core.c 216.5 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2
 *  kernel/sched/core.c
L
Linus Torvalds 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 *
 *  Kernel scheduler and related syscalls
 *
 *  Copyright (C) 1991-2002  Linus Torvalds
 *
 *  1996-12-23  Modified by Dave Grothe to fix bugs in semaphores and
 *		make semaphores SMP safe
 *  1998-11-19	Implemented schedule_timeout() and related stuff
 *		by Andrea Arcangeli
 *  2002-01-04	New ultra-scalable O(1) scheduler by Ingo Molnar:
 *		hybrid priority-list and round-robin design with
 *		an array-switch method of distributing timeslices
 *		and per-CPU runqueues.  Cleanups and useful suggestions
 *		by Davide Libenzi, preemptible kernel bits by Robert Love.
 *  2003-09-03	Interactivity tuning by Con Kolivas.
 *  2004-04-02	Scheduler domains code by Nick Piggin
I
Ingo Molnar 已提交
19 20 21 22 23 24
 *  2007-04-15  Work begun on replacing all interactivity tuning with a
 *              fair scheduling design by Con Kolivas.
 *  2007-05-05  Load balancing (smp-nice) and other improvements
 *              by Peter Williams
 *  2007-05-06  Interactivity improvements to CFS by Mike Galbraith
 *  2007-07-01  Group scheduling enhancements by Srivatsa Vaddagiri
25 26
 *  2007-11-29  RT balancing improvements by Steven Rostedt, Gregory Haskins,
 *              Thomas Gleixner, Mike Kravetz
L
Linus Torvalds 已提交
27 28
 */

29
#include <linux/kasan.h>
L
Linus Torvalds 已提交
30 31 32 33
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/nmi.h>
#include <linux/init.h>
34
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
35
#include <linux/highmem.h>
36
#include <linux/mmu_context.h>
L
Linus Torvalds 已提交
37
#include <linux/interrupt.h>
38
#include <linux/capability.h>
L
Linus Torvalds 已提交
39 40
#include <linux/completion.h>
#include <linux/kernel_stat.h>
41
#include <linux/debug_locks.h>
42
#include <linux/perf_event.h>
L
Linus Torvalds 已提交
43 44 45
#include <linux/security.h>
#include <linux/notifier.h>
#include <linux/profile.h>
46
#include <linux/freezer.h>
47
#include <linux/vmalloc.h>
L
Linus Torvalds 已提交
48 49
#include <linux/blkdev.h>
#include <linux/delay.h>
50
#include <linux/pid_namespace.h>
L
Linus Torvalds 已提交
51 52 53 54 55 56 57
#include <linux/smp.h>
#include <linux/threads.h>
#include <linux/timer.h>
#include <linux/rcupdate.h>
#include <linux/cpu.h>
#include <linux/cpuset.h>
#include <linux/percpu.h>
58
#include <linux/proc_fs.h>
L
Linus Torvalds 已提交
59
#include <linux/seq_file.h>
60
#include <linux/sysctl.h>
L
Linus Torvalds 已提交
61 62
#include <linux/syscalls.h>
#include <linux/times.h>
63
#include <linux/tsacct_kern.h>
64
#include <linux/kprobes.h>
65
#include <linux/delayacct.h>
66
#include <linux/unistd.h>
J
Jens Axboe 已提交
67
#include <linux/pagemap.h>
P
Peter Zijlstra 已提交
68
#include <linux/hrtimer.h>
R
Reynes Philippe 已提交
69
#include <linux/tick.h>
P
Peter Zijlstra 已提交
70
#include <linux/ctype.h>
71
#include <linux/ftrace.h>
72
#include <linux/slab.h>
73
#include <linux/init_task.h>
74
#include <linux/context_tracking.h>
75
#include <linux/compiler.h>
76
#include <linux/frame.h>
77
#include <linux/prefetch.h>
L
Linus Torvalds 已提交
78

79
#include <asm/switch_to.h>
80
#include <asm/tlb.h>
81
#include <asm/irq_regs.h>
82
#include <asm/mutex.h>
G
Glauber Costa 已提交
83 84 85
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
#endif
L
Linus Torvalds 已提交
86

87
#include "sched.h"
88
#include "../workqueue_internal.h"
89
#include "../smpboot.h"
90

91
#define CREATE_TRACE_POINTS
92
#include <trace/events/sched.h>
93

94 95
DEFINE_MUTEX(sched_domains_mutex);
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
96

97
static void update_rq_clock_task(struct rq *rq, s64 delta);
98

99
void update_rq_clock(struct rq *rq)
100
{
101
	s64 delta;
102

103 104 105
	lockdep_assert_held(&rq->lock);

	if (rq->clock_skip_update & RQCF_ACT_SKIP)
106
		return;
107

108
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
109 110
	if (delta < 0)
		return;
111 112
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
113 114
}

