since irq_time is only updated on {soft,}irq_exit, we might run into
	 * this case when a previous update_rq_clock() happened inside a
	 * {soft,}irq region.
	 *
	 * When this happens, we stop ->clock_task and only update the
	 * prev_irq_time stamp to account for the part that fit, so that a next
	 * update will consume the rest. This ensures ->clock_task is
	 * monotonic.
	 *
	 * It does however cause some slight miss-attribution of {soft,}irq
	 * time, a more accurate solution would be to update the irq_time using
	 * the current rq->clock timestamp, except that would require using
	 * atomic ops.
	 */
	if (irq_delta > delta)
		irq_delta = delta;

	rq->prev_irq_time += irq_delta;
	delta -= irq_delta;
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
	if (static_key_false((&paravirt_steal_rq_enabled))) {
		steal = paravirt_steal_clock(cpu_of(rq));
		steal -= rq->prev_steal_time_rq;

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

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

	rq->clock_task += delta;

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

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

	lockdep_assert_held(&rq->lock);

	if (rq->clock_update_flags & RQCF_ACT_SKIP)
		return;

#ifdef CONFIG_SCHED_DEBUG
202 203
	if (sched_feat(WARN_DOUBLE_CLOCK))
		SCHED_WARN_ON(rq->clock_update_flags & RQCF_UPDATED);
204 205
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
206

207 208 209 210 211 212 213 214
	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
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
#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);
233
	struct rq_flags rf;
P
Peter Zijlstra 已提交
234 235 236

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

237
	rq_lock(rq, &rf);
238
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
239
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
240
	rq_unlock(rq, &rf);
P
Peter Zijlstra 已提交
241 242 243 244

	return HRTIMER_NORESTART;
}

245
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
246

247
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
248 249 250
{
	struct hrtimer *timer = &rq->hrtick_timer;

251
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
252 253
}

254 255 256 257
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
258
{
259
	struct rq *rq = arg;
260
	struct rq_flags rf;
261

262
	rq_lock(rq, &rf);
P
Peter Zijlstra 已提交
263
	__hrtick_restart(rq);
264
	rq->hrtick_csd_pending = 0;
265
	rq_unlock(rq, &rf);
266 267
}

268 269 270 271 272
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
273
void hrtick_start(struct rq *rq, u64 delay)
274
{
275
	struct hrtimer *timer = &rq->hrtick_timer;
276 277 278 279 280 281 282 283 284
	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);
285

286
	hrtimer_set_expires(timer, time);
287 288

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
289
		__hrtick_restart(rq);
290
	} else if (!rq->hrtick_csd_pending) {
291
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
292 293
		rq->hrtick_csd_pending = 1;
	}
294 295
}

296 297 298 299 300 301
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
302
void hrtick_start(struct rq *rq, u64 delay)
303
{
W
Wanpeng Li 已提交
304 305 306 307 308
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
309 310
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
311 312
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
313

314
static void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
315
{
316 317
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
318

319 320 321 322
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
323

324 325
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
326
}
A
Andrew Morton 已提交
327
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
328 329 330 331
static inline void hrtick_clear(struct rq *rq)
{
}

332
static inline void hrtick_rq_init(struct rq *rq)
P
Peter Zijlstra 已提交
333 334
{
}
A
Andrew Morton 已提交
335
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
336

337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
/*
 * 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;								\
})

355
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
356 357 358 359 360 361 362 363 364 365
/*
 * 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);
}
366 367 368 369 370 371 372 373 374 375

/*
 * 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);
376
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
377 378 379 380 381 382 383 384 385 386 387 388 389 390

	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;
}

391 392 393 394 395 396
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
397 398 399 400 401 402 403

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
404 405
#endif

406 407 408 409 410 411 412 413 414 415
void wake_q_add(struct wake_q_head *head, struct task_struct *task)
{
	struct wake_q_node *node = &task->wake_q;

	/*
	 * Atomically grab the task, if ->wake_q is !nil already it means
	 * its already queued (either by us or someone else) and will get the
	 * wakeup due to that.
	 *
	 * This cmpxchg() implies a full barrier, which pairs with the write
416
	 * barrier implied by the wakeup in wake_up_q().
417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438
	 */
	if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL))
		return;

	get_task_struct(task);

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

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

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

		task = container_of(node, struct task_struct, wake_q);
		BUG_ON(!task);
I
Ingo Molnar 已提交
439
		/* Task can safely be re-inserted now: */
440 441 442 443 444 445 446 447 448 449 450 451
		node = node->next;
		task->wake_q.next = NULL;

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

I
Ingo Molnar 已提交
452
/*
453
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
454 455 456 457 458
 *
 * 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.
 */
459
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
460
{
461
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
462 463
	int cpu;

464
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
465

466
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
467 468
		return;

469
	cpu = cpu_of(rq);
470

471
	if (cpu == smp_processor_id()) {
472
		set_tsk_need_resched(curr);
473
		set_preempt_need_resched();
I
Ingo Molnar 已提交
474
		return;
475
	}
I
Ingo Molnar 已提交
476

477
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
478
		smp_send_reschedule(cpu);
479 480
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
481 482
}

483
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
484 485 486 487
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

488
	raw_spin_lock_irqsave(&rq->lock, flags);
489 490
	if (cpu_online(cpu) || cpu == smp_processor_id())
		resched_curr(rq);
491
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
492
}
493

494
#ifdef CONFIG_SMP
495
#ifdef CONFIG_NO_HZ_COMMON
496
/*
I
Ingo Molnar 已提交
497 498
 * In the semi idle case, use the nearest busy CPU for migrating timers
 * from an idle CPU.  This is good for power-savings.
499 500
 *
 * We don't do similar optimization for completely idle system, as
I
Ingo Molnar 已提交
501 502
 * 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).
503
 */
504
int get_nohz_timer_target(void)
505
{
506
	int i, cpu = smp_processor_id();
507 508
	struct sched_domain *sd;

509
	if (!idle_cpu(cpu) && housekeeping_cpu(cpu, HK_FLAG_TIMER))
510 511
		return cpu;

512
	rcu_read_lock();
513
	for_each_domain(cpu, sd) {
514
		for_each_cpu(i, sched_domain_span(sd)) {
515 516 517
			if (cpu == i)
				continue;

518
			if (!idle_cpu(i) && housekeeping_cpu(i, HK_FLAG_TIMER)) {
519 520 521 522
				cpu = i;
				goto unlock;
			}
		}
523
	}
524

525 526
	if (!housekeeping_cpu(cpu, HK_FLAG_TIMER))
		cpu = housekeeping_any_cpu(HK_FLAG_TIMER);
527 528
unlock:
	rcu_read_unlock();
529 530
	return cpu;
}
I
Ingo Molnar 已提交
531

532 533 534 535 536 537 538 539 540 541
/*
 * 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.
 */
542
static void wake_up_idle_cpu(int cpu)
543 544 545 546 547 548
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

549
	if (set_nr_and_not_polling(rq->idle))
550
		smp_send_reschedule(cpu);
551 552
	else
		trace_sched_wake_idle_without_ipi(cpu);
553 554
}

555
static bool wake_up_full_nohz_cpu(int cpu)
556
{
557 558 559 560 561 562
	/*
	 * 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.
	 */
563 564
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
565
	if (tick_nohz_full_cpu(cpu)) {
566 567
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
568
			tick_nohz_full_kick_cpu(cpu);
569 570 571 572 573 574
		return true;
	}

	return false;
}

575 576 577 578 579
/*
 * 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.
 */
580 581
void wake_up_nohz_cpu(int cpu)
{
582
	if (!wake_up_full_nohz_cpu(cpu))
583 584 585
		wake_up_idle_cpu(cpu);
}

586
static inline bool got_nohz_idle_kick(void)
587
{
588
	int cpu = smp_processor_id();
589

P
Peter Zijlstra 已提交
590
	if (!(atomic_read(nohz_flags(cpu)) & NOHZ_KICK_MASK))
591 592 593 594 595 596 597 598 599
		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
600
	atomic_andnot(NOHZ_KICK_MASK, nohz_flags(cpu));
601
	return false;
602 603
}

604
#else /* CONFIG_NO_HZ_COMMON */
605

606
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
607
{
608
	return false;
P
Peter Zijlstra 已提交
609 610
}

611
#endif /* CONFIG_NO_HZ_COMMON */
612

613
#ifdef CONFIG_NO_HZ_FULL
614
bool sched_can_stop_tick(struct rq *rq)
615
{
616 617 618 619 620 621
	int fifo_nr_running;

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

622
	/*
623 624
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
625
	 */
626 627 628 629 630
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
631 632
	}

633 634 635 636 637 638 639 640 641 642 643 644 645 646
	/*
	 * 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)
647
		return false;
648

649
	return true;
650 651
}
#endif /* CONFIG_NO_HZ_FULL */
652

653
void sched_avg_update(struct rq *rq)
654
{
655 656
	s64 period = sched_avg_period();

657
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
658 659 660 661 662 663
		/*
		 * Inline assembly required to prevent the compiler
		 * optimising this loop into a divmod call.
		 * See __iter_div_u64_rem() for another example of this.
		 */
		asm("" : "+rm" (rq->age_stamp));
664 665 666
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
667 668
}

669
#endif /* CONFIG_SMP */
670

671 672
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
673
/*
674 675 676 677
 * 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.
678
 */
679
int walk_tg_tree_from(struct task_group *from,
680
			     tg_visitor down, tg_visitor up, void *data)
681 682
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
683
	int ret;
684

685 686
	parent = from;

687
down:
P
Peter Zijlstra 已提交
688 689
	ret = (*down)(parent, data);
	if (ret)
690
		goto out;
691 692 693 694 695 696 697
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
698
	ret = (*up)(parent, data);
699 700
	if (ret || parent == from)
		goto out;
701 702 703 704 705

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
706
out:
P
Peter Zijlstra 已提交
707
	return ret;
708 709
}

710
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
711
{
712
	return 0;
P
Peter Zijlstra 已提交
713
}
714 715
#endif

716
static void set_load_weight(struct task_struct *p, bool update_load)
717
{
N
Nikhil Rao 已提交
718 719 720
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
721 722 723
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
724
	if (idle_policy(p->policy)) {
725
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
726
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
727 728
		return;
	}
729

730 731 732 733 734 735 736 737 738 739
	/*
	 * 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];
	}
740 741
}

742
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
743
{
744 745 746
	if (!(flags & ENQUEUE_NOCLOCK))
		update_rq_clock(rq);

747 748
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
749

750
	p->sched_class->enqueue_task(rq, p, flags);
751 752
}

753
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
754
{
755 756 757
	if (!(flags & DEQUEUE_NOCLOCK))
		update_rq_clock(rq);

758 759
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
760

761
	p->sched_class->dequeue_task(rq, p, flags);
762 763
}

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

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

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

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

780
/*
I
Ingo Molnar 已提交
781
 * __normal_prio - return the priority that is based on the static prio
782 783 784
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
785
	return p->static_prio;
786 787
}

788 789 790 791 792 793 794
/*
 * 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.
 */
795
static inline int normal_prio(struct task_struct *p)
796 797 798
{
	int prio;

799 800 801
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
802 803 804 805 806 807 808 809 810 811 812 813 814
		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.
 */
815
static int effective_prio(struct task_struct *p)
816 817 818 819 820 821 822 823 824 825 826 827
{
	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
828 829 830
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
831 832
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
833
 */
834
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
835 836 837 838
{
	return cpu_curr(task_cpu(p)) == p;
}

839
/*
840 841 842 843 844
 * 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().
845
 */
846 847
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
848
				       int oldprio)
849 850 851
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
852
			prev_class->switched_from(rq, p);
853

P
Peter Zijlstra 已提交
854
		p->sched_class->switched_to(rq, p);
855
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
856
		p->sched_class->prio_changed(rq, p, oldprio);
857 858
}

859
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
860 861 862 863 864 865 866 867 868 869
{
	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) {
870
				resched_curr(rq);
871 872 873 874 875 876 877 878 879
				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.
	 */
880
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
881
		rq_clock_skip_update(rq);
882 883
}

L
Linus Torvalds 已提交
884
#ifdef CONFIG_SMP
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911

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
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930
/*
 * 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.
 */
931 932
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
933 934 935 936
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
937
	dequeue_task(rq, p, DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
938
	set_task_cpu(p, new_cpu);
939
	rq_unlock(rq, rf);
P
Peter Zijlstra 已提交
940 941 942

	rq = cpu_rq(new_cpu);

943
	rq_lock(rq, rf);
P
Peter Zijlstra 已提交
944 945
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
946
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
947 948 949 950 951 952 953 954 955 956 957
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
I
Ingo Molnar 已提交
958
 * Move (not current) task off this CPU, onto the destination CPU. We're doing
P
Peter Zijlstra 已提交
959 960 961 962 963 964 965
 * 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.
 */
966 967
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
968 969
{
	/* Affinity changed (again). */
970
	if (!is_cpu_allowed(p, dest_cpu))
971
		return rq;
P
Peter Zijlstra 已提交
972

973
	update_rq_clock(rq);
974
	rq = move_queued_task(rq, rf, p, dest_cpu);
975 976

	return rq;
P
Peter Zijlstra 已提交
977 978 979 980 981 982 983 984 985 986
}

/*
 * 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;
987 988
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
989
	struct rq_flags rf;
P
Peter Zijlstra 已提交
990 991

	/*
I
Ingo Molnar 已提交
992 993
	 * 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
994 995 996 997 998 999 1000 1001
	 */
	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();
1002 1003

	raw_spin_lock(&p->pi_lock);
1004
	rq_lock(rq, &rf);
1005 1006 1007 1008 1009
	/*
	 * 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.
	 */
1010 1011
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
1012
			rq = __migrate_task(rq, &rf, p, arg->dest_cpu);
1013 1014 1015
		else
			p->wake_cpu = arg->dest_cpu;
	}
1016
	rq_unlock(rq, &rf);
1017 1018
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1019 1020 1021 1022
	local_irq_enable();
	return 0;
}

1023 1024 1025 1026 1027
/*
 * 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
1028 1029 1030 1031 1032
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1033 1034
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1035 1036 1037
	struct rq *rq = task_rq(p);
	bool queued, running;

1038
	lockdep_assert_held(&p->pi_lock);
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048

	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);
1049
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
1050 1051 1052 1053
	}
	if (running)
		put_prev_task(rq, p);

1054
	p->sched_class->set_cpus_allowed(p, new_mask);
1055 1056

	if (queued)
1057
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
1058
	if (running)
1059
		set_curr_task(rq, p);
1060 1061
}

P
Peter Zijlstra 已提交
1062 1063 1064 1065 1066 1067 1068 1069 1070
/*
 * 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.
 */
1071 1072
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1073
{
1074
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1075
	unsigned int dest_cpu;
1076 1077
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1078 1079
	int ret = 0;

1080
	rq = task_rq_lock(p, &rf);
1081
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
1082

1083 1084 1085 1086 1087 1088 1089
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1090 1091 1092 1093 1094 1095 1096 1097 1098
	/*
	 * 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
1099 1100 1101
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1102
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1103 1104 1105 1106 1107 1108
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1109 1110 1111
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
I
Ingo Molnar 已提交
1112
		 * !active we want to ensure they are strict per-CPU threads.
1113 1114 1115 1116 1117 1118
		 */
		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
1119 1120 1121 1122
	/* 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;