I
Ingo Molnar 已提交
115 116 117
/*
 * Debugging: various feature bits
 */
P
Peter Zijlstra 已提交
118 119 120 121

#define SCHED_FEAT(name, enabled)	\
	(1UL << __SCHED_FEAT_##name) * enabled |

I
Ingo Molnar 已提交
122
const_debug unsigned int sysctl_sched_features =
123
#include "features.h"
P
Peter Zijlstra 已提交
124 125 126 127
	0;

#undef SCHED_FEAT

128 129 130 131 132 133
/*
 * 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;

134 135 136 137 138 139 140 141
/*
 * 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
142
/*
P
Peter Zijlstra 已提交
143
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
144 145
 * default: 1s
 */
P
Peter Zijlstra 已提交
146
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
147

148
__read_mostly int scheduler_running;
149

P
Peter Zijlstra 已提交
150 151 152 153 154
/*
 * 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
155

156 157 158
/* cpus with isolated domains */
cpumask_var_t cpu_isolated_map;

L
Linus Torvalds 已提交
159
/*
160
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
161
 */
A
Alexey Dobriyan 已提交
162
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
163 164
	__acquires(rq->lock)
{
165
	struct rq *rq;
L
Linus Torvalds 已提交
166 167 168

	local_irq_disable();
	rq = this_rq();
169
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
170 171 172 173

	return rq;
}

174 175 176
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
177
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
178 179 180 181 182 183 184 185 186 187
	__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))) {
188
			rf->cookie = lockdep_pin_lock(&rq->lock);
189 190 191 192 193 194 195 196 197 198 199 200
			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.
 */
201
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
202 203 204 205 206 207
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
208
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
		rq = task_rq(p);
		raw_spin_lock(&rq->lock);
		/*
		 *	move_queued_task()		task_rq_lock()
		 *
		 *	ACQUIRE (rq->lock)
		 *	[S] ->on_rq = MIGRATING		[L] rq = task_rq()
		 *	WMB (__set_task_cpu())		ACQUIRE (rq->lock);
		 *	[S] ->cpu = new_cpu		[L] task_rq()
		 *					[L] ->on_rq
		 *	RELEASE (rq->lock)
		 *
		 * If we observe the old cpu in task_rq_lock, the acquire of
		 * the old rq->lock will fully serialize against the stores.
		 *
		 * If we observe the new cpu in task_rq_lock, the acquire will
		 * pair with the WMB to ensure we must then also see migrating.
		 */
		if (likely(rq == task_rq(p) && !task_on_rq_migrating(p))) {
228
			rf->cookie = lockdep_pin_lock(&rq->lock);
229 230 231
			return rq;
		}
		raw_spin_unlock(&rq->lock);
232
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
233 234 235 236 237 238

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

P
Peter Zijlstra 已提交
239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259
#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);

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

260
	raw_spin_lock(&rq->lock);
261
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
262
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
263
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
264 265 266 267

	return HRTIMER_NORESTART;
}

268
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
269

270
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
271 272 273
{
	struct hrtimer *timer = &rq->hrtick_timer;

274
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
275 276
}

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

284
	raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
285
	__hrtick_restart(rq);
286
	rq->hrtick_csd_pending = 0;
287
	raw_spin_unlock(&rq->lock);
288 289
}

290 291 292 293 294
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
295
void hrtick_start(struct rq *rq, u64 delay)
296
{
297
	struct hrtimer *timer = &rq->hrtick_timer;
298 299 300 301 302 303 304 305 306
	ktime_t time;
	s64 delta;

	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense and can cause timer DoS.
	 */
	delta = max_t(s64, delay, 10000LL);
	time = ktime_add_ns(timer->base->get_time(), delta);
307

308
	hrtimer_set_expires(timer, time);
309 310

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
311
		__hrtick_restart(rq);
312
	} else if (!rq->hrtick_csd_pending) {
313
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
314 315
		rq->hrtick_csd_pending = 1;
	}
316 317
}

318 319 320 321 322 323
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
324
void hrtick_start(struct rq *rq, u64 delay)
325
{
W
Wanpeng Li 已提交
326 327 328 329 330
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
331 332
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
333 334
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
335

336
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
337
{
338 339
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
340

341 342 343 344
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
345

346 347
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
348
}
A
Andrew Morton 已提交
349
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
350 351 352 353 354 355 356
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}
A
Andrew Morton 已提交
357
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
358

359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
/*
 * cmpxchg based fetch_or, macro so it works for different integer types
 */
#define fetch_or(ptr, mask)						\
	({								\
		typeof(ptr) _ptr = (ptr);				\
		typeof(mask) _mask = (mask);				\
		typeof(*_ptr) _old, _val = *_ptr;			\
									\
		for (;;) {						\
			_old = cmpxchg(_ptr, _val, _val | _mask);	\
			if (_old == _val)				\
				break;					\
			_val = _old;					\
		}							\
	_old;								\
})

377
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
378 379 380 381 382 383 384 385 386 387
/*
 * Atomically set TIF_NEED_RESCHED and test for TIF_POLLING_NRFLAG,
 * this avoids any races wrt polling state changes and thereby avoids
 * spurious IPIs.
 */
static bool set_nr_and_not_polling(struct task_struct *p)
{
	struct thread_info *ti = task_thread_info(p);
	return !(fetch_or(&ti->flags, _TIF_NEED_RESCHED) & _TIF_POLLING_NRFLAG);
}
388 389 390 391 392 393 394 395 396 397

/*
 * Atomically set TIF_NEED_RESCHED if TIF_POLLING_NRFLAG is set.
 *
 * If this returns true, then the idle task promises to call
 * sched_ttwu_pending() and reschedule soon.
 */
static bool set_nr_if_polling(struct task_struct *p)
{
	struct thread_info *ti = task_thread_info(p);
398
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
399 400 401 402 403 404 405 406 407 408 409 410 411 412

	for (;;) {
		if (!(val & _TIF_POLLING_NRFLAG))
			return false;
		if (val & _TIF_NEED_RESCHED)
			return true;
		old = cmpxchg(&ti->flags, val, val | _TIF_NEED_RESCHED);
		if (old == val)
			break;
		val = old;
	}
	return true;
}

413 414 415 416 417 418
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
419 420 421 422 423 424 425

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
426 427
#endif

428 429 430 431 432 433 434 435 436 437
void wake_q_add(struct wake_q_head *head, struct task_struct *task)
{
	struct wake_q_node *node = &task->wake_q;

	/*
	 * Atomically grab the task, if ->wake_q is !nil already it means
	 * its already queued (either by us or someone else) and will get the
	 * wakeup due to that.
	 *
	 * This cmpxchg() implies a full barrier, which pairs with the write
438
	 * barrier implied by the wakeup in wake_up_q().
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
	 */
	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);
		/* task can safely be re-inserted now */
		node = node->next;
		task->wake_q.next = NULL;

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

I
Ingo Molnar 已提交
474
/*
475
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
476 477 478 479 480
 *
 * On UP this means the setting of the need_resched flag, on SMP it
 * might also involve a cross-CPU call to trigger the scheduler on
 * the target CPU.
 */
481
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
482
{
483
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
484 485
	int cpu;

486
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
487

488
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
489 490
		return;

491
	cpu = cpu_of(rq);
492

493
	if (cpu == smp_processor_id()) {
494
		set_tsk_need_resched(curr);
495
		set_preempt_need_resched();
I
Ingo Molnar 已提交
496
		return;
497
	}
I
Ingo Molnar 已提交
498

499
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
500
		smp_send_reschedule(cpu);
501 502
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
503 504
}

505
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
506 507 508 509
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

510
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
511
		return;
512
	resched_curr(rq);
513
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
514
}
515

516
#ifdef CONFIG_SMP
517
#ifdef CONFIG_NO_HZ_COMMON
518 519 520 521 522 523 524 525
/*
 * In the semi idle case, use the nearest busy cpu for migrating timers
 * from an idle cpu.  This is good for power-savings.
 *
 * We don't do similar optimization for completely idle system, as
 * 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).
 */
526
int get_nohz_timer_target(void)
527
{
528
	int i, cpu = smp_processor_id();
529 530
	struct sched_domain *sd;

531
	if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
532 533
		return cpu;

534
	rcu_read_lock();
535
	for_each_domain(cpu, sd) {
536
		for_each_cpu(i, sched_domain_span(sd)) {
537 538 539 540
			if (cpu == i)
				continue;

			if (!idle_cpu(i) && is_housekeeping_cpu(i)) {
541 542 543 544
				cpu = i;
				goto unlock;
			}
		}
545
	}
546 547 548

	if (!is_housekeeping_cpu(cpu))
		cpu = housekeeping_any_cpu();
549 550
unlock:
	rcu_read_unlock();
551 552
	return cpu;
}
553 554 555 556 557 558 559 560 561 562
/*
 * 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.
 */
563
static void wake_up_idle_cpu(int cpu)
564 565 566 567 568 569
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

570
	if (set_nr_and_not_polling(rq->idle))
571
		smp_send_reschedule(cpu);
572 573
	else
		trace_sched_wake_idle_without_ipi(cpu);
574 575
}