1123
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1124 1125 1126
	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. */
1127
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1128 1129 1130
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1131 1132 1133 1134 1135
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1136
		rq = move_queued_task(rq, &rf, p, dest_cpu);
1137
	}
P
Peter Zijlstra 已提交
1138
out:
1139
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1140 1141 1142

	return ret;
}
1143 1144 1145 1146 1147

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
1148 1149
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1150
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1151
{
1152 1153 1154 1155 1156
#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
1157
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1158
			!p->on_rq);
1159

1160 1161 1162 1163 1164 1165 1166 1167 1168
	/*
	 * 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)));

1169
#ifdef CONFIG_LOCKDEP
1170 1171 1172 1173 1174
	/*
	 * 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
1175
	 * see task_group().
1176 1177 1178 1179
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1180 1181 1182
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1183 1184 1185 1186
	/*
	 * Clearly, migrating tasks to offline CPUs is a fairly daft thing.
	 */
	WARN_ON_ONCE(!cpu_online(new_cpu));
1187 1188
#endif

1189
	trace_sched_migrate_task(p, new_cpu);
1190

1191
	if (task_cpu(p) != new_cpu) {
1192
		if (p->sched_class->migrate_task_rq)
1193
			p->sched_class->migrate_task_rq(p);
1194
		p->se.nr_migrations++;
1195
		rseq_migrate(p);
1196
		perf_event_task_migrate(p);
1197
	}
I
Ingo Molnar 已提交
1198 1199

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1200 1201
}

1202 1203
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1204
	if (task_on_rq_queued(p)) {
1205
		struct rq *src_rq, *dst_rq;
1206
		struct rq_flags srf, drf;
1207 1208 1209 1210

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

1211 1212 1213
		rq_pin_lock(src_rq, &srf);
		rq_pin_lock(dst_rq, &drf);

1214
		p->on_rq = TASK_ON_RQ_MIGRATING;
1215 1216 1217
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1218
		p->on_rq = TASK_ON_RQ_QUEUED;
1219
		check_preempt_curr(dst_rq, p, 0);
1220 1221 1222 1223

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

1224 1225 1226 1227
	} 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
1228
		 * previous CPU our target instead of where it really is.
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
		 */
		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;

1245 1246 1247
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1248 1249 1250
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1251 1252
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1253
	double_rq_lock(src_rq, dst_rq);
1254

1255 1256 1257 1258 1259 1260
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

1261
	if (!cpumask_test_cpu(arg->dst_cpu, &arg->src_task->cpus_allowed))
1262 1263
		goto unlock;

1264
	if (!cpumask_test_cpu(arg->src_cpu, &arg->dst_task->cpus_allowed))
1265 1266 1267 1268 1269 1270 1271 1272 1273
		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);
1274 1275
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297

	return ret;
}

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

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

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

1298 1299 1300 1301
	/*
	 * 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.
	 */
1302 1303 1304
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

1305
	if (!cpumask_test_cpu(arg.dst_cpu, &arg.src_task->cpus_allowed))
1306 1307
		goto out;

1308
	if (!cpumask_test_cpu(arg.src_cpu, &arg.dst_task->cpus_allowed))
1309 1310
		goto out;

1311
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1312 1313 1314 1315 1316 1317
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1318 1319 1320
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1321 1322 1323 1324 1325 1326 1327
 * 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
1328 1329 1330 1331 1332 1333
 * 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
1334
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1335
{
1336
	int running, queued;
1337
	struct rq_flags rf;
R
Roland McGrath 已提交
1338
	unsigned long ncsw;
1339
	struct rq *rq;
L
Linus Torvalds 已提交
1340

1341 1342 1343 1344 1345 1346 1347 1348
	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);
1349

1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360
		/*
		 * 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
1361 1362 1363
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1364
			cpu_relax();
R
Roland McGrath 已提交
1365
		}
1366

1367 1368 1369 1370 1371
		/*
		 * 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.
		 */
1372
		rq = task_rq_lock(p, &rf);
1373
		trace_sched_wait_task(p);
1374
		running = task_running(rq, p);
1375
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1376
		ncsw = 0;
1377
		if (!match_state || p->state == match_state)
1378
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1379
		task_rq_unlock(rq, p, &rf);
1380

R
Roland McGrath 已提交
1381 1382 1383 1384 1385 1386
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1387 1388 1389 1390 1391 1392 1393 1394 1395 1396
		/*
		 * 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;
		}
1397

1398 1399 1400 1401 1402
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1403
		 * So if it was still runnable (but just not actively
1404 1405 1406
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1407
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1408
			ktime_t to = NSEC_PER_SEC / HZ;
1409 1410 1411

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1412 1413
			continue;
		}
1414

1415 1416 1417 1418 1419 1420 1421
		/*
		 * 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
1422 1423

	return ncsw;
L
Linus Torvalds 已提交
1424 1425 1426 1427 1428 1429 1430 1431 1432
}

/***
 * 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
1433
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1434 1435 1436 1437 1438
 * 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.
 */
1439
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1440 1441 1442 1443 1444 1445 1446 1447 1448
{
	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
1449
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1450

1451
/*
1452
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1453 1454 1455 1456 1457
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
1458
 *  - on CPU-up we allow per-CPU kthreads on the online && !active CPU,
1459
 *    see __set_cpus_allowed_ptr(). At this point the newly online
I
Ingo Molnar 已提交
1460
 *    CPU isn't yet part of the sched domains, and balancing will not
1461 1462
 *    see it.
 *
I
Ingo Molnar 已提交
1463
 *  - on CPU-down we clear cpu_active() to mask the sched domains and
1464
 *    avoid the load balancer to place new tasks on the to be removed
I
Ingo Molnar 已提交
1465
 *    CPU. Existing tasks will remain running there and will be taken
1466 1467 1468 1469 1470 1471
 *    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.
1472
 */
1473 1474
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1475 1476
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1477 1478
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1479

1480
	/*
I
Ingo Molnar 已提交
1481 1482 1483
	 * 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.
1484 1485 1486 1487 1488 1489 1490 1491
	 */
	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;
1492
			if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed))
1493 1494
				return dest_cpu;
		}
1495
	}
1496

1497 1498
	for (;;) {
		/* Any allowed, online CPU? */
1499
		for_each_cpu(dest_cpu, &p->cpus_allowed) {
1500
			if (!is_cpu_allowed(p, dest_cpu))
1501
				continue;
1502

1503 1504
			goto out;
		}
1505

1506
		/* No more Mr. Nice Guy. */
1507 1508
		switch (state) {
		case cpuset:
1509 1510 1511 1512 1513
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
I
Ingo Molnar 已提交
1514
			/* Fall-through */
1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
		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()) {
1534
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1535 1536
					task_pid_nr(p), p->comm, cpu);
		}
1537 1538 1539 1540 1541
	}

	return dest_cpu;
}

1542
/*
1543
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1544
 */
1545
static inline
1546
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1547
{
1548 1549
	lockdep_assert_held(&p->pi_lock);

1550
	if (p->nr_cpus_allowed > 1)
1551
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1552
	else
1553
		cpu = cpumask_any(&p->cpus_allowed);
1554 1555 1556 1557

	/*
	 * 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
1558
	 * CPU.
1559 1560 1561 1562 1563 1564
	 *
	 * 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 ]
	 */
1565
	if (unlikely(!is_cpu_allowed(p, cpu)))
1566
		cpu = select_fallback_rq(task_cpu(p), p);
1567 1568

	return cpu;
1569
}
1570 1571 1572 1573 1574 1575

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1576

1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606
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;
	}
}

1607 1608 1609 1610 1611 1612 1613 1614
#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
1615
#endif /* CONFIG_SMP */
1616

P
Peter Zijlstra 已提交
1617
static void
1618
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1619
{
1620
	struct rq *rq;
1621

1622 1623 1624 1625
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1626

1627 1628
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1629 1630
		__schedstat_inc(rq->ttwu_local);
		__schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1631 1632 1633
	} else {
		struct sched_domain *sd;

1634
		__schedstat_inc(p->se.statistics.nr_wakeups_remote);
1635
		rcu_read_lock();
1636
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1637
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1638
				__schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1639 1640 1641
				break;
			}
		}
1642
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1643
	}
1644 1645

	if (wake_flags & WF_MIGRATED)
1646
		__schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1647 1648
#endif /* CONFIG_SMP */

1649 1650
	__schedstat_inc(rq->ttwu_count);
	__schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1651 1652

	if (wake_flags & WF_SYNC)
1653
		__schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1654 1655
}

1656
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1657
{
T
Tejun Heo 已提交
1658
	activate_task(rq, p, en_flags);
1659
	p->on_rq = TASK_ON_RQ_QUEUED;
1660

I
Ingo Molnar 已提交
1661
	/* If a worker is waking up, notify the workqueue: */
1662 1663
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1664 1665
}

1666 1667 1668
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1669
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1670
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1671 1672 1673
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1674 1675
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1676
#ifdef CONFIG_SMP
1677 1678
	if (p->sched_class->task_woken) {
		/*
1679 1680
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1681
		 */
1682
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1683
		p->sched_class->task_woken(rq, p);
1684
		rq_repin_lock(rq, rf);
1685
	}
T
Tejun Heo 已提交
1686

1687
	if (rq->idle_stamp) {
1688
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1689
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1690

1691 1692 1693
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1694
			rq->avg_idle = max;
1695

T
Tejun Heo 已提交
1696 1697 1698 1699 1700
		rq->idle_stamp = 0;
	}
#endif
}

1701
static void
1702
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1703
		 struct rq_flags *rf)
1704
{
1705
	int en_flags = ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK;
1706

1707 1708
	lockdep_assert_held(&rq->lock);

1709 1710 1711
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1712 1713

	if (wake_flags & WF_MIGRATED)
1714
		en_flags |= ENQUEUE_MIGRATED;
1715 1716
#endif

1717
	ttwu_activate(rq, p, en_flags);
1718
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1719 1720 1721 1722 1723 1724 1725 1726 1727 1728
}

/*
 * 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)
{
1729
	struct rq_flags rf;
1730 1731 1732
	struct rq *rq;
	int ret = 0;

1733
	rq = __task_rq_lock(p, &rf);
1734
	if (task_on_rq_queued(p)) {
1735 1736
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1737
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1738 1739
		ret = 1;
	}
1740
	__task_rq_unlock(rq, &rf);
1741 1742 1743 1744

	return ret;
}

1745
#ifdef CONFIG_SMP
1746
void sched_ttwu_pending(void)
1747 1748
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1749
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1750
	struct task_struct *p, *t;
1751
	struct rq_flags rf;
1752

1753 1754 1755
	if (!llist)
		return;

1756
	rq_lock_irqsave(rq, &rf);
1757
	update_rq_clock(rq);
1758

1759 1760
	llist_for_each_entry_safe(p, t, llist, wake_entry)
		ttwu_do_activate(rq, p, p->sched_remote_wakeup ? WF_MIGRATED : 0, &rf);
1761

1762
	rq_unlock_irqrestore(rq, &rf);
1763 1764 1765 1766
}

void scheduler_ipi(void)
{
1767 1768 1769 1770 1771
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1772
	preempt_fold_need_resched();
1773

1774
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790
		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
1791
	sched_ttwu_pending();
1792 1793 1794 1795

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1796
	if (unlikely(got_nohz_idle_kick())) {
1797
		this_rq()->idle_balance = 1;
1798
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1799
	}
1800
	irq_exit();
1801 1802
}

P
Peter Zijlstra 已提交
1803
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1804
{
1805 1806
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1807 1808
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1809 1810 1811 1812 1813 1814
	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);
	}
1815
}
1816

1817 1818 1819
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
1820
	struct rq_flags rf;
1821

1822 1823 1824 1825
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1826 1827 1828 1829

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
1830
		rq_lock_irqsave(rq, &rf);
1831 1832
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
I
Ingo Molnar 已提交
1833
		/* Else CPU is not idle, do nothing here: */
1834
		rq_unlock_irqrestore(rq, &rf);
1835
	}
1836 1837 1838

out:
	rcu_read_unlock();
1839 1840
}

1841
bool cpus_share_cache(int this_cpu, int that_cpu)
1842 1843 1844
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1845
#endif /* CONFIG_SMP */
1846

1847
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1848 1849
{
	struct rq *rq = cpu_rq(cpu);
1850
	struct rq_flags rf;
1851

1852
#if defined(CONFIG_SMP)
1853
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
I
Ingo Molnar 已提交
1854
		sched_clock_cpu(cpu); /* Sync clocks across CPUs */
P
Peter Zijlstra 已提交
1855
		ttwu_queue_remote(p, cpu, wake_flags);
1856 1857 1858 1859
		return;
	}
#endif

1860
	rq_lock(rq, &rf);
1861
	update_rq_clock(rq);
1862
	ttwu_do_activate(rq, p, wake_flags, &rf);
1863
	rq_unlock(rq, &rf);
T
Tejun Heo 已提交
1864 1865
}

1866 1867 1868 1869 1870 1871
/*
 * 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
1872 1873
 * migrates, all its activity on its old CPU [c0] happens-before any subsequent
 * execution on its new CPU [c1].
1874 1875 1876 1877 1878 1879 1880 1881 1882 1883
 *
 * For migration (of runnable tasks) this is provided by the following means:
 *
 *  A) UNLOCK of the rq(c0)->lock scheduling out task t
 *  B) migration for t is required to synchronize *both* rq(c0)->lock and
 *     rq(c1)->lock (if not at the same time, then in that order).
 *  C) LOCK of the rq(c1)->lock scheduling in task
 *
 * Transitivity guarantees that B happens after A and C after B.
 * Note: we only require RCpc transitivity.
I
Ingo Molnar 已提交
1884
 * Note: the CPU doing B need not be c0 or c1
1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915
 *
 * 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)
1916
 *   2) smp_cond_load_acquire(!X->on_cpu)
1917 1918 1919 1920 1921 1922 1923 1924 1925 1926
 *
 * 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);
 *
1927
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952
 *                    X->state = WAKING
 *                    set_task_cpu(X,2)
 *
 *                    LOCK rq(2)->lock
 *                    enqueue X
 *                    X->state = RUNNING
 *                    UNLOCK rq(2)->lock
 *
 *                                          LOCK rq(2)->lock // orders against CPU1
 *                                          sched-out Z
 *                                          sched-in X
 *                                          UNLOCK rq(2)->lock
 *
 *                    UNLOCK X->pi_lock
 *   UNLOCK rq(0)->lock
 *
 *
 * However; for wakeups there is a second guarantee we must provide, namely we
 * must observe the state that lead to our wakeup. That is, not only must our
 * task observe its own prior state, it must also observe the stores prior to
 * its wakeup.
 *
 * This means that any means of doing remote wakeups must order the CPU doing
 * the wakeup against the CPU the task is going to end up running on. This,
 * however, is already required for the regular Program-Order guarantee above,
1953
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1954 1955 1956
 *
 */

T
Tejun Heo 已提交
1957
/**
L
Linus Torvalds 已提交
1958
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1959
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1960
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1961
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1962
 *
1963
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1964
 *
1965 1966 1967 1968 1969 1970 1971
 * If the task was not queued/runnable, also place it back on a runqueue.
 *
 * Atomic against schedule() which would dequeue a task, also see
 * set_current_state().
 *
 * Return: %true if @p->state changes (an actual wakeup was done),
 *	   %false otherwise.
L
Linus Torvalds 已提交
1972
 */
1973 1974
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1975 1976
{
	unsigned long flags;
1977
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1978

1979 1980 1981 1982 1983 1984
	/*
	 * 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.
	 */
1985
	raw_spin_lock_irqsave(&p->pi_lock, flags);
1986
	smp_mb__after_spinlock();
P
Peter Zijlstra 已提交
1987
	if (!(p->state & state))
L
Linus Torvalds 已提交
1988 1989
		goto out;

1990 1991
	trace_sched_waking(p);

I
Ingo Molnar 已提交
1992 1993
	/* We're going to change ->state: */
	success = 1;
L
Linus Torvalds 已提交
1994 1995
	cpu = task_cpu(p);

1996 1997 1998 1999 2000
	/*
	 * 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.
	 *
2001 2002 2003 2004 2005 2006 2007 2008
	 * 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
2009 2010
	 *
	 * [task p]
2011
	 *   STORE p->state = UNINTERRUPTIBLE	  LOAD p->on_rq
2012
	 *
2013 2014
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
2015 2016
	 */
	smp_rmb();
2017 2018
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2019 2020

#ifdef CONFIG_SMP
2021 2022 2023 2024 2025 2026 2027
	/*
	 * 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.
	 *
2028 2029 2030 2031 2032 2033 2034 2035
	 * __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
2036
	 *
2037 2038
	 * Pairs with the LOCK+smp_mb__after_spinlock() on rq->lock in
	 * __schedule().  See the comment for smp_mb__after_spinlock().
2039 2040 2041
	 */
	smp_rmb();

P
Peter Zijlstra 已提交
2042
	/*
I
Ingo Molnar 已提交
2043
	 * If the owning (remote) CPU is still in the middle of schedule() with
2044
	 * this task as prev, wait until its done referencing the task.
2045
	 *
2046
	 * Pairs with the smp_store_release() in finish_task().
2047 2048 2049
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2050
	 */
2051
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2052

2053
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2054
	p->state = TASK_WAKING;
2055

2056
	if (p->in_iowait) {
2057
		delayacct_blkio_end(p);
2058 2059 2060
		atomic_dec(&task_rq(p)->nr_iowait);
	}

2061
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2062 2063
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2064
		set_task_cpu(p, cpu);
2065
	}
2066 2067 2068 2069

#else /* CONFIG_SMP */

	if (p->in_iowait) {
2070
		delayacct_blkio_end(p);
2071 2072 2073
		atomic_dec(&task_rq(p)->nr_iowait);
	}

L
Linus Torvalds 已提交
2074 2075
#endif /* CONFIG_SMP */

2076
	ttwu_queue(p, cpu, wake_flags);
2077
stat:
2078
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2079
out:
2080
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2081 2082 2083 2084

	return success;
}

T
Tejun Heo 已提交
2085 2086 2087
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2088
 * @rf: request-queue flags for pinning
T
Tejun Heo 已提交
2089
 *
2090
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2091
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2092
 * the current task.
T
Tejun Heo 已提交
2093
 */
2094
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2095 2096 2097
{
	struct rq *rq = task_rq(p);

2098 2099 2100 2101
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2102 2103
	lockdep_assert_held(&rq->lock);

2104
	if (!raw_spin_trylock(&p->pi_lock)) {
2105 2106 2107 2108 2109 2110
		/*
		 * 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.
		 */
2111
		rq_unlock(rq, rf);
2112
		raw_spin_lock(&p->pi_lock);
2113
		rq_relock(rq, rf);
2114 2115
	}

T
Tejun Heo 已提交
2116
	if (!(p->state & TASK_NORMAL))
2117
		goto out;
T
Tejun Heo 已提交
2118

2119 2120
	trace_sched_waking(p);

2121 2122
	if (!task_on_rq_queued(p)) {
		if (p->in_iowait) {
2123
			delayacct_blkio_end(p);
2124 2125
			atomic_dec(&rq->nr_iowait);
		}
2126
		ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_NOCLOCK);
2127
	}
P
Peter Zijlstra 已提交
2128

2129
	ttwu_do_wakeup(rq, p, 0, rf);
2130
	ttwu_stat(p, smp_processor_id(), 0);
2131 2132
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2133 2134
}

2135 2136 2137 2138 2139
/**
 * 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
2140 2141 2142
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2143 2144 2145 2146
 *
 * It may be assumed that this function implies a write memory barrier before
 * changing the task state if and only if any tasks are woken up.
 */
2147
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2148
{
2149
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2150 2151 2152
}
EXPORT_SYMBOL(wake_up_process);

2153
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2154 2155 2156 2157 2158 2159 2160
{
	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
2161 2162 2163
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2164
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2165
{
P
Peter Zijlstra 已提交
2166 2167 2168
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2169 2170
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2171
	p->se.prev_sum_exec_runtime	= 0;
2172
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2173
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2174
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2175

2176 2177 2178 2179
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2180
#ifdef CONFIG_SCHEDSTATS
2181
	/* Even if schedstat is disabled, there should not be garbage */
2182
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2183
#endif
N
Nick Piggin 已提交
2184

2185
	RB_CLEAR_NODE(&p->dl.rb_node);
2186
	init_dl_task_timer(&p->dl);
2187
	init_dl_inactive_task_timer(&p->dl);
2188
	__dl_clear_params(p);
2189

P
Peter Zijlstra 已提交
2190
	INIT_LIST_HEAD(&p->rt.run_list);
2191 2192 2193 2194
	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
2195

2196 2197 2198
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2199

2200
	init_numa_balancing(clone_flags, p);
I
Ingo Molnar 已提交
2201 2202
}

2203 2204
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2205
#ifdef CONFIG_NUMA_BALANCING
2206

2207 2208 2209
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2210
		static_branch_enable(&sched_numa_balancing);
2211
	else
2212
		static_branch_disable(&sched_numa_balancing);
2213
}
2214 2215 2216 2217 2218 2219 2220

#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;
2221
	int state = static_branch_likely(&sched_numa_balancing);
2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236

	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
2237

2238 2239
#ifdef CONFIG_SCHEDSTATS

2240
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2241
static bool __initdata __sched_schedstats = false;
2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264

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;

2265 2266 2267 2268 2269
	/*
	 * 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.
	 */
2270
	if (!strcmp(str, "enable")) {
2271
		__sched_schedstats = true;
2272 2273
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2274
		__sched_schedstats = false;
2275 2276 2277 2278 2279 2280 2281 2282 2283 2284
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2285 2286 2287 2288 2289
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309
#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;
}
2310 2311 2312 2313
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2314 2315 2316 2317

/*
 * fork()/clone()-time setup:
 */
2318
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2319
{
2320
	unsigned long flags;
I
Ingo Molnar 已提交
2321 2322
	int cpu = get_cpu();

2323
	__sched_fork(clone_flags, p);
2324
	/*
2325
	 * We mark the process as NEW here. This guarantees that
2326 2327 2328
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2329
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2330

2331 2332 2333 2334 2335
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2336 2337 2338 2339
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2340
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2341
			p->policy = SCHED_NORMAL;
2342
			p->static_prio = NICE_TO_PRIO(0);
2343 2344 2345 2346 2347
			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);
2348
		set_load_weight(p, false);
2349

2350 2351 2352 2353 2354 2355
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2356

2357 2358 2359 2360 2361 2362
	if (dl_prio(p->prio)) {
		put_cpu();
		return -EAGAIN;
	} else if (rt_prio(p->prio)) {
		p->sched_class = &rt_sched_class;
	} else {
H
Hiroshi Shimamoto 已提交
2363
		p->sched_class = &fair_sched_class;
2364
	}
2365

2366
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2367

2368 2369 2370 2371 2372 2373 2374
	/*
	 * 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.
	 */
2375
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2376
	/*
I
Ingo Molnar 已提交
2377
	 * We're setting the CPU for the first time, we don't migrate,
2378 2379 2380 2381 2382
	 * so use __set_task_cpu().
	 */
	__set_task_cpu(p, cpu);
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2383
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2384

2385
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2386
	if (likely(sched_info_on()))
2387
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2388
#endif
P
Peter Zijlstra 已提交
2389 2390
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2391
#endif
2392
	init_task_preempt_count(p);
2393
#ifdef CONFIG_SMP
2394
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2395
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2396
#endif
2397

N
Nick Piggin 已提交
2398
	put_cpu();
2399
	return 0;
L
Linus Torvalds 已提交
2400 2401
}

2402 2403 2404
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
2405
		return BW_UNIT;
2406 2407 2408 2409 2410 2411 2412 2413 2414

	/*
	 * 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;

2415
	return div64_u64(runtime << BW_SHIFT, period);
2416 2417
}

L
Linus Torvalds 已提交
2418 2419 2420 2421 2422 2423 2424
/*
 * 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.
 */
2425
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2426
{
2427
	struct rq_flags rf;
I
Ingo Molnar 已提交
2428
	struct rq *rq;
2429

2430
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2431
	p->state = TASK_RUNNING;
2432 2433 2434 2435
#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
2436
	 *  - any previously selected CPU might disappear through hotplug
2437 2438 2439
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2440
	 */
2441
	p->recent_used_cpu = task_cpu(p);
2442
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2443
#endif
2444
	rq = __task_rq_lock(p, &rf);
2445
	update_rq_clock(rq);
2446
	post_init_entity_util_avg(&p->se);
2447

2448
	activate_task(rq, p, ENQUEUE_NOCLOCK);
2449
	p->on_rq = TASK_ON_RQ_QUEUED;
2450
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2451
	check_preempt_curr(rq, p, WF_FORK);
2452
#ifdef CONFIG_SMP
2453 2454 2455 2456 2457
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2458
		rq_unpin_lock(rq, &rf);
2459
		p->sched_class->task_woken(rq, p);
2460
		rq_repin_lock(rq, &rf);
2461
	}
2462
#endif
2463
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2464 2465
}

2466 2467
#ifdef CONFIG_PREEMPT_NOTIFIERS

2468
static DEFINE_STATIC_KEY_FALSE(preempt_notifier_key);
2469

2470 2471
void preempt_notifier_inc(void)
{
2472
	static_branch_inc(&preempt_notifier_key);
2473 2474 2475 2476 2477
}
EXPORT_SYMBOL_GPL(preempt_notifier_inc);

void preempt_notifier_dec(void)
{
2478
	static_branch_dec(&preempt_notifier_key);
2479 2480 2481
}
EXPORT_SYMBOL_GPL(preempt_notifier_dec);

2482
/**
2483
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2484
 * @notifier: notifier struct to register
2485 2486 2487
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2488
	if (!static_branch_unlikely(&preempt_notifier_key))
2489 2490
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2491 2492 2493 2494 2495 2496
	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
2497
 * @notifier: notifier struct to unregister
2498
 *
2499
 * This is *not* safe to call from within a preemption notifier.
2500 2501 2502 2503 2504 2505 2506
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2507
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2508 2509 2510
{
	struct preempt_notifier *notifier;

2511
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2512 2513 2514
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2515 2516
static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
2517
	if (static_branch_unlikely(&preempt_notifier_key))
2518 2519 2520
		__fire_sched_in_preempt_notifiers(curr);
}

2521
static void
2522 2523
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2524 2525 2526
{
	struct preempt_notifier *notifier;

2527
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2528 2529 2530
		notifier->ops->sched_out(notifier, next);
}

2531 2532 2533 2534
static __always_inline void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
2535
	if (static_branch_unlikely(&preempt_notifier_key))
2536 2537 2538
		__fire_sched_out_preempt_notifiers(curr, next);
}

2539
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2540

2541
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2542 2543 2544
{
}

2545
static inline void
2546 2547 2548 2549 2550
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2551
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2552

2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580
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
}

2581 2582
static inline void
prepare_lock_switch(struct rq *rq, struct task_struct *next, struct rq_flags *rf)
2583
{
2584 2585 2586 2587 2588 2589 2590 2591
	/*
	 * 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_);
2592 2593
#ifdef CONFIG_DEBUG_SPINLOCK
	/* this is a valid case when another task releases the spinlock */
2594
	rq->lock.owner = next;
2595
#endif
2596 2597 2598 2599
}

static inline void finish_lock_switch(struct rq *rq)
{
2600 2601 2602 2603 2604 2605 2606 2607 2608
	/*
	 * 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);
}

2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620
/*
 * 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

2621 2622 2623
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2624
 * @prev: the current task that is being switched out
2625 2626 2627 2628 2629 2630 2631 2632 2633
 * @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.
 */
2634 2635 2636
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2637
{
2638
	kcov_prepare_switch(prev);
2639
	sched_info_switch(rq, prev, next);
2640
	perf_event_task_sched_out(prev, next);
2641
	rseq_preempt(prev);
2642
	fire_sched_out_preempt_notifiers(prev, next);
2643
	prepare_task(next);
2644 2645 2646
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2647 2648 2649 2650
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2651 2652 2653 2654
 * 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
2655 2656
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2657
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2658 2659
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2660 2661 2662 2663 2664
 *
 * 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
2665
 */
2666
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2667 2668
	__releases(rq->lock)
{
2669
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2670
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2671
	long prev_state;
L
Linus Torvalds 已提交
2672

2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683
	/*
	 * 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.
	 */
2684 2685 2686 2687
	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);
2688

L
Linus Torvalds 已提交
2689 2690 2691 2692
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2693
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2694 2695
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2696 2697
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
2698
	 * finish_task), otherwise a concurrent wakeup can get prev
2699 2700
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2701
	 */
O
Oleg Nesterov 已提交
2702
	prev_state = prev->state;
2703
	vtime_task_switch(prev);
2704
	perf_event_task_sched_in(prev, current);
2705 2706
	finish_task(prev);
	finish_lock_switch(rq);
2707
	finish_arch_post_lock_switch();
2708
	kcov_finish_switch(current);
S
Steven Rostedt 已提交
2709

2710
	fire_sched_in_preempt_notifiers(current);
2711
	/*
2712 2713 2714 2715 2716 2717 2718 2719 2720 2721
	 * 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.
2722
	 */
2723 2724
	if (mm) {
		membarrier_mm_sync_core_before_usermode(mm);
L
Linus Torvalds 已提交
2725
		mmdrop(mm);
2726
	}
2727 2728 2729
	if (unlikely(prev_state == TASK_DEAD)) {
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);
2730

2731 2732 2733 2734 2735 2736 2737 2738 2739 2740
		/*
		 * 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);
2741
	}
2742

2743
	tick_nohz_task_switch();
2744
	return rq;
L
Linus Torvalds 已提交
2745 2746
}

2747 2748 2749
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2750
static void __balance_callback(struct rq *rq)
2751
{
2752 2753 2754
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2755

2756 2757 2758 2759 2760 2761 2762 2763
	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;
2764

2765
		func(rq);
2766
	}
2767 2768 2769 2770 2771 2772 2773
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2774 2775 2776
}

#else
2777

2778
static inline void balance_callback(struct rq *rq)
2779
{
L
Linus Torvalds 已提交
2780 2781
}

2782 2783
#endif

L
Linus Torvalds 已提交
2784 2785 2786 2787
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2788
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2789 2790
	__releases(rq->lock)
{
2791
	struct rq *rq;
2792

2793 2794 2795 2796 2797 2798 2799 2800 2801
	/*
	 * 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).
	 */

2802
	rq = finish_task_switch(prev);
2803
	balance_callback(rq);
2804
	preempt_enable();
2805

L
Linus Torvalds 已提交
2806
	if (current->set_child_tid)
2807
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2808 2809 2810
}

/*
2811
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2812
 */
2813
static __always_inline struct rq *
2814
context_switch(struct rq *rq, struct task_struct *prev,
2815
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2816
{
I
Ingo Molnar 已提交
2817
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2818