576
static bool wake_up_full_nohz_cpu(int cpu)
577
{
578 579 580 581 582 583
	/*
	 * 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.
	 */
584 585
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
586
	if (tick_nohz_full_cpu(cpu)) {
587 588
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
589
			tick_nohz_full_kick_cpu(cpu);
590 591 592 593 594 595
		return true;
	}

	return false;
}

596 597 598 599 600
/*
 * 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.
 */
601 602
void wake_up_nohz_cpu(int cpu)
{
603
	if (!wake_up_full_nohz_cpu(cpu))
604 605 606
		wake_up_idle_cpu(cpu);
}

607
static inline bool got_nohz_idle_kick(void)
608
{
609
	int cpu = smp_processor_id();
610 611 612 613 614 615 616 617 618 619 620 621 622

	if (!test_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu)))
		return false;

	if (idle_cpu(cpu) && !need_resched())
		return true;

	/*
	 * We can't run Idle Load Balance on this CPU for this time so we
	 * cancel it and clear NOHZ_BALANCE_KICK
	 */
	clear_bit(NOHZ_BALANCE_KICK, nohz_flags(cpu));
	return false;
623 624
}

625
#else /* CONFIG_NO_HZ_COMMON */
626

627
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
628
{
629
	return false;
P
Peter Zijlstra 已提交
630 631
}

632
#endif /* CONFIG_NO_HZ_COMMON */
633

634
#ifdef CONFIG_NO_HZ_FULL
635
bool sched_can_stop_tick(struct rq *rq)
636
{
637 638 639 640 641 642
	int fifo_nr_running;

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

643
	/*
644 645
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
646
	 */
647 648 649 650 651
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
652 653
	}

654 655 656 657 658 659 660 661 662 663 664 665 666 667
	/*
	 * 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)
668
		return false;
669

670
	return true;
671 672
}
#endif /* CONFIG_NO_HZ_FULL */
673

674
void sched_avg_update(struct rq *rq)
675
{
676 677
	s64 period = sched_avg_period();

678
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
679 680 681 682 683 684
		/*
		 * 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));
685 686 687
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
688 689
}

690
#endif /* CONFIG_SMP */
691

692 693
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
694
/*
695 696 697 698
 * 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.
699
 */
700
int walk_tg_tree_from(struct task_group *from,
701
			     tg_visitor down, tg_visitor up, void *data)
702 703
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
704
	int ret;
705

706 707
	parent = from;

708
down:
P
Peter Zijlstra 已提交
709 710
	ret = (*down)(parent, data);
	if (ret)
711
		goto out;
712 713 714 715 716 717 718
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
719
	ret = (*up)(parent, data);
720 721
	if (ret || parent == from)
		goto out;
722 723 724 725 726

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
727
out:
P
Peter Zijlstra 已提交
728
	return ret;
729 730
}

731
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
732
{
733
	return 0;
P
Peter Zijlstra 已提交
734
}
735 736
#endif

737 738
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
739 740 741
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
742 743 744
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
745
	if (idle_policy(p->policy)) {
746
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
747
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
748 749
		return;
	}
750

751 752
	load->weight = scale_load(sched_prio_to_weight[prio]);
	load->inv_weight = sched_prio_to_wmult[prio];
753 754
}

755
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
756
{
757
	update_rq_clock(rq);
758 759
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
760
	p->sched_class->enqueue_task(rq, p, flags);
761 762
}

763
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
764
{
765
	update_rq_clock(rq);
766 767
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
768
	p->sched_class->dequeue_task(rq, p, flags);
769 770
}

771
void activate_task(struct rq *rq, struct task_struct *p, int flags)
772 773 774 775
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

776
	enqueue_task(rq, p, flags);
777 778
}

779
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
780 781 782 783
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

784
	dequeue_task(rq, p, flags);
785 786
}

787
static void update_rq_clock_task(struct rq *rq, s64 delta)
788
{
789 790 791 792 793 794 795 796
/*
 * 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
797
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818

	/*
	 * 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;
819 820
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
821
	if (static_key_false((&paravirt_steal_rq_enabled))) {
822 823 824 825 826 827 828 829 830 831 832
		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

833 834
	rq->clock_task += delta;

835
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
836
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
837 838
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
839 840
}

841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
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;
	}
}

871
/*
I
Ingo Molnar 已提交
872
 * __normal_prio - return the priority that is based on the static prio
873 874 875
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
876
	return p->static_prio;
877 878
}

879 880 881 882 883 884 885
/*
 * 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.
 */
886
static inline int normal_prio(struct task_struct *p)
887 888 889
{
	int prio;