2819
	prepare_task_switch(rq, prev, next);
2820

I
Ingo Molnar 已提交
2821 2822
	mm = next->mm;
	oldmm = prev->active_mm;
2823 2824 2825 2826 2827
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2828
	arch_start_context_switch(prev);
2829

2830 2831 2832 2833 2834 2835 2836
	/*
	 * 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.
	 */
2837
	if (!mm) {
L
Linus Torvalds 已提交
2838
		next->active_mm = oldmm;
V
Vegard Nossum 已提交
2839
		mmgrab(oldmm);
L
Linus Torvalds 已提交
2840 2841
		enter_lazy_tlb(oldmm, next);
	} else
2842
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2843

2844
	if (!prev->mm) {
L
Linus Torvalds 已提交
2845 2846 2847
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2848

2849
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2850

2851
	prepare_lock_switch(rq, next, rf);
L
Linus Torvalds 已提交
2852 2853 2854

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2855
	barrier();
2856 2857

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2858 2859 2860
}

/*
2861
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2862 2863
 *
 * externally visible scheduler statistics: current number of runnable
2864
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2865 2866 2867 2868 2869 2870 2871 2872 2873
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2874
}
L
Linus Torvalds 已提交
2875

2876
/*
I
Ingo Molnar 已提交
2877
 * Check if only the current task is running on the CPU.
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887
 *
 * 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)
2888 2889 2890
 */
bool single_task_running(void)
{
2891
	return raw_rq()->nr_running == 1;
2892 2893 2894
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2895
unsigned long long nr_context_switches(void)
2896
{
2897 2898
	int i;
	unsigned long long sum = 0;
2899

2900
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2901
		sum += cpu_rq(i)->nr_switches;
2902

L
Linus Torvalds 已提交
2903 2904
	return sum;
}
2905

2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935
/*
 * 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
2936 2937 2938
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2939

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

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

2946 2947 2948 2949 2950 2951 2952
/*
 * 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.
 */

2953
unsigned long nr_iowait_cpu(int cpu)
2954
{
2955
	struct rq *this = cpu_rq(cpu);
2956 2957
	return atomic_read(&this->nr_iowait);
}
2958

2959 2960
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2961 2962 2963
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2964 2965
}

I
Ingo Molnar 已提交
2966
#ifdef CONFIG_SMP
2967

2968
/*
P
Peter Zijlstra 已提交
2969 2970
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2971
 */
P
Peter Zijlstra 已提交
2972
void sched_exec(void)
2973
{
P
Peter Zijlstra 已提交
2974
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2975
	unsigned long flags;
2976
	int dest_cpu;
2977

2978
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2979
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2980 2981
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2982

2983
	if (likely(cpu_active(dest_cpu))) {
2984
		struct migration_arg arg = { p, dest_cpu };
2985

2986 2987
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2988 2989
		return;
	}
2990
unlock:
2991
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2992
}
I
Ingo Molnar 已提交
2993

L
Linus Torvalds 已提交
2994 2995 2996
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2997
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2998 2999

EXPORT_PER_CPU_SYMBOL(kstat);
3000
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
3001

3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018
/*
 * 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);
}

3019 3020 3021 3022 3023 3024 3025
/*
 * 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)
{
3026
	struct rq_flags rf;
3027
	struct rq *rq;
3028
	u64 ns;
3029

3030 3031
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
3032
	 * 64-bit doesn't need locks to atomically read a 64-bit value.
3033 3034 3035
	 * 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
3036 3037
	 * 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
3038
	 * indistinguishable from the read occurring a few cycles earlier.
3039 3040
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3041
	 */
3042
	if (!p->on_cpu || !task_on_rq_queued(p))
3043 3044 3045
		return p->se.sum_exec_runtime;
#endif

3046
	rq = task_rq_lock(p, &rf);
3047 3048 3049 3050 3051 3052
	/*
	 * 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)) {
3053
		prefetch_curr_exec_start(p);
3054 3055 3056 3057
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3058
	task_rq_unlock(rq, p, &rf);
3059 3060 3061

	return ns;
}
3062

3063 3064 3065 3066 3067 3068 3069 3070
/*
 * 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
3071
	struct task_struct *curr = rq->curr;
3072
	struct rq_flags rf;
3073 3074

	sched_clock_tick();
I
Ingo Molnar 已提交
3075

3076 3077
	rq_lock(rq, &rf);

3078
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3079
	curr->sched_class->task_tick(rq, curr, 0);
3080
	cpu_load_update_active(rq);
3081
	calc_global_load_tick(rq);
3082 3083

	rq_unlock(rq, &rf);
3084

3085
	perf_event_task_tick();
3086

3087
#ifdef CONFIG_SMP
3088
	rq->idle_balance = idle_cpu(cpu);
3089
	trigger_load_balance(rq);
3090
#endif
L
Linus Torvalds 已提交
3091 3092
}

3093
#ifdef CONFIG_NO_HZ_FULL
3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107

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

	/*
	 * 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.
	 */
3119 3120
	if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
		goto out_requeue;
3121

3122 3123 3124 3125
	rq_lock_irq(rq, &rf);
	curr = rq->curr;
	if (is_idle_task(curr))
		goto out_unlock;
3126

3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138
	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);
3139

3140
out_requeue:
3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
	/*
	 * 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) { }
3190
#endif
L
Linus Torvalds 已提交
3191

3192 3193
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207
/*
 * 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);
	}
}
3208

3209
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3210
{
3211
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3212 3213 3214
	/*
	 * Underflow?
	 */
3215 3216
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3217
#endif
3218
	__preempt_count_add(val);
3219
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3220 3221 3222
	/*
	 * Spinlock count overflowing soon?
	 */
3223 3224
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3225
#endif
3226
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3227
}
3228
EXPORT_SYMBOL(preempt_count_add);
3229
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3230

3231 3232 3233 3234 3235 3236 3237 3238 3239 3240
/*
 * 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());
}

3241
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3242
{
3243
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3244 3245 3246
	/*
	 * Underflow?
	 */
3247
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3248
		return;
L
Linus Torvalds 已提交
3249 3250 3251
	/*
	 * Is the spinlock portion underflowing?
	 */
3252 3253 3254
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3255
#endif
3256

3257
	preempt_latency_stop(val);
3258
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3259
}
3260
EXPORT_SYMBOL(preempt_count_sub);
3261
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3262

3263 3264 3265
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3266 3267
#endif

3268 3269 3270 3271 3272 3273 3274 3275 3276
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
3277
/*
I
Ingo Molnar 已提交
3278
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3279
 */
I
Ingo Molnar 已提交
3280
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3281
{
3282 3283 3284
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3285 3286 3287
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3288 3289
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3290

I
Ingo Molnar 已提交
3291
	debug_show_held_locks(prev);
3292
	print_modules();
I
Ingo Molnar 已提交
3293 3294
	if (irqs_disabled())
		print_irqtrace_events(prev);
3295 3296
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3297
		pr_err("Preemption disabled at:");
3298
		print_ip_sym(preempt_disable_ip);
3299 3300
		pr_cont("\n");
	}
3301 3302 3303
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3304
	dump_stack();
3305
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3306
}
L
Linus Torvalds 已提交
3307

I
Ingo Molnar 已提交
3308 3309 3310 3311 3312
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3313
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3314 3315
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3316
#endif
3317

3318
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3319
		__schedule_bug(prev);
3320 3321
		preempt_count_set(PREEMPT_DISABLED);
	}
3322
	rcu_sleep_check();
I
Ingo Molnar 已提交
3323

L
Linus Torvalds 已提交
3324 3325
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3326
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3327 3328 3329 3330 3331 3332
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3333
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3334
{
3335
	const struct sched_class *class;
I
Ingo Molnar 已提交
3336
	struct task_struct *p;
L
Linus Torvalds 已提交
3337 3338

	/*
3339 3340 3341 3342
	 * 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
3343
	 */
3344 3345 3346 3347
	if (likely((prev->sched_class == &idle_sched_class ||
		    prev->sched_class == &fair_sched_class) &&
		   rq->nr_running == rq->cfs.h_nr_running)) {

3348
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3349 3350 3351
		if (unlikely(p == RETRY_TASK))
			goto again;

I
Ingo Molnar 已提交
3352
		/* Assumes fair_sched_class->next == idle_sched_class */
3353
		if (unlikely(!p))
3354
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3355 3356

		return p;
L
Linus Torvalds 已提交
3357 3358
	}

3359
again:
3360
	for_each_class(class) {
3361
		p = class->pick_next_task(rq, prev, rf);
3362 3363 3364
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3365
			return p;
3366
		}
I
Ingo Molnar 已提交
3367
	}
3368

I
Ingo Molnar 已提交
3369 3370
	/* The idle class should always have a runnable task: */
	BUG();
I
Ingo Molnar 已提交
3371
}
L
Linus Torvalds 已提交
3372

I
Ingo Molnar 已提交
3373
/*
3374
 * __schedule() is the main scheduler function.
3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408
 *
 * 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
3409
 *
3410
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3411
 */
3412
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3413 3414
{
	struct task_struct *prev, *next;
3415
	unsigned long *switch_count;
3416
	struct rq_flags rf;
I
Ingo Molnar 已提交
3417
	struct rq *rq;
3418
	int cpu;
I
Ingo Molnar 已提交
3419 3420 3421 3422 3423 3424

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3425

3426
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3427
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3428

3429
	local_irq_disable();
3430
	rcu_note_context_switch(preempt);
3431

3432 3433 3434 3435
	/*
	 * 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().
3436 3437 3438
	 *
	 * The membarrier system call requires a full memory barrier
	 * after coming from user-space, before storing to rq->curr.
3439
	 */
3440
	rq_lock(rq, &rf);
3441
	smp_mb__after_spinlock();
L
Linus Torvalds 已提交
3442

I
Ingo Molnar 已提交
3443 3444
	/* Promote REQ to ACT */
	rq->clock_update_flags <<= 1;
3445
	update_rq_clock(rq);
3446

3447
	switch_count = &prev->nivcsw;
3448
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3449
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3450
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3451
		} else {
3452
			deactivate_task(rq, prev, DEQUEUE_SLEEP | DEQUEUE_NOCLOCK);
3453 3454
			prev->on_rq = 0;

3455 3456 3457 3458 3459
			if (prev->in_iowait) {
				atomic_inc(&rq->nr_iowait);
				delayacct_blkio_start();
			}

T
Tejun Heo 已提交
3460
			/*
3461 3462 3463
			 * 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
3464 3465 3466 3467
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3468
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3469
				if (to_wakeup)
3470
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3471 3472
			}
		}
I
Ingo Molnar 已提交
3473
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3474 3475
	}

3476
	next = pick_next_task(rq, prev, &rf);
3477
	clear_tsk_need_resched(prev);
3478
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3479 3480 3481 3482

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
3483 3484 3485
		/*
		 * The membarrier system call requires each architecture
		 * to have a full memory barrier after updating
3486 3487 3488 3489 3490 3491 3492 3493 3494 3495
		 * 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),
3496
		 */
L
Linus Torvalds 已提交
3497 3498
		++*switch_count;

3499
		trace_sched_switch(preempt, prev, next);
I
Ingo Molnar 已提交
3500 3501 3502

		/* Also unlocks the rq: */
		rq = context_switch(rq, prev, next, &rf);
3503
	} else {
3504
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3505
		rq_unlock_irq(rq, &rf);
3506
	}
L
Linus Torvalds 已提交
3507

3508
	balance_callback(rq);
L
Linus Torvalds 已提交
3509
}
3510

3511 3512
void __noreturn do_task_dead(void)
{
I
Ingo Molnar 已提交
3513
	/* Causes final put_task_struct in finish_task_switch(): */
3514
	set_special_state(TASK_DEAD);
I
Ingo Molnar 已提交
3515 3516 3517 3518

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

3519 3520
	__schedule(false);
	BUG();
I
Ingo Molnar 已提交
3521 3522

	/* Avoid "noreturn function does return" - but don't continue if BUG() is a NOP: */
3523
	for (;;)
I
Ingo Molnar 已提交
3524
		cpu_relax();
3525 3526
}

3527 3528
static inline void sched_submit_work(struct task_struct *tsk)
{
3529
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3530 3531 3532 3533 3534 3535 3536 3537 3538
		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);
}

3539
asmlinkage __visible void __sched schedule(void)
3540
{
3541 3542 3543
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3544
	do {
3545
		preempt_disable();
3546
		__schedule(false);
3547
		sched_preempt_enable_no_resched();
3548
	} while (need_resched());
3549
}
L
Linus Torvalds 已提交
3550 3551
EXPORT_SYMBOL(schedule);

3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576
/*
 * 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());
}

3577
#ifdef CONFIG_CONTEXT_TRACKING
3578
asmlinkage __visible void __sched schedule_user(void)
3579 3580 3581 3582 3583 3584
{
	/*
	 * 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.
3585 3586
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3587
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3588
	 * too frequently to make sense yet.
3589
	 */
3590
	enum ctx_state prev_state = exception_enter();
3591
	schedule();
3592
	exception_exit(prev_state);
3593 3594 3595
}
#endif

3596 3597 3598 3599 3600 3601 3602
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3603
	sched_preempt_enable_no_resched();
3604 3605 3606 3607
	schedule();
	preempt_disable();
}

3608
static void __sched notrace preempt_schedule_common(void)
3609 3610
{
	do {
3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623
		/*
		 * 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.
		 */
3624
		preempt_disable_notrace();
3625
		preempt_latency_start(1);
3626
		__schedule(true);
3627
		preempt_latency_stop(1);
3628
		preempt_enable_no_resched_notrace();
3629 3630 3631 3632 3633 3634 3635 3636

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

L
Linus Torvalds 已提交
3637 3638
#ifdef CONFIG_PREEMPT
/*
3639
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3640
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3641 3642
 * occur there and call schedule directly.
 */
3643
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3644 3645 3646
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3647
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3648
	 */
3649
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3650 3651
		return;

3652
	preempt_schedule_common();
L
Linus Torvalds 已提交
3653
}
3654
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3655
EXPORT_SYMBOL(preempt_schedule);
3656 3657

/**
3658
 * preempt_schedule_notrace - preempt_schedule called by tracing
3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670
 *
 * 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.
 */
3671
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3672 3673 3674 3675 3676 3677 3678
{
	enum ctx_state prev_ctx;

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

	do {
3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691
		/*
		 * 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.
		 */
3692
		preempt_disable_notrace();
3693
		preempt_latency_start(1);
3694 3695 3696 3697 3698 3699
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3700
		__schedule(true);
3701 3702
		exception_exit(prev_ctx);

3703
		preempt_latency_stop(1);
3704
		preempt_enable_no_resched_notrace();
3705 3706
	} while (need_resched());
}
3707
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3708

3709
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3710 3711

/*
3712
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3713 3714 3715 3716
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3717
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3718
{
3719
	enum ctx_state prev_state;
3720

3721
	/* Catch callers which need to be fixed */
3722
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3723

3724 3725
	prev_state = exception_enter();

3726
	do {
3727
		preempt_disable();
3728
		local_irq_enable();
3729
		__schedule(true);
3730
		local_irq_disable();
3731
		sched_preempt_enable_no_resched();
3732
	} while (need_resched());
3733 3734

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3735 3736
}

3737
int default_wake_function(wait_queue_entry_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3738
			  void *key)
L
Linus Torvalds 已提交
3739
{
P
Peter Zijlstra 已提交
3740
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3741 3742 3743
}
EXPORT_SYMBOL(default_wake_function);

3744 3745
#ifdef CONFIG_RT_MUTEXES

3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760
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);
}

3761 3762
/*
 * rt_mutex_setprio - set the current priority of a task
3763 3764
 * @p: task to boost
 * @pi_task: donor task
3765 3766 3767 3768
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3769 3770
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3771
 */
3772
void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
3773
{
3774
	int prio, oldprio, queued, running, queue_flag =
3775
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
3776
	const struct sched_class *prev_class;
3777 3778
	struct rq_flags rf;
	struct rq *rq;
3779

3780 3781 3782 3783 3784 3785 3786 3787
	/* 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;
3788

3789
	rq = __task_rq_lock(p, &rf);
3790
	update_rq_clock(rq);
3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807
	/*
	 * 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;
3808

3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826
	/*
	 * 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;
	}

3827
	trace_sched_pi_setprio(p, pi_task);
3828
	oldprio = p->prio;
3829 3830 3831 3832

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

3833
	prev_class = p->sched_class;
3834
	queued = task_on_rq_queued(p);
3835
	running = task_current(rq, p);
3836
	if (queued)
3837
		dequeue_task(rq, p, queue_flag);
3838
	if (running)
3839
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3840

3841 3842 3843 3844 3845 3846 3847 3848 3849 3850
	/*
	 * 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)) {
3851 3852
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3853
			p->dl.dl_boosted = 1;
3854
			queue_flag |= ENQUEUE_REPLENISH;
3855 3856
		} else
			p->dl.dl_boosted = 0;
3857
		p->sched_class = &dl_sched_class;
3858 3859 3860 3861
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3862
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3863
		p->sched_class = &rt_sched_class;
3864 3865 3866
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3867 3868
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3869
		p->sched_class = &fair_sched_class;
3870
	}
I
Ingo Molnar 已提交
3871

3872 3873
	p->prio = prio;

3874
	if (queued)
3875
		enqueue_task(rq, p, queue_flag);
3876
	if (running)
3877
		set_curr_task(rq, p);
3878

P
Peter Zijlstra 已提交
3879
	check_class_changed(rq, p, prev_class, oldprio);
3880
out_unlock:
I
Ingo Molnar 已提交
3881 3882
	/* Avoid rq from going away on us: */
	preempt_disable();
3883
	__task_rq_unlock(rq, &rf);
3884 3885 3886

	balance_callback(rq);
	preempt_enable();
3887
}
3888 3889 3890 3891 3892
#else
static inline int rt_effective_prio(struct task_struct *p, int prio)
{
	return prio;
}
3893
#endif
3894

3895
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3896
{
P
Peter Zijlstra 已提交
3897 3898
	bool queued, running;
	int old_prio, delta;
3899
	struct rq_flags rf;
3900
	struct rq *rq;
L
Linus Torvalds 已提交
3901

3902
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3903 3904 3905 3906 3907
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3908
	rq = task_rq_lock(p, &rf);
3909 3910
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3911 3912 3913 3914
	/*
	 * 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
3915
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3916
	 */
3917
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3918 3919 3920
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3921
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3922
	running = task_current(rq, p);
3923
	if (queued)
3924
		dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK);
P
Peter Zijlstra 已提交
3925 3926
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3927 3928

	p->static_prio = NICE_TO_PRIO(nice);
3929
	set_load_weight(p, true);
3930 3931 3932
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3933

3934
	if (queued) {
3935
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
L
Linus Torvalds 已提交
3936
		/*
3937 3938
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3939
		 */
3940
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3941
			resched_curr(rq);
L
Linus Torvalds 已提交
3942
	}
P
Peter Zijlstra 已提交
3943 3944
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3945
out_unlock:
3946
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3947 3948 3949
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3950 3951 3952 3953 3954
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3955
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3956
{
I
Ingo Molnar 已提交
3957
	/* Convert nice value [19,-20] to rlimit style value [1,40]: */
3958
	int nice_rlim = nice_to_rlimit(nice);
3959

3960
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3961 3962 3963
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3964 3965 3966 3967 3968 3969 3970 3971 3972
#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.
 */
3973
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3974
{
3975
	long nice, retval;
L
Linus Torvalds 已提交
3976 3977 3978 3979 3980 3981

	/*
	 * 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.
	 */
3982
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3983
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3984

3985
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3986 3987 3988
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002
	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.
 *
4003
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
4004 4005 4006
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
4007
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
4008 4009 4010 4011 4012
{
	return p->prio - MAX_RT_PRIO;
}

/**
I
Ingo Molnar 已提交
4013
 * idle_cpu - is a given CPU idle currently?
L
Linus Torvalds 已提交
4014
 * @cpu: the processor in question.
4015 4016
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
4017 4018 4019
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033
	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
4034 4035
}

4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046
/**
 * 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;

4047 4048 4049
	if (vcpu_is_preempted(cpu))
		return 0;

T
Thomas Gleixner 已提交
4050
	return 1;
L
Linus Torvalds 已提交
4051 4052 4053
}

/**
I
Ingo Molnar 已提交
4054
 * idle_task - return the idle task for a given CPU.
L
Linus Torvalds 已提交
4055
 * @cpu: the processor in question.
4056
 *
I
Ingo Molnar 已提交
4057
 * Return: The idle task for the CPU @cpu.
L
Linus Torvalds 已提交
4058
 */
4059
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
4060 4061 4062 4063 4064 4065 4066
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
4067 4068
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
4069
 */
A
Alexey Dobriyan 已提交
4070
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
4071
{
4072
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
4073 4074
}

4075 4076 4077 4078 4079 4080
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

4081 4082
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
4083
{
4084 4085
	int policy = attr->sched_policy;

4086
	if (policy == SETPARAM_POLICY)
4087 4088
		policy = p->policy;

L
Linus Torvalds 已提交
4089
	p->policy = policy;
4090

4091 4092
	if (dl_policy(policy))
		__setparam_dl(p, attr);
4093
	else if (fair_policy(policy))
4094 4095
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

4096 4097 4098 4099 4100 4101
	/*
	 * __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;
4102
	p->normal_prio = normal_prio(p);
4103
	set_load_weight(p, true);
4104
}
4105

4106 4107
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
4108
			   const struct sched_attr *attr, bool keep_boost)
4109 4110
{
	__setscheduler_params(p, attr);
4111

4112
	/*
4113 4114
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
4115
	 */
4116
	p->prio = normal_prio(p);
4117
	if (keep_boost)
4118
		p->prio = rt_effective_prio(p, p->prio);
4119

4120 4121 4122
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
4123 4124 4125
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
4126
}
4127

4128
/*
I
Ingo Molnar 已提交
4129
 * Check the target process has a UID that matches the current process's:
4130 4131 4132 4133 4134 4135 4136 4137
 */
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);
4138 4139
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4140 4141 4142 4143
	rcu_read_unlock();
	return match;
}

4144 4145
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4146
				bool user, bool pi)
L
Linus Torvalds 已提交
4147
{
4148 4149
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4150
	int retval, oldprio, oldpolicy = -1, queued, running;
4151
	int new_effective_prio, policy = attr->sched_policy;
4152
	const struct sched_class *prev_class;
4153
	struct rq_flags rf;
4154
	int reset_on_fork;
4155
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
4156
	struct rq *rq;
L
Linus Torvalds 已提交
4157

4158 4159
	/* The pi code expects interrupts enabled */
	BUG_ON(pi && in_interrupt());
L
Linus Torvalds 已提交
4160
recheck:
I
Ingo Molnar 已提交
4161
	/* Double check policy once rq lock held: */
4162 4163
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4164
		policy = oldpolicy = p->policy;
4165
	} else {
4166
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4167

4168
		if (!valid_policy(policy))
4169 4170 4171
			return -EINVAL;
	}

4172
	if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV))
4173 4174
		return -EINVAL;

L
Linus Torvalds 已提交
4175 4176
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4177 4178
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4179
	 */
4180
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4181
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4182
		return -EINVAL;
4183 4184
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4185 4186
		return -EINVAL;

4187 4188 4189
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4190
	if (user && !capable(CAP_SYS_NICE)) {
4191
		if (fair_policy(policy)) {
4192
			if (attr->sched_nice < task_nice(p) &&
4193
			    !can_nice(p, attr->sched_nice))
4194 4195 4196
				return -EPERM;
		}

4197
		if (rt_policy(policy)) {
4198 4199
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4200

I
Ingo Molnar 已提交
4201
			/* Can't set/change the rt policy: */
4202 4203 4204
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

I
Ingo Molnar 已提交
4205
			/* Can't increase priority: */
4206 4207
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4208 4209
				return -EPERM;
		}
4210

4211 4212 4213 4214 4215 4216 4217 4218 4219
		 /*
		  * 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
4220
		/*
4221 4222
		 * 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
4223
		 */
4224
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4225
			if (!can_nice(p, task_nice(p)))
4226 4227
				return -EPERM;
		}
4228

I
Ingo Molnar 已提交
4229
		/* Can't change other user's priorities: */
4230
		if (!check_same_owner(p))
4231
			return -EPERM;
4232

I
Ingo Molnar 已提交
4233
		/* Normal users shall not reset the sched_reset_on_fork flag: */
4234 4235
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4236
	}
L
Linus Torvalds 已提交
4237

4238
	if (user) {
4239 4240 4241
		if (attr->sched_flags & SCHED_FLAG_SUGOV)
			return -EINVAL;

4242
		retval = security_task_setscheduler(p);
4243 4244 4245 4246
		if (retval)
			return retval;
	}

4247
	/*
I
Ingo Molnar 已提交
4248
	 * Make sure no PI-waiters arrive (or leave) while we are
4249
	 * changing the priority of the task:
4250
	 *
L
Lucas De Marchi 已提交
4251
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4252 4253
	 * runqueue lock must be held.
	 */
4254
	rq = task_rq_lock(p, &rf);
4255
	update_rq_clock(rq);
4256

4257
	/*
I
Ingo Molnar 已提交
4258
	 * Changing the policy of the stop threads its a very bad idea:
4259 4260
	 */
	if (p == rq->stop) {
4261
		task_rq_unlock(rq, p, &rf);
4262 4263 4264
		return -EINVAL;
	}

4265
	/*
4266 4267
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4268
	 */
4269
	if (unlikely(policy == p->policy)) {
4270
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4271 4272 4273
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4274
		if (dl_policy(policy) && dl_param_changed(p, attr))
4275
			goto change;
4276

4277
		p->sched_reset_on_fork = reset_on_fork;
4278
		task_rq_unlock(rq, p, &rf);
4279 4280
		return 0;
	}
4281
change:
4282

4283
	if (user) {
4284
#ifdef CONFIG_RT_GROUP_SCHED
4285 4286 4287 4288 4289
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4290 4291
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4292
			task_rq_unlock(rq, p, &rf);
4293 4294 4295
			return -EPERM;
		}
#endif
4296
#ifdef CONFIG_SMP
4297 4298
		if (dl_bandwidth_enabled() && dl_policy(policy) &&
				!(attr->sched_flags & SCHED_FLAG_SUGOV)) {
4299 4300 4301 4302 4303 4304 4305
			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.
			 */
4306 4307
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4308
				task_rq_unlock(rq, p, &rf);
4309 4310 4311 4312 4313
				return -EPERM;
			}
		}
#endif
	}
4314

I
Ingo Molnar 已提交
4315
	/* Re-check policy now with rq lock held: */
L
Linus Torvalds 已提交
4316 4317
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4318
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4319 4320
		goto recheck;
	}
4321 4322 4323 4324 4325 4326

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4327
	if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) {
4328
		task_rq_unlock(rq, p, &rf);
4329 4330 4331
		return -EBUSY;
	}

4332 4333 4334
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4335 4336 4337 4338 4339 4340 4341 4342
	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.
		 */
4343
		new_effective_prio = rt_effective_prio(p, newprio);
4344 4345
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4346 4347
	}

4348
	queued = task_on_rq_queued(p);
4349
	running = task_current(rq, p);
4350
	if (queued)
4351
		dequeue_task(rq, p, queue_flags);
4352
	if (running)
4353
		put_prev_task(rq, p);
4354

4355
	prev_class = p->sched_class;
4356
	__setscheduler(rq, p, attr, pi);
4357

4358
	if (queued) {
4359 4360 4361 4362
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4363 4364
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4365

4366
		enqueue_task(rq, p, queue_flags);
4367
	}
4368
	if (running)
4369
		set_curr_task(rq, p);
4370

P
Peter Zijlstra 已提交
4371
	check_class_changed(rq, p, prev_class, oldprio);
I
Ingo Molnar 已提交
4372 4373 4374

	/* Avoid rq from going away on us: */
	preempt_disable();
4375
	task_rq_unlock(rq, p, &rf);
4376

4377 4378
	if (pi)
		rt_mutex_adjust_pi(p);
4379

I
Ingo Molnar 已提交
4380
	/* Run balance callbacks after we've adjusted the PI chain: */
4381 4382
	balance_callback(rq);
	preempt_enable();
4383

L
Linus Torvalds 已提交
4384 4385
	return 0;
}
4386

4387 4388 4389 4390 4391 4392 4393 4394 4395
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),
	};

4396 4397
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4398 4399 4400 4401 4402
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4403
	return __sched_setscheduler(p, &attr, check, true);
4404
}
4405 4406 4407 4408 4409 4410
/**
 * 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.
 *
4411 4412
 * Return: 0 on success. An error code otherwise.
 *
4413 4414 4415
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4416
		       const struct sched_param *param)
4417
{
4418
	return _sched_setscheduler(p, policy, param, true);
4419
}
L
Linus Torvalds 已提交
4420 4421
EXPORT_SYMBOL_GPL(sched_setscheduler);

4422 4423
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4424
	return __sched_setscheduler(p, attr, true, true);
4425 4426 4427
}
EXPORT_SYMBOL_GPL(sched_setattr);

4428 4429 4430 4431 4432
int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr)
{
	return __sched_setscheduler(p, attr, false, true);
}

4433 4434 4435 4436 4437 4438 4439 4440 4441 4442
/**
 * 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.
4443 4444
 *
 * Return: 0 on success. An error code otherwise.
4445 4446
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4447
			       const struct sched_param *param)
4448
{
4449
	return _sched_setscheduler(p, policy, param, false);
4450
}
4451
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4452

I
Ingo Molnar 已提交
4453 4454
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4455 4456 4457
{
	struct sched_param lparam;
	struct task_struct *p;
4458
	int retval;
L
Linus Torvalds 已提交
4459 4460 4461 4462 4463

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4464 4465 4466

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4467
	p = find_process_by_pid(pid);
4468 4469 4470
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4471

L
Linus Torvalds 已提交
4472 4473 4474
	return retval;
}

4475 4476 4477
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
I
Ingo Molnar 已提交
4478
static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr)
4479 4480 4481 4482 4483 4484 4485
{
	u32 size;
	int ret;

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

I
Ingo Molnar 已提交
4486
	/* Zero the full structure, so that a short copy will be nice: */
4487 4488 4489 4490 4491 4492
	memset(attr, 0, sizeof(*attr));

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

I
Ingo Molnar 已提交
4493 4494
	/* Bail out on silly large: */
	if (size > PAGE_SIZE)
4495 4496
		goto err_size;

I
Ingo Molnar 已提交
4497 4498
	/* ABI compatibility quirk: */
	if (!size)
4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532
		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
4533
	 * XXX: Do we want to be lenient like existing syscalls; or do we want
4534 4535
	 * to be strict and return an error on out-of-bounds values?
	 */
4536
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4537