890 891 892
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
893 894 895 896 897 898 899 900 901 902 903 904 905
		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.
 */
906
static int effective_prio(struct task_struct *p)
907 908 909 910 911 912 913 914 915 916 917 918
{
	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
919 920 921
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
922 923
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
924
 */
925
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
926 927 928 929
{
	return cpu_curr(task_cpu(p)) == p;
}

930
/*
931 932 933 934 935
 * 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().
936
 */
937 938
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
939
				       int oldprio)
940 941 942
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
943
			prev_class->switched_from(rq, p);
944

P
Peter Zijlstra 已提交
945
		p->sched_class->switched_to(rq, p);
946
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
947
		p->sched_class->prio_changed(rq, p, oldprio);
948 949
}

950
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
951 952 953 954 955 956 957 958 959 960
{
	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) {
961
				resched_curr(rq);
962 963 964 965 966 967 968 969 970
				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.
	 */
971
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
972
		rq_clock_skip_update(rq, true);
973 974
}

L
Linus Torvalds 已提交
975
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994
/*
 * 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.
 */
995
static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
996 997 998 999
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
1000
	dequeue_task(rq, p, 0);
P
Peter Zijlstra 已提交
1001 1002 1003 1004 1005 1006 1007 1008
	set_task_cpu(p, new_cpu);
	raw_spin_unlock(&rq->lock);

	rq = cpu_rq(new_cpu);

	raw_spin_lock(&rq->lock);
	BUG_ON(task_cpu(p) != new_cpu);
	enqueue_task(rq, p, 0);
1009
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028
	check_preempt_curr(rq, p, 0);

	return rq;
}

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

/*
 * Move (not current) task off this cpu, onto dest cpu. We're doing
 * 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.
 */
1029
static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
1030 1031
{
	if (unlikely(!cpu_active(dest_cpu)))
1032
		return rq;
P
Peter Zijlstra 已提交
1033 1034 1035

	/* Affinity changed (again). */
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
1036
		return rq;
P
Peter Zijlstra 已提交
1037

1038 1039 1040
	rq = move_queued_task(rq, p, dest_cpu);

	return rq;
P
Peter Zijlstra 已提交
1041 1042 1043 1044 1045 1046 1047 1048 1049 1050
}

/*
 * 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;
1051 1052
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064

	/*
	 * The original target cpu might have gone down and we might
	 * be on another cpu but it doesn't matter.
	 */
	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();
1065 1066 1067 1068 1069 1070 1071 1072

	raw_spin_lock(&p->pi_lock);
	raw_spin_lock(&rq->lock);
	/*
	 * 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.
	 */
1073 1074 1075 1076 1077 1078
	if (task_rq(p) == rq) {
		if (task_on_rq_queued(p))
			rq = __migrate_task(rq, p, arg->dest_cpu);
		else
			p->wake_cpu = arg->dest_cpu;
	}
1079 1080 1081
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1082 1083 1084 1085
	local_irq_enable();
	return 0;
}

1086 1087 1088 1089 1090
/*
 * 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
1091 1092 1093 1094 1095
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1096 1097
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1098 1099 1100
	struct rq *rq = task_rq(p);
	bool queued, running;

1101
	lockdep_assert_held(&p->pi_lock);
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111

	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);
1112
		dequeue_task(rq, p, DEQUEUE_SAVE);
1113 1114 1115 1116
	}
	if (running)
		put_prev_task(rq, p);

1117
	p->sched_class->set_cpus_allowed(p, new_mask);
1118 1119

	if (queued)
1120
		enqueue_task(rq, p, ENQUEUE_RESTORE);
1121
	if (running)
1122
		set_curr_task(rq, p);
1123 1124
}

P
Peter Zijlstra 已提交
1125 1126 1127 1128 1129 1130 1131 1132 1133
/*
 * 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.
 */
1134 1135
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1136
{
1137
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1138
	unsigned int dest_cpu;
1139 1140
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1141 1142
	int ret = 0;

1143
	rq = task_rq_lock(p, &rf);
P
Peter Zijlstra 已提交
1144

1145 1146 1147 1148 1149 1150 1151
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1152 1153 1154 1155 1156 1157 1158 1159 1160
	/*
	 * 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
1161 1162 1163
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1164
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1165 1166 1167 1168 1169 1170
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1171 1172 1173 1174 1175 1176 1177 1178 1179 1180
	if (p->flags & PF_KTHREAD) {
		/*
		 * For kernel threads that do indeed end up on online &&
		 * !active we want to ensure they are strict per-cpu threads.
		 */
		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
1181 1182 1183 1184
	/* 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;