4538
	return 0;
4539 4540 4541

err_size:
	put_user(sizeof(*attr), &uattr->size);
4542
	return -E2BIG;
4543 4544
}

L
Linus Torvalds 已提交
4545 4546 4547 4548 4549
/**
 * 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.
4550 4551
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4552
 */
I
Ingo Molnar 已提交
4553
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4554
{
4555 4556 4557
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4558 4559 4560 4561 4562 4563 4564
	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.
4565 4566
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4567
 */
4568
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4569
{
4570
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4571 4572
}

4573 4574 4575
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4576
 * @uattr: structure containing the extended parameters.
4577
 * @flags: for future extension.
4578
 */
4579 4580
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4581 4582 4583 4584 4585
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4586
	if (!uattr || pid < 0 || flags)
4587 4588
		return -EINVAL;

4589 4590 4591
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4592

4593
	if ((int)attr.sched_policy < 0)
4594
		return -EINVAL;
4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605

	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
4606 4607 4608
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4609 4610 4611
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4612
 */
4613
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4614
{
4615
	struct task_struct *p;
4616
	int retval;
L
Linus Torvalds 已提交
4617 4618

	if (pid < 0)
4619
		return -EINVAL;
L
Linus Torvalds 已提交
4620 4621

	retval = -ESRCH;
4622
	rcu_read_lock();
L
Linus Torvalds 已提交
4623 4624 4625 4626
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4627 4628
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4629
	}
4630
	rcu_read_unlock();
L
Linus Torvalds 已提交
4631 4632 4633 4634
	return retval;
}

/**
4635
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4636 4637
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4638 4639 4640
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4641
 */
4642
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4643
{
4644
	struct sched_param lp = { .sched_priority = 0 };
4645
	struct task_struct *p;
4646
	int retval;
L
Linus Torvalds 已提交
4647 4648

	if (!param || pid < 0)
4649
		return -EINVAL;
L
Linus Torvalds 已提交
4650

4651
	rcu_read_lock();
L
Linus Torvalds 已提交
4652 4653 4654 4655 4656 4657 4658 4659 4660
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4661 4662
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4663
	rcu_read_unlock();
L
Linus Torvalds 已提交
4664 4665 4666 4667 4668 4669 4670 4671 4672

	/*
	 * 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:
4673
	rcu_read_unlock();
L
Linus Torvalds 已提交
4674 4675 4676
	return retval;
}

4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699
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)
4700
				return -EFBIG;
4701 4702 4703 4704 4705
		}

		attr->size = usize;
	}

4706
	ret = copy_to_user(uattr, attr, attr->size);
4707 4708 4709
	if (ret)
		return -EFAULT;

4710
	return 0;
4711 4712 4713
}

/**
4714
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4715
 * @pid: the pid in question.
J
Juri Lelli 已提交
4716
 * @uattr: structure containing the extended parameters.
4717
 * @size: sizeof(attr) for fwd/bwd comp.
4718
 * @flags: for future extension.
4719
 */
4720 4721
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4722 4723 4724 4725 4726 4727 4728 4729
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4730
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743
		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;
4744 4745
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4746 4747 4748
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4749 4750
		attr.sched_priority = p->rt_priority;
	else
4751
		attr.sched_nice = task_nice(p);
4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4763
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4764
{
4765
	cpumask_var_t cpus_allowed, new_mask;
4766 4767
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4768

4769
	rcu_read_lock();
L
Linus Torvalds 已提交
4770 4771 4772

	p = find_process_by_pid(pid);
	if (!p) {
4773
		rcu_read_unlock();
L
Linus Torvalds 已提交
4774 4775 4776
		return -ESRCH;
	}

4777
	/* Prevent p going away */
L
Linus Torvalds 已提交
4778
	get_task_struct(p);
4779
	rcu_read_unlock();
L
Linus Torvalds 已提交
4780

4781 4782 4783 4784
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4785 4786 4787 4788 4789 4790 4791 4792
	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
4793
	retval = -EPERM;
E
Eric W. Biederman 已提交
4794 4795 4796 4797
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4798
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4799 4800 4801
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4802

4803
	retval = security_task_setscheduler(p);
4804
	if (retval)
4805
		goto out_free_new_mask;
4806

4807 4808 4809 4810

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

4811 4812 4813 4814 4815 4816 4817
	/*
	 * 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
4818 4819 4820
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4821
			retval = -EBUSY;
4822
			rcu_read_unlock();
4823
			goto out_free_new_mask;
4824
		}
4825
		rcu_read_unlock();
4826 4827
	}
#endif
P
Peter Zijlstra 已提交
4828
again:
4829
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4830

P
Paul Menage 已提交
4831
	if (!retval) {
4832 4833
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4834 4835 4836 4837 4838
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4839
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4840 4841 4842
			goto again;
		}
	}
4843
out_free_new_mask:
4844 4845 4846 4847
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4848 4849 4850 4851 4852
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4853
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4854
{
4855 4856 4857 4858 4859
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4860 4861 4862 4863
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
I
Ingo Molnar 已提交
4864
 * sys_sched_setaffinity - set the CPU affinity of a process
L
Linus Torvalds 已提交
4865 4866
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4867
 * @user_mask_ptr: user-space pointer to the new CPU mask
4868 4869
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4870
 */
4871 4872
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4873
{
4874
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4875 4876
	int retval;

4877 4878
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4879

4880 4881 4882 4883 4884
	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
4885 4886
}

4887
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4888
{
4889
	struct task_struct *p;
4890
	unsigned long flags;
L
Linus Torvalds 已提交
4891 4892
	int retval;

4893
	rcu_read_lock();
L
Linus Torvalds 已提交
4894 4895 4896 4897 4898 4899

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

4900 4901 4902 4903
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4904
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4905
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4906
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4907 4908

out_unlock:
4909
	rcu_read_unlock();
L
Linus Torvalds 已提交
4910

4911
	return retval;
L
Linus Torvalds 已提交
4912 4913 4914
}

/**
I
Ingo Molnar 已提交
4915
 * sys_sched_getaffinity - get the CPU affinity of a process
L
Linus Torvalds 已提交
4916 4917
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
I
Ingo Molnar 已提交
4918
 * @user_mask_ptr: user-space pointer to hold the current CPU mask
4919
 *
4920 4921
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4922
 */
4923 4924
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4925 4926
{
	int ret;
4927
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4928

A
Anton Blanchard 已提交
4929
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4930 4931
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4932 4933
		return -EINVAL;

4934 4935
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4936

4937 4938
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4939
		unsigned int retlen = min(len, cpumask_size());
4940 4941

		if (copy_to_user(user_mask_ptr, mask, retlen))
4942 4943
			ret = -EFAULT;
		else
4944
			ret = retlen;
4945 4946
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4947

4948
	return ret;
L
Linus Torvalds 已提交
4949 4950 4951 4952 4953
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4954 4955
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4956 4957
 *
 * Return: 0.
L
Linus Torvalds 已提交
4958
 */
4959
static void do_sched_yield(void)
L
Linus Torvalds 已提交
4960
{
4961 4962 4963 4964 4965 4966
	struct rq_flags rf;
	struct rq *rq;

	local_irq_disable();
	rq = this_rq();
	rq_lock(rq, &rf);
L
Linus Torvalds 已提交
4967

4968
	schedstat_inc(rq->yld_count);
4969
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4970 4971 4972 4973 4974

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
4975 4976
	preempt_disable();
	rq_unlock(rq, &rf);
4977
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4978 4979

	schedule();
4980
}
L
Linus Torvalds 已提交
4981

4982 4983 4984
SYSCALL_DEFINE0(sched_yield)
{
	do_sched_yield();
L
Linus Torvalds 已提交
4985 4986 4987
	return 0;
}

4988
#ifndef CONFIG_PREEMPT
4989
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4990
{
4991
	if (should_resched(0)) {
4992
		preempt_schedule_common();
L
Linus Torvalds 已提交
4993 4994
		return 1;
	}
4995
	rcu_all_qs();
L
Linus Torvalds 已提交
4996 4997
	return 0;
}
4998
EXPORT_SYMBOL(_cond_resched);
4999
#endif
L
Linus Torvalds 已提交
5000 5001

/*
5002
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
5003 5004
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
5005
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
5006 5007 5008
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
5009
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
5010
{
5011
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
5012 5013
	int ret = 0;

5014 5015
	lockdep_assert_held(lock);

5016
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
5017
		spin_unlock(lock);
P
Peter Zijlstra 已提交
5018
		if (resched)
5019
			preempt_schedule_common();
N
Nick Piggin 已提交
5020 5021
		else
			cpu_relax();
J
Jan Kara 已提交
5022
		ret = 1;
L
Linus Torvalds 已提交
5023 5024
		spin_lock(lock);
	}
J
Jan Kara 已提交
5025
	return ret;
L
Linus Torvalds 已提交
5026
}
5027
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
5028 5029 5030 5031

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
5032 5033 5034 5035 5036 5037 5038 5039 5040
 * 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
5041
 *	yield();
P
Peter Zijlstra 已提交
5042 5043 5044 5045 5046 5047 5048 5049
 *
 * 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
5050 5051 5052 5053
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
5054
	do_sched_yield();
L
Linus Torvalds 已提交
5055 5056 5057
}
EXPORT_SYMBOL(yield);

5058 5059 5060 5061
/**
 * 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
5062 5063
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5064 5065 5066 5067
 *
 * 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.
 *
5068
 * Return:
5069 5070 5071
 *	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.
5072
 */
5073
int __sched yield_to(struct task_struct *p, bool preempt)
5074 5075 5076 5077
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5078
	int yielded = 0;
5079 5080 5081 5082 5083 5084

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5085 5086 5087 5088 5089 5090 5091 5092 5093
	/*
	 * 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;
	}

5094
	double_rq_lock(rq, p_rq);
5095
	if (task_rq(p) != p_rq) {
5096 5097 5098 5099 5100
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5101
		goto out_unlock;
5102 5103

	if (curr->sched_class != p->sched_class)
5104
		goto out_unlock;
5105 5106

	if (task_running(p_rq, p) || p->state)
5107
		goto out_unlock;
5108 5109

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5110
	if (yielded) {
5111
		schedstat_inc(rq->yld_count);
5112 5113 5114 5115 5116
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5117
			resched_curr(p_rq);
5118
	}
5119

5120
out_unlock:
5121
	double_rq_unlock(rq, p_rq);
5122
out_irq:
5123 5124
	local_irq_restore(flags);

5125
	if (yielded > 0)
5126 5127 5128 5129 5130 5131
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146
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
5147
/*
I
Ingo Molnar 已提交
5148
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5149 5150 5151 5152
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5153
	int token;
L
Linus Torvalds 已提交
5154 5155
	long ret;

5156
	token = io_schedule_prepare();
L
Linus Torvalds 已提交
5157
	ret = schedule_timeout(timeout);
5158
	io_schedule_finish(token);
5159

L
Linus Torvalds 已提交
5160 5161
	return ret;
}
5162
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5163

5164 5165 5166 5167 5168 5169 5170 5171 5172 5173
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
5174 5175 5176 5177
/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5178 5179 5180
 * 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
5181
 */
5182
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5183 5184 5185 5186 5187 5188 5189 5190
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5191
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5192
	case SCHED_NORMAL:
5193
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5194
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5195 5196 5197 5198 5199 5200 5201 5202 5203 5204
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5205 5206 5207
 * 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
5208
 */
5209
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5210 5211 5212 5213 5214 5215 5216 5217
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5218
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5219
	case SCHED_NORMAL:
5220
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5221
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5222 5223 5224 5225 5226
		ret = 0;
	}
	return ret;
}

5227
static int sched_rr_get_interval(pid_t pid, struct timespec64 *t)
L
Linus Torvalds 已提交
5228
{
5229
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5230
	unsigned int time_slice;
5231
	struct rq_flags rf;
5232
	struct rq *rq;
5233
	int retval;
L
Linus Torvalds 已提交
5234 5235

	if (pid < 0)
5236
		return -EINVAL;
L
Linus Torvalds 已提交
5237 5238

	retval = -ESRCH;
5239
	rcu_read_lock();
L
Linus Torvalds 已提交
5240 5241 5242 5243 5244 5245 5246 5247
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5248
	rq = task_rq_lock(p, &rf);
5249 5250 5251
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5252
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5253

5254
	rcu_read_unlock();
5255 5256
	jiffies_to_timespec64(time_slice, t);
	return 0;
5257

L
Linus Torvalds 已提交
5258
out_unlock:
5259
	rcu_read_unlock();
L
Linus Torvalds 已提交
5260 5261 5262
	return retval;
}

5263 5264 5265 5266 5267 5268 5269 5270 5271 5272 5273
/**
 * 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.
 */
5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299
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

5300
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5301 5302
{
	unsigned long free = 0;
5303
	int ppid;
5304

5305 5306
	if (!try_get_task_stack(p))
		return;
5307 5308 5309 5310

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

	if (p->state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5311
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5312
#ifdef CONFIG_DEBUG_STACK_USAGE
5313
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5314
#endif
5315
	ppid = 0;
5316
	rcu_read_lock();
5317 5318
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5319
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5320
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5321
		task_pid_nr(p), ppid,
5322
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5323

5324
	print_worker_info(KERN_INFO, p);
5325
	show_stack(p, NULL);
5326
	put_task_stack(p);
L
Linus Torvalds 已提交
5327
}
5328
EXPORT_SYMBOL_GPL(sched_show_task);
L
Linus Torvalds 已提交
5329

5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351
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
5352
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5353
{
5354
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5355

5356
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5357 5358
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5359
#else
P
Peter Zijlstra 已提交
5360 5361
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5362
#endif
5363
	rcu_read_lock();
5364
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5365 5366
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5367
		 * console might take a lot of time:
5368 5369 5370
		 * 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
5371 5372
		 */
		touch_nmi_watchdog();
5373
		touch_all_softlockup_watchdogs();
5374
		if (state_filter_match(state_filter, p))
5375
			sched_show_task(p);
5376
	}
L
Linus Torvalds 已提交
5377

I
Ingo Molnar 已提交
5378
#ifdef CONFIG_SCHED_DEBUG
5379 5380
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5381
#endif
5382
	rcu_read_unlock();
I
Ingo Molnar 已提交
5383 5384 5385
	/*
	 * Only show locks if all tasks are dumped:
	 */
5386
	if (!state_filter)
I
Ingo Molnar 已提交
5387
		debug_show_all_locks();
L
Linus Torvalds 已提交
5388 5389
}

5390 5391 5392
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
I
Ingo Molnar 已提交
5393
 * @cpu: CPU the idle task belongs to
5394 5395 5396 5397
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5398
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5399
{
5400
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5401 5402
	unsigned long flags;

5403 5404
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5405

5406
	__sched_fork(0, idle);
5407
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5408
	idle->se.exec_start = sched_clock();
5409
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5410

5411 5412
	kasan_unpoison_task_stack(idle);

5413 5414 5415 5416 5417 5418 5419 5420 5421
#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
5422 5423
	/*
	 * We're having a chicken and egg problem, even though we are
I
Ingo Molnar 已提交
5424
	 * holding rq->lock, the CPU isn't yet set to this CPU so the
5425 5426 5427 5428 5429 5430 5431 5432
	 * 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
5433
	__set_task_cpu(idle, cpu);
5434
	rcu_read_unlock();
L
Linus Torvalds 已提交
5435 5436

	rq->curr = rq->idle = idle;
5437
	idle->on_rq = TASK_ON_RQ_QUEUED;
5438
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5439
	idle->on_cpu = 1;
5440
#endif
5441 5442
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5443 5444