1185
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1186 1187 1188
	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. */
1189
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1190 1191 1192
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1193 1194 1195 1196 1197
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1198
		lockdep_unpin_lock(&rq->lock, rf.cookie);
1199
		rq = move_queued_task(rq, p, dest_cpu);
1200
		lockdep_repin_lock(&rq->lock, rf.cookie);
1201
	}
P
Peter Zijlstra 已提交
1202
out:
1203
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1204 1205 1206

	return ret;
}
1207 1208 1209 1210 1211

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
1212 1213
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1214
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1215
{
1216 1217 1218 1219 1220
#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
1221
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1222
			!p->on_rq);
1223

1224 1225 1226 1227 1228 1229 1230 1231 1232
	/*
	 * 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)));

1233
#ifdef CONFIG_LOCKDEP
1234 1235 1236 1237 1238
	/*
	 * 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
1239
	 * see task_group().
1240 1241 1242 1243
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1244 1245 1246
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1247 1248
#endif

1249
	trace_sched_migrate_task(p, new_cpu);
1250

1251
	if (task_cpu(p) != new_cpu) {
1252
		if (p->sched_class->migrate_task_rq)
1253
			p->sched_class->migrate_task_rq(p);
1254
		p->se.nr_migrations++;
1255
		perf_event_task_migrate(p);
1256
	}
I
Ingo Molnar 已提交
1257 1258

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1259 1260
}

1261 1262
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1263
	if (task_on_rq_queued(p)) {
1264 1265 1266 1267 1268
		struct rq *src_rq, *dst_rq;

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

1269
		p->on_rq = TASK_ON_RQ_MIGRATING;
1270 1271 1272
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1273
		p->on_rq = TASK_ON_RQ_QUEUED;
1274 1275 1276 1277 1278
		check_preempt_curr(dst_rq, p, 0);
	} 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
L
Leo Yan 已提交
1279
		 * previous cpu our target instead of where it really is.
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
		 */
		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;

1296 1297 1298
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1299 1300 1301
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1302 1303
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1304
	double_rq_lock(src_rq, dst_rq);
1305

1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
	if (task_cpu(arg->dst_task) != arg->dst_cpu)
		goto unlock;

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

	if (!cpumask_test_cpu(arg->dst_cpu, tsk_cpus_allowed(arg->src_task)))
		goto unlock;

	if (!cpumask_test_cpu(arg->src_cpu, tsk_cpus_allowed(arg->dst_task)))
		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);
1325 1326
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348

	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;

1349 1350 1351 1352
	/*
	 * 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.
	 */
1353 1354 1355 1356 1357 1358 1359 1360 1361
	if (!cpu_active(arg.src_cpu) || !cpu_active(arg.dst_cpu))
		goto out;

	if (!cpumask_test_cpu(arg.dst_cpu, tsk_cpus_allowed(arg.src_task)))
		goto out;

	if (!cpumask_test_cpu(arg.src_cpu, tsk_cpus_allowed(arg.dst_task)))
		goto out;

1362
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1363 1364 1365 1366 1367 1368
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1369 1370 1371
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1372 1373 1374 1375 1376 1377 1378
 * 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
1379 1380 1381 1382 1383 1384
 * 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
1385
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1386
{
1387
	int running, queued;
1388
	struct rq_flags rf;
R
Roland McGrath 已提交
1389
	unsigned long ncsw;
1390
	struct rq *rq;
L
Linus Torvalds 已提交
1391

1392 1393 1394 1395 1396 1397 1398 1399
	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);
1400

1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
		/*
		 * 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
1412 1413 1414
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1415
			cpu_relax();
R
Roland McGrath 已提交
1416
		}
1417

1418 1419 1420 1421 1422
		/*
		 * 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.
		 */
1423
		rq = task_rq_lock(p, &rf);
1424
		trace_sched_wait_task(p);
1425
		running = task_running(rq, p);
1426
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1427
		ncsw = 0;
1428
		if (!match_state || p->state == match_state)
1429
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1430
		task_rq_unlock(rq, p, &rf);
1431

R
Roland McGrath 已提交
1432 1433 1434 1435 1436 1437
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1438 1439 1440 1441 1442 1443 1444 1445 1446 1447
		/*
		 * 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;
		}
1448

1449 1450 1451 1452 1453
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1454
		 * So if it was still runnable (but just not actively
1455 1456 1457
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1458
		if (unlikely(queued)) {
1459 1460 1461 1462
			ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1463 1464
			continue;
		}
1465

1466 1467 1468 1469 1470 1471 1472
		/*
		 * 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
1473 1474

	return ncsw;
L
Linus Torvalds 已提交
1475 1476 1477 1478 1479 1480 1481 1482 1483
}

/***
 * 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
1484
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1485 1486 1487 1488 1489
 * 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.
 */
1490
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1491 1492 1493 1494 1495 1496 1497 1498 1499
{
	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
1500
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1501