	/* Set the preempt count _outside_ the spinlocks! */
5445
	init_idle_preempt_count(idle, cpu);
5446

I
Ingo Molnar 已提交
5447 5448 5449 5450
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5451
	ftrace_graph_init_idle_task(idle, cpu);
5452
	vtime_init_idle(idle, cpu);
5453
#ifdef CONFIG_SMP
5454 5455
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5456 5457
}

5458 5459
#ifdef CONFIG_SMP

5460 5461 5462
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
5463
	int ret = 1;
5464

5465 5466 5467
	if (!cpumask_weight(cur))
		return ret;

5468
	ret = dl_cpuset_cpumask_can_shrink(cur, trial);
5469 5470 5471 5472

	return ret;
}

5473 5474 5475 5476 5477 5478 5479
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
5480
	 * to a new cpuset; we don't want to change their CPU
5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492
	 * 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,
5493 5494
					      cs_cpus_allowed))
		ret = dl_task_can_attach(p, cs_cpus_allowed);
5495 5496 5497 5498 5499

out:
	return ret;
}

5500
bool sched_smp_initialized __read_mostly;
5501

5502 5503 5504 5505 5506 5507 5508 5509 5510 5511
#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;

5512
	if (!cpumask_test_cpu(target_cpu, &p->cpus_allowed))
5513 5514 5515 5516
		return -EINVAL;

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

5517
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5518 5519
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5520 5521 5522 5523 5524 5525 5526

/*
 * 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)
{
5527
	bool queued, running;
5528 5529
	struct rq_flags rf;
	struct rq *rq;
5530

5531
	rq = task_rq_lock(p, &rf);
5532
	queued = task_on_rq_queued(p);
5533 5534
	running = task_current(rq, p);

5535
	if (queued)
5536
		dequeue_task(rq, p, DEQUEUE_SAVE);
5537
	if (running)
5538
		put_prev_task(rq, p);
5539 5540 5541

	p->numa_preferred_nid = nid;

5542
	if (queued)
5543
		enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK);
5544
	if (running)
5545
		set_curr_task(rq, p);
5546
	task_rq_unlock(rq, p, &rf);
5547
}
P
Peter Zijlstra 已提交
5548
#endif /* CONFIG_NUMA_BALANCING */
5549

L
Linus Torvalds 已提交
5550
#ifdef CONFIG_HOTPLUG_CPU
5551
/*
I
Ingo Molnar 已提交
5552
 * Ensure that the idle task is using init_mm right before its CPU goes
5553
 * offline.
5554
 */
5555
void idle_task_exit(void)
L
Linus Torvalds 已提交
5556
{
5557
	struct mm_struct *mm = current->active_mm;
5558

5559
	BUG_ON(cpu_online(smp_processor_id()));
5560

5561
	if (mm != &init_mm) {
5562
		switch_mm(mm, &init_mm, current);
5563
		current->active_mm = &init_mm;
5564 5565
		finish_arch_post_lock_switch();
	}
5566
	mmdrop(mm);
L
Linus Torvalds 已提交
5567 5568 5569
}

/*
5570 5571
 * 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
5572 5573 5574
 * 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.
5575 5576
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5577
 */
5578
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5579
{
5580
	long delta = calc_load_fold_active(rq, 1);
5581 5582
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5583 5584
}

5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600
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,
};

5601
/*
5602 5603 5604 5605 5606 5607
 * 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
5608
 */
5609
static void migrate_tasks(struct rq *dead_rq, struct rq_flags *rf)
L
Linus Torvalds 已提交
5610
{
5611
	struct rq *rq = dead_rq;
5612
	struct task_struct *next, *stop = rq->stop;
5613
	struct rq_flags orf = *rf;
5614
	int dest_cpu;
L
Linus Torvalds 已提交
5615 5616

	/*
5617 5618 5619 5620 5621 5622 5623
	 * 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
5624
	 */
5625
	rq->stop = NULL;
5626

5627 5628 5629 5630 5631 5632 5633
	/*
	 * 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);

5634
	for (;;) {
5635 5636
		/*
		 * There's this thread running, bail when that's the only
I
Ingo Molnar 已提交
5637
		 * remaining thread:
5638 5639
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5640
			break;
5641

5642
		/*
I
Ingo Molnar 已提交
5643
		 * pick_next_task() assumes pinned rq->lock:
5644
		 */
5645
		next = pick_next_task(rq, &fake_task, rf);
5646
		BUG_ON(!next);
V
Viresh Kumar 已提交
5647
		put_prev_task(rq, next);
5648

W
Wanpeng Li 已提交
5649 5650 5651 5652 5653 5654 5655 5656 5657
		/*
		 * 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.
		 */
5658
		rq_unlock(rq, rf);
W
Wanpeng Li 已提交
5659
		raw_spin_lock(&next->pi_lock);
5660
		rq_relock(rq, rf);
W
Wanpeng Li 已提交
5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671

		/*
		 * 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;
		}

5672
		/* Find suitable destination for @next, with force if needed. */
5673
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5674
		rq = __migrate_task(rq, rf, next, dest_cpu);
5675
		if (rq != dead_rq) {
5676
			rq_unlock(rq, rf);
5677
			rq = dead_rq;
5678 5679
			*rf = orf;
			rq_relock(rq, rf);
5680
		}
W
Wanpeng Li 已提交
5681
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5682
	}
5683

5684
	rq->stop = stop;
5685
}
L
Linus Torvalds 已提交
5686 5687
#endif /* CONFIG_HOTPLUG_CPU */

5688
void set_rq_online(struct rq *rq)
5689 5690 5691 5692
{
	if (!rq->online) {
		const struct sched_class *class;

5693
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5694 5695 5696 5697 5698 5699 5700 5701 5702
		rq->online = 1;

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

5703
void set_rq_offline(struct rq *rq)
5704 5705 5706 5707 5708 5709 5710 5711 5712
{
	if (rq->online) {
		const struct sched_class *class;

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

5713
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5714 5715 5716 5717
		rq->online = 0;
	}
}

5718
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5719
{
5720
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5721

5722 5723 5724
	rq->age_stamp = sched_clock_cpu(cpu);
}

I
Ingo Molnar 已提交
5725 5726 5727 5728
/*
 * used to mark begin/end of suspend/resume:
 */
static int num_cpus_frozen;
5729

L
Linus Torvalds 已提交
5730
/*
5731 5732 5733
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
5734 5735 5736
 *
 * 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
5737
 */
5738
static void cpuset_cpu_active(void)
5739
{
5740
	if (cpuhp_tasks_frozen) {
5741 5742 5743 5744 5745 5746
		/*
		 * 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.
		 */
5747 5748
		partition_sched_domains(1, NULL, NULL);
		if (--num_cpus_frozen)
5749
			return;
5750 5751 5752 5753 5754
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
5755
		cpuset_force_rebuild();
5756
	}
5757
	cpuset_update_active_cpus();
5758
}
5759

5760
static int cpuset_cpu_inactive(unsigned int cpu)
5761
{
5762
	if (!cpuhp_tasks_frozen) {
5763
		if (dl_cpu_busy(cpu))
5764
			return -EBUSY;
5765
		cpuset_update_active_cpus();
5766
	} else {
5767 5768
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
5769
	}
5770
	return 0;
5771 5772
}

5773
int sched_cpu_activate(unsigned int cpu)
5774
{
5775
	struct rq *rq = cpu_rq(cpu);
5776
	struct rq_flags rf;
5777

5778
	set_cpu_active(cpu, true);
5779

5780
	if (sched_smp_initialized) {
5781
		sched_domains_numa_masks_set(cpu);
5782
		cpuset_cpu_active();
5783
	}
5784 5785 5786 5787 5788

	/*
	 * 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
5789
	 *    after all CPUs have been brought up.
5790 5791 5792 5793
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
5794
	rq_lock_irqsave(rq, &rf);
5795 5796 5797 5798
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
5799
	rq_unlock_irqrestore(rq, &rf);
5800 5801 5802

	update_max_interval();

5803
	return 0;
5804 5805
}

5806
int sched_cpu_deactivate(unsigned int cpu)
5807 5808 5809
{
	int ret;

5810
	set_cpu_active(cpu, false);
5811 5812 5813 5814 5815 5816 5817
	/*
	 * 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.
	 */
5818
	synchronize_rcu_mult(call_rcu, call_rcu_sched);
5819 5820 5821 5822 5823 5824 5825 5826

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
5827
	}
5828 5829
	sched_domains_numa_masks_clear(cpu);
	return 0;
5830 5831
}

5832 5833 5834 5835 5836 5837 5838 5839
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();
}

5840 5841 5842
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
5843
	sched_rq_cpu_starting(cpu);
5844
	sched_tick_start(cpu);
5845
	return 0;
5846 5847
}

5848 5849 5850 5851
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
5852
	struct rq_flags rf;
5853 5854 5855

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
5856
	sched_tick_stop(cpu);
5857 5858

	rq_lock_irqsave(rq, &rf);
5859 5860 5861 5862
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
5863
	migrate_tasks(rq, &rf);
5864
	BUG_ON(rq->nr_running != 1);
5865 5866
	rq_unlock_irqrestore(rq, &rf);

5867 5868
	calc_load_migrate(rq);
	update_max_interval();
5869
	nohz_balance_exit_idle(rq);
5870
	hrtick_clear(rq);
5871 5872 5873 5874
	return 0;
}
#endif

P
Peter Zijlstra 已提交
5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890
#ifdef CONFIG_SCHED_SMT
DEFINE_STATIC_KEY_FALSE(sched_smt_present);

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

L
Linus Torvalds 已提交
5891 5892
void __init sched_init_smp(void)
{
5893 5894
	sched_init_numa();

5895 5896
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
I
Ingo Molnar 已提交
5897
	 * CPU masks are stable and all blatant races in the below code cannot
5898 5899
	 * happen.
	 */
5900
	mutex_lock(&sched_domains_mutex);
P
Peter Zijlstra 已提交
5901
	sched_init_domains(cpu_active_mask);
5902
	mutex_unlock(&sched_domains_mutex);
5903

5904
	/* Move init over to a non-isolated CPU */
5905
	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_FLAG_DOMAIN)) < 0)
5906
		BUG();
I
Ingo Molnar 已提交
5907
	sched_init_granularity();
5908

5909
	init_sched_rt_class();
5910
	init_sched_dl_class();
P
Peter Zijlstra 已提交
5911 5912 5913

	sched_init_smt();

5914
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
5915
}
5916 5917 5918

static int __init migration_init(void)
{
5919
	sched_rq_cpu_starting(smp_processor_id());
5920
	return 0;
L
Linus Torvalds 已提交
5921
}
5922 5923
early_initcall(migration_init);

L
Linus Torvalds 已提交
5924 5925 5926
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
5927
	sched_init_granularity();
L
Linus Torvalds 已提交
5928 5929 5930 5931 5932 5933 5934 5935 5936 5937
}
#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);
}

5938
#ifdef CONFIG_CGROUP_SCHED
5939 5940 5941 5942
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
5943
struct task_group root_task_group;
5944
LIST_HEAD(task_groups);
5945 5946 5947

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
5948
#endif
P
Peter Zijlstra 已提交
5949

5950
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
5951
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
5952

L
Linus Torvalds 已提交
5953 5954
void __init sched_init(void)
{
I
Ingo Molnar 已提交
5955
	int i, j;
5956 5957
	unsigned long alloc_size = 0, ptr;

5958
	sched_clock_init();
5959
	wait_bit_init();
5960

5961 5962 5963 5964 5965 5966 5967
#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) {
5968
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
5969 5970

#ifdef CONFIG_FAIR_GROUP_SCHED
5971
		root_task_group.se = (struct sched_entity **)ptr;
5972 5973
		ptr += nr_cpu_ids * sizeof(void **);

5974
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
5975
		ptr += nr_cpu_ids * sizeof(void **);
5976

5977
#endif /* CONFIG_FAIR_GROUP_SCHED */
5978
#ifdef CONFIG_RT_GROUP_SCHED
5979
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
5980 5981
		ptr += nr_cpu_ids * sizeof(void **);

5982
		root_task_group.rt_rq = (struct rt_rq **)ptr;
5983 5984
		ptr += nr_cpu_ids * sizeof(void **);

5985
#endif /* CONFIG_RT_GROUP_SCHED */
5986
	}
5987
#ifdef CONFIG_CPUMASK_OFFSTACK
5988 5989 5990
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5991 5992
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
5993
	}
5994
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
5995

I
Ingo Molnar 已提交
5996 5997
	init_rt_bandwidth(&def_rt_bandwidth, global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth, global_rt_period(), global_rt_runtime());
5998

G
Gregory Haskins 已提交
5999 6000 6001 6002
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

6003
#ifdef CONFIG_RT_GROUP_SCHED
6004
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
6005
			global_rt_period(), global_rt_runtime());
6006
#endif /* CONFIG_RT_GROUP_SCHED */
6007

D
Dhaval Giani 已提交
6008
#ifdef CONFIG_CGROUP_SCHED
6009 6010
	task_group_cache = KMEM_CACHE(task_group, 0);

6011 6012
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
6013
	INIT_LIST_HEAD(&root_task_group.siblings);
6014
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
6015
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
6016

6017
	for_each_possible_cpu(i) {
6018
		struct rq *rq;
L
Linus Torvalds 已提交
6019 6020

		rq = cpu_rq(i);
6021
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
6022
		rq->nr_running = 0;
6023 6024
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
6025
		init_cfs_rq(&rq->cfs);
6026 6027
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
6028
#ifdef CONFIG_FAIR_GROUP_SCHED
6029
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
6030
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
6031
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
6032
		/*
I
Ingo Molnar 已提交
6033
		 * How much CPU bandwidth does root_task_group get?
D
Dhaval Giani 已提交
6034 6035
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
I
Ingo Molnar 已提交
6036 6037
		 * gets 100% of the CPU resources in the system. This overall
		 * system CPU resource is divided among the tasks of
6038
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
6039 6040 6041
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
6042
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
6043
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
I
Ingo Molnar 已提交
6044
		 * then A0's share of the CPU resource is:
D
Dhaval Giani 已提交
6045
		 *
6046
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
6047
		 *
6048 6049
		 * 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
6050
		 */
6051
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
6052
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
6053 6054 6055
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
6056
#ifdef CONFIG_RT_GROUP_SCHED
6057
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
6058
#endif
L
Linus Torvalds 已提交
6059

I
Ingo Molnar 已提交
6060 6061
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
6062

L
Linus Torvalds 已提交
6063
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
6064
		rq->sd = NULL;
G
Gregory Haskins 已提交
6065
		rq->rd = NULL;
6066
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
6067
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
6068
		rq->active_balance = 0;
I
Ingo Molnar 已提交
6069
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
6070
		rq->push_cpu = 0;
6071
		rq->cpu = i;
6072
		rq->online = 0;
6073 6074
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
6075
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
6076 6077 6078

		INIT_LIST_HEAD(&rq->cfs_tasks);

6079
		rq_attach_root(rq, &def_root_domain);
6080
#ifdef CONFIG_NO_HZ_COMMON
6081
		rq->last_load_update_tick = jiffies;
6082
		rq->last_blocked_load_update_tick = jiffies;
6083
		atomic_set(&rq->nohz_flags, 0);
6084
#endif
6085
#endif /* CONFIG_SMP */
6086
		hrtick_rq_init(rq);
L
Linus Torvalds 已提交
6087 6088 6089
		atomic_set(&rq->nr_iowait, 0);
	}