1502
/*
1503
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522
 *
 * A few notes on cpu_active vs cpu_online:
 *
 *  - cpu_active must be a subset of cpu_online
 *
 *  - on cpu-up we allow per-cpu kthreads on the online && !active cpu,
 *    see __set_cpus_allowed_ptr(). At this point the newly online
 *    cpu isn't yet part of the sched domains, and balancing will not
 *    see it.
 *
 *  - on cpu-down we clear cpu_active() to mask the sched domains and
 *    avoid the load balancer to place new tasks on the to be removed
 *    cpu. Existing tasks will remain running there and will be taken
 *    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.
1523
 */
1524 1525
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1526 1527
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1528 1529
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1530

1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
	/*
	 * 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.
	 */
	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;
			if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
				return dest_cpu;
		}
1546
	}
1547

1548 1549
	for (;;) {
		/* Any allowed, online CPU? */
1550
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1551 1552 1553
			if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu))
				continue;
			if (!cpu_online(dest_cpu))
1554 1555 1556
				continue;
			goto out;
		}
1557

1558
		/* No more Mr. Nice Guy. */
1559 1560
		switch (state) {
		case cpuset:
1561 1562 1563 1564 1565 1566
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
			/* fall-through */
1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585
		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()) {
1586
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1587 1588
					task_pid_nr(p), p->comm, cpu);
		}
1589 1590 1591 1592 1593
	}

	return dest_cpu;
}

1594
/*
1595
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1596
 */
1597
static inline
1598
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1599
{
1600 1601
	lockdep_assert_held(&p->pi_lock);

1602
	if (tsk_nr_cpus_allowed(p) > 1)
1603
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1604 1605
	else
		cpu = cpumask_any(tsk_cpus_allowed(p));
1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616

	/*
	 * 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
	 * cpu.
	 *
	 * 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 ]
	 */
1617
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1618
		     !cpu_online(cpu)))
1619
		cpu = select_fallback_rq(task_cpu(p), p);
1620 1621

	return cpu;
1622
}
1623 1624 1625 1626 1627 1628

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1629 1630 1631 1632 1633 1634 1635 1636 1637

#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
1638
#endif /* CONFIG_SMP */
1639

P
Peter Zijlstra 已提交
1640
static void
1641
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1642
{
1643
	struct rq *rq;
1644

1645 1646 1647 1648
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1649

1650 1651
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1652 1653
		schedstat_inc(rq->ttwu_local);
		schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1654 1655 1656
	} else {
		struct sched_domain *sd;

1657
		schedstat_inc(p->se.statistics.nr_wakeups_remote);
1658
		rcu_read_lock();
1659
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1660
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1661
				schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1662 1663 1664
				break;
			}
		}
1665
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1666
	}
1667 1668

	if (wake_flags & WF_MIGRATED)
1669
		schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1670 1671
#endif /* CONFIG_SMP */

1672 1673
	schedstat_inc(rq->ttwu_count);
	schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1674 1675

	if (wake_flags & WF_SYNC)
1676
		schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1677 1678
}

1679
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1680
{
T
Tejun Heo 已提交
1681
	activate_task(rq, p, en_flags);
1682
	p->on_rq = TASK_ON_RQ_QUEUED;
1683 1684 1685 1686

	/* if a worker is waking up, notify workqueue */
	if (p->flags & PF_WQ_WORKER)
		wq_worker_waking_up(p, cpu_of(rq));
T
Tejun Heo 已提交
1687 1688
}

1689 1690 1691
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1692 1693
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
			   struct pin_cookie cookie)
T
Tejun Heo 已提交
1694 1695 1696
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1697 1698
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1699
#ifdef CONFIG_SMP
1700 1701
	if (p->sched_class->task_woken) {
		/*
1702 1703
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1704
		 */
1705
		lockdep_unpin_lock(&rq->lock, cookie);
T
Tejun Heo 已提交
1706
		p->sched_class->task_woken(rq, p);
1707
		lockdep_repin_lock(&rq->lock, cookie);
1708
	}
T
Tejun Heo 已提交
1709

1710
	if (rq->idle_stamp) {
1711
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1712
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1713

1714 1715 1716
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1717
			rq->avg_idle = max;
1718

T
Tejun Heo 已提交
1719 1720 1721 1722 1723
		rq->idle_stamp = 0;
	}
#endif
}

1724
static void
1725 1726
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
		 struct pin_cookie cookie)
1727
{
1728 1729
	int en_flags = ENQUEUE_WAKEUP;

1730 1731
	lockdep_assert_held(&rq->lock);

1732 1733 1734
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1735 1736

	if (wake_flags & WF_MIGRATED)
1737
		en_flags |= ENQUEUE_MIGRATED;
1738 1739
#endif