6090
	set_load_weight(&init_task, false);
6091

L
Linus Torvalds 已提交
6092 6093 6094
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
V
Vegard Nossum 已提交
6095
	mmgrab(&init_mm);
L
Linus Torvalds 已提交
6096 6097 6098 6099 6100 6101 6102 6103 6104
	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());
6105 6106 6107

	calc_load_update = jiffies + LOAD_FREQ;

6108
#ifdef CONFIG_SMP
6109
	idle_thread_set_boot_cpu();
6110
	set_cpu_rq_start_time(smp_processor_id());
6111 6112
#endif
	init_sched_fair_class();
6113

6114 6115
	init_schedstats();

6116
	scheduler_running = 1;
L
Linus Torvalds 已提交
6117 6118
}

6119
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
6120 6121
static inline int preempt_count_equals(int preempt_offset)
{
6122
	int nested = preempt_count() + rcu_preempt_depth();
6123

A
Arnd Bergmann 已提交
6124
	return (nested == preempt_offset);
6125 6126
}

6127
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
6128
{
P
Peter Zijlstra 已提交
6129 6130 6131 6132 6133
	/*
	 * 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.
	 */
6134
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
6135 6136 6137 6138
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
6139
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
6140

6141 6142 6143 6144 6145
	___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
6146
{
I
Ingo Molnar 已提交
6147 6148 6149
	/* Ratelimiting timestamp: */
	static unsigned long prev_jiffy;

6150
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
6151

I
Ingo Molnar 已提交
6152 6153 6154
	/* WARN_ON_ONCE() by default, no rate limit required: */
	rcu_sleep_check();

6155 6156
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
6157 6158
	    system_state == SYSTEM_BOOTING || system_state > SYSTEM_RUNNING ||
	    oops_in_progress)
I
Ingo Molnar 已提交
6159
		return;
6160

I
Ingo Molnar 已提交
6161 6162 6163 6164
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

I
Ingo Molnar 已提交
6165
	/* Save this before calling printk(), since that will clobber it: */
6166 6167
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
6168 6169 6170 6171 6172 6173 6174
	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
6175

6176 6177 6178
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
6179 6180 6181
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
6182 6183
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
6184
		pr_err("Preemption disabled at:");
6185
		print_ip_sym(preempt_disable_ip);
6186 6187
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
6188
	dump_stack();
6189
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
6190
}
6191
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
6192 6193 6194
#endif

#ifdef CONFIG_MAGIC_SYSRQ
6195
void normalize_rt_tasks(void)
6196
{
6197
	struct task_struct *g, *p;
6198 6199 6200
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
6201

6202
	read_lock(&tasklist_lock);
6203
	for_each_process_thread(g, p) {
6204 6205 6206
		/*
		 * Only normalize user tasks:
		 */
6207
		if (p->flags & PF_KTHREAD)
6208 6209
			continue;

6210 6211 6212 6213
		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
6214

6215
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
6216 6217 6218 6219
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
6220
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
6221
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
6222
			continue;
I
Ingo Molnar 已提交
6223
		}
L
Linus Torvalds 已提交
6224

6225
		__sched_setscheduler(p, &attr, false, false);
6226
	}
6227
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
6228 6229 6230
}

#endif /* CONFIG_MAGIC_SYSRQ */
6231

6232
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
6233
/*
6234
 * These functions are only useful for the IA64 MCA handling, or kdb.
6235 6236 6237 6238 6239 6240 6241 6242 6243
 *
 * 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
6244
 * curr_task - return the current task for a given CPU.
6245 6246 6247
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
6248 6249
 *
 * Return: The current task for @cpu.
6250
 */
6251
struct task_struct *curr_task(int cpu)
6252 6253 6254 6255
{
	return cpu_curr(cpu);
}

6256 6257 6258
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
6259
/**
I
Ingo Molnar 已提交
6260
 * set_curr_task - set the current task for a given CPU.
6261 6262 6263 6264
 * @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
6265
 * are serviced on a separate stack. It allows the architecture to switch the
I
Ingo Molnar 已提交
6266
 * notion of the current task on a CPU in a non-blocking manner. This function
6267 6268 6269 6270 6271 6272 6273
 * 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!
 */
6274
void ia64_set_curr_task(int cpu, struct task_struct *p)
6275 6276 6277 6278 6279
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
6280

D
Dhaval Giani 已提交
6281
#ifdef CONFIG_CGROUP_SCHED
6282 6283 6284
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

6285
static void sched_free_group(struct task_group *tg)
6286 6287 6288
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
6289
	autogroup_free(tg);
6290
	kmem_cache_free(task_group_cache, tg);
6291 6292 6293
}

/* allocate runqueue etc for a new task group */
6294
struct task_group *sched_create_group(struct task_group *parent)
6295 6296 6297
{
	struct task_group *tg;

6298
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
6299 6300 6301
	if (!tg)
		return ERR_PTR(-ENOMEM);

6302
	if (!alloc_fair_sched_group(tg, parent))
6303 6304
		goto err;

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

6308 6309 6310
	return tg;

err:
6311
	sched_free_group(tg);
6312 6313 6314 6315 6316 6317 6318
	return ERR_PTR(-ENOMEM);
}

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

6319
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6320
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
6321

I
Ingo Molnar 已提交
6322 6323
	/* Root should already exist: */
	WARN_ON(!parent);
P
Peter Zijlstra 已提交
6324 6325 6326

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
6327
	list_add_rcu(&tg->siblings, &parent->children);
6328
	spin_unlock_irqrestore(&task_group_lock, flags);
6329 6330

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
6331 6332
}

6333
/* rcu callback to free various structures associated with a task group */
6334
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
6335
{
I
Ingo Molnar 已提交
6336
	/* Now it should be safe to free those cfs_rqs: */
6337
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
6338 6339
}

6340
void sched_destroy_group(struct task_group *tg)
6341
{
I
Ingo Molnar 已提交
6342
	/* Wait for possible concurrent references to cfs_rqs complete: */
6343
	call_rcu(&tg->rcu, sched_free_group_rcu);
6344 6345 6346
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
6347
{
6348
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
6349

I
Ingo Molnar 已提交
6350
	/* End participation in shares distribution: */
6351
	unregister_fair_sched_group(tg);
6352 6353

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
6354
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
6355
	list_del_rcu(&tg->siblings);
6356
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
6357 6358
}

6359
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
6360
{
P
Peter Zijlstra 已提交
6361
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
6362

6363 6364 6365 6366 6367 6368
	/*
	 * 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
6369 6370 6371 6372
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
6373
#ifdef CONFIG_FAIR_GROUP_SCHED
6374 6375
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
6376
	else
P
Peter Zijlstra 已提交
6377
#endif
6378
		set_task_rq(tsk, task_cpu(tsk));
6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389
}

/*
 * 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)
{
6390 6391
	int queued, running, queue_flags =
		DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK;
6392 6393 6394 6395
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);
6396
	update_rq_clock(rq);
6397 6398 6399 6400 6401

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

	if (queued)
6402
		dequeue_task(rq, tsk, queue_flags);
6403
	if (running)
6404 6405 6406
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
6407

6408
	if (queued)
6409
		enqueue_task(rq, tsk, queue_flags);
6410
	if (running)
6411
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
6412

6413
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
6414
}
6415

6416
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
6417
{
6418
	return css ? container_of(css, struct task_group, css) : NULL;
6419 6420
}

6421 6422
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
6423
{
6424 6425
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
6426

6427
	if (!parent) {
6428
		/* This is early initialization for the top cgroup */
6429
		return &root_task_group.css;
6430 6431
	}

6432
	tg = sched_create_group(parent);
6433 6434 6435 6436 6437 6438
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449
/* 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;
}

6450
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
6451
{
6452
	struct task_group *tg = css_tg(css);
6453

6454
	sched_offline_group(tg);
6455 6456
}

6457
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
6458
{
6459
	struct task_group *tg = css_tg(css);
6460

6461 6462 6463 6464
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
6465 6466
}

6467 6468 6469 6470
/*
 * 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.
 */
6471
static void cpu_cgroup_fork(struct task_struct *task)
6472
{
6473 6474 6475 6476 6477
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

6478
	update_rq_clock(rq);
6479 6480 6481
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
6482 6483
}

6484
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
6485
{
6486
	struct task_struct *task;
6487
	struct cgroup_subsys_state *css;
6488
	int ret = 0;
6489

6490
	cgroup_taskset_for_each(task, css, tset) {
6491
#ifdef CONFIG_RT_GROUP_SCHED
6492
		if (!sched_rt_can_attach(css_tg(css), task))
6493
			return -EINVAL;
6494
#else
6495 6496 6497
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
6498
#endif
6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514
		/*
		 * 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;
6515
	}
6516
	return ret;
6517
}
6518

6519
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
6520
{
6521
	struct task_struct *task;
6522
	struct cgroup_subsys_state *css;
6523

6524
	cgroup_taskset_for_each(task, css, tset)
6525
		sched_move_task(task);
6526 6527
}

6528
#ifdef CONFIG_FAIR_GROUP_SCHED
6529 6530
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
6531
{
6532
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
6533 6534
}

6535 6536
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
6537
{
6538
	struct task_group *tg = css_tg(css);
6539

6540
	return (u64) scale_load_down(tg->shares);
6541
}
6542 6543

#ifdef CONFIG_CFS_BANDWIDTH
6544 6545
static DEFINE_MUTEX(cfs_constraints_mutex);

6546 6547 6548
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

6549 6550
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

6551 6552
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
6553
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
6554
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574

	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;

6575 6576 6577 6578 6579
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
6580 6581 6582 6583 6584
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

6585
	runtime_enabled = quota != RUNTIME_INF;
6586
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
6587 6588 6589 6590 6591 6592
	/*
	 * 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();
6593 6594 6595
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
6596

P
Paul Turner 已提交
6597
	__refill_cfs_bandwidth_runtime(cfs_b);
I
Ingo Molnar 已提交
6598 6599

	/* Restart the period timer (if active) to handle new period expiry: */
P
Peter Zijlstra 已提交
6600 6601
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
I
Ingo Molnar 已提交
6602

6603 6604
	raw_spin_unlock_irq(&cfs_b->lock);

6605
	for_each_online_cpu(i) {
6606
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
6607
		struct rq *rq = cfs_rq->rq;
6608
		struct rq_flags rf;
6609

6610
		rq_lock_irq(rq, &rf);
6611
		cfs_rq->runtime_enabled = runtime_enabled;
6612
		cfs_rq->runtime_remaining = 0;
6613

6614
		if (cfs_rq->throttled)
6615
			unthrottle_cfs_rq(cfs_rq);
6616
		rq_unlock_irq(rq, &rf);
6617
	}
6618 6619
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
6620 6621
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
6622
	put_online_cpus();
6623

6624
	return ret;
6625 6626 6627 6628 6629 6630
}

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

6631
	period = ktime_to_ns(tg->cfs_bandwidth.period);
6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643
	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;

6644
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
6645 6646
		return -1;

6647
	quota_us = tg->cfs_bandwidth.quota;
6648 6649 6650 6651 6652 6653 6654 6655 6656 6657
	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;
6658
	quota = tg->cfs_bandwidth.quota;
6659 6660 6661 6662 6663 6664 6665 6666

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

6667
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
6668 6669 6670 6671 6672
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

6673 6674
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
6675
{
6676
	return tg_get_cfs_quota(css_tg(css));
6677 6678
}

6679 6680
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
6681
{
6682
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
6683 6684
}

6685 6686
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6687
{
6688
	return tg_get_cfs_period(css_tg(css));
6689 6690
}

6691 6692
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
6693
{
6694
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
6695 6696
}

6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728
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;
6729
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6730 6731 6732 6733 6734
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
6735
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
6736 6737

		quota = normalize_cfs_quota(tg, d);
6738
		parent_quota = parent_b->hierarchical_quota;
6739 6740

		/*
6741 6742
		 * Ensure max(child_quota) <= parent_quota.  On cgroup2,
		 * always take the min.  On cgroup1, only inherit when no
I
Ingo Molnar 已提交
6743
		 * limit is set:
6744
		 */
6745 6746 6747 6748 6749 6750 6751 6752
		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;
		}
6753
	}
6754
	cfs_b->hierarchical_quota = quota;
6755 6756 6757 6758 6759 6760

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
6761
	int ret;
6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772
	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);
	}

6773 6774 6775 6776 6777
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
6778
}
6779

6780
static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
6781
{
6782
	struct task_group *tg = css_tg(seq_css(sf));
6783
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
6784

6785 6786 6787
	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);
6788 6789 6790

	return 0;
}
6791
#endif /* CONFIG_CFS_BANDWIDTH */
6792
#endif /* CONFIG_FAIR_GROUP_SCHED */
6793

6794
#ifdef CONFIG_RT_GROUP_SCHED
6795 6796
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
6797
{
6798
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
6799 6800
}

6801 6802
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
6803
{
6804
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
6805
}
6806

6807 6808
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
6809
{
6810
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
6811 6812
}

6813 6814
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
6815
{
6816
	return sched_group_rt_period(css_tg(css));
6817
}
6818
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
6819

6820
static struct cftype cpu_legacy_files[] = {
6821
#ifdef CONFIG_FAIR_GROUP_SCHED
6822 6823
	{
		.name = "shares",
6824 6825
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
6826
	},
6827
#endif
6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838
#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,
	},
6839 6840
	{
		.name = "stat",
6841
		.seq_show = cpu_cfs_stat_show,
6842
	},
6843
#endif
6844
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
6845
	{
P
Peter Zijlstra 已提交
6846
		.name = "rt_runtime_us",
6847 6848
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
6849
	},
6850 6851
	{
		.name = "rt_period_us",
6852 6853
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
6854
	},
6855
#endif
I
Ingo Molnar 已提交
6856
	{ }	/* Terminate */
6857 6858
};

6859 6860
static int cpu_extra_stat_show(struct seq_file *sf,
			       struct cgroup_subsys_state *css)
6861 6862 6863
{
#ifdef CONFIG_CFS_BANDWIDTH
	{
6864
		struct task_group *tg = css_tg(css);
6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930
		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;
6931
	int idx;
6932 6933 6934 6935

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

6936 6937 6938 6939
	idx = NICE_TO_PRIO(nice) - MAX_RT_PRIO;
	idx = array_index_nospec(idx, 40);
	weight = sched_prio_to_weight[idx];

6940 6941 6942 6943 6944 6945 6946 6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957 6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991 6992 6993 6994 6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025
	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 */
};

7026
struct cgroup_subsys cpu_cgrp_subsys = {
7027
	.css_alloc	= cpu_cgroup_css_alloc,
7028
	.css_online	= cpu_cgroup_css_online,
7029
	.css_released	= cpu_cgroup_css_released,
7030
	.css_free	= cpu_cgroup_css_free,
7031
	.css_extra_stat_show = cpu_extra_stat_show,
7032
	.fork		= cpu_cgroup_fork,
7033 7034
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
7035
	.legacy_cftypes	= cpu_legacy_files,
7036
	.dfl_cftypes	= cpu_files,
7037
	.early_init	= true,
7038
	.threaded	= true,
7039 7040
};

7041
#endif	/* CONFIG_CGROUP_SCHED */
7042

7043 7044 7045 7046 7047
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060 7061 7062 7063 7064 7065 7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088

/*
 * 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,
};
7089 7090

#undef CREATE_TRACE_POINTS