1740
	ttwu_activate(rq, p, en_flags);
1741
	ttwu_do_wakeup(rq, p, wake_flags, cookie);
1742 1743 1744 1745 1746 1747 1748 1749 1750 1751
}

/*
 * 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)
{
1752
	struct rq_flags rf;
1753 1754 1755
	struct rq *rq;
	int ret = 0;

1756
	rq = __task_rq_lock(p, &rf);
1757
	if (task_on_rq_queued(p)) {
1758 1759
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1760
		ttwu_do_wakeup(rq, p, wake_flags, rf.cookie);
1761 1762
		ret = 1;
	}
1763
	__task_rq_unlock(rq, &rf);
1764 1765 1766 1767

	return ret;
}

1768
#ifdef CONFIG_SMP
1769
void sched_ttwu_pending(void)
1770 1771
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1772
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1773
	struct pin_cookie cookie;
P
Peter Zijlstra 已提交
1774
	struct task_struct *p;
1775
	unsigned long flags;
1776

1777 1778 1779 1780
	if (!llist)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
1781
	cookie = lockdep_pin_lock(&rq->lock);
1782

P
Peter Zijlstra 已提交
1783
	while (llist) {
P
Peter Zijlstra 已提交
1784 1785
		int wake_flags = 0;

P
Peter Zijlstra 已提交
1786 1787
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
P
Peter Zijlstra 已提交
1788 1789 1790 1791 1792

		if (p->sched_remote_wakeup)
			wake_flags = WF_MIGRATED;

		ttwu_do_activate(rq, p, wake_flags, cookie);
1793 1794
	}

1795
	lockdep_unpin_lock(&rq->lock, cookie);
1796
	raw_spin_unlock_irqrestore(&rq->lock, flags);
1797 1798 1799 1800
}

void scheduler_ipi(void)
{
1801 1802 1803 1804 1805
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1806
	preempt_fold_need_resched();
1807

1808
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824
		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
1825
	sched_ttwu_pending();
1826 1827 1828 1829

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1830
	if (unlikely(got_nohz_idle_kick())) {
1831
		this_rq()->idle_balance = 1;
1832
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1833
	}
1834
	irq_exit();
1835 1836
}

P
Peter Zijlstra 已提交
1837
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1838
{
1839 1840
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1841 1842
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1843 1844 1845 1846 1847 1848
	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);
	}
1849
}
1850

1851 1852 1853 1854 1855
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

1856 1857 1858 1859
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1860 1861 1862 1863 1864 1865 1866 1867 1868 1869

	if (set_nr_if_polling(rq->idle)) {
		trace_sched_wake_idle_without_ipi(cpu);
	} else {
		raw_spin_lock_irqsave(&rq->lock, flags);
		if (is_idle_task(rq->curr))
			smp_send_reschedule(cpu);
		/* Else cpu is not in idle, do nothing here */
		raw_spin_unlock_irqrestore(&rq->lock, flags);
	}
1870 1871 1872

out:
	rcu_read_unlock();
1873 1874
}

1875
bool cpus_share_cache(int this_cpu, int that_cpu)
1876 1877 1878
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1879
#endif /* CONFIG_SMP */
1880

1881
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1882 1883
{
	struct rq *rq = cpu_rq(cpu);
1884
	struct pin_cookie cookie;
1885

1886
#if defined(CONFIG_SMP)
1887
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1888
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
P
Peter Zijlstra 已提交
1889
		ttwu_queue_remote(p, cpu, wake_flags);
1890 1891 1892 1893
		return;
	}
#endif

1894
	raw_spin_lock(&rq->lock);
1895
	cookie = lockdep_pin_lock(&rq->lock);
1896
	ttwu_do_activate(rq, p, wake_flags, cookie);
1897
	lockdep_unpin_lock(&rq->lock, cookie);
1898
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
1899 1900
}

1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950
/*
 * Notes on Program-Order guarantees on SMP systems.
 *
 *  MIGRATION
 *
 * The basic program-order guarantee on SMP systems is that when a task [t]
 * migrates, all its activity on its old cpu [c0] happens-before any subsequent
 * execution on its new cpu [c1].
 *
 * 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.
 * Note: the cpu doing B need not be c0 or c1
 *
 * 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)
1951
 *   2) smp_cond_load_acquire(!X->on_cpu)
1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
 *
 * 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);
 *
1962
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987
 *                    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,
1988
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1989 1990 1991
 *
 */

T
Tejun Heo 已提交
1992
/**
L
Linus Torvalds 已提交
1993
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1994
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1995
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1996
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1997
 *
1998
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
1999
 *
2000 2001 2002 2003 2004 2005 2006
 * 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
2007
 */
2008 2009
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
2010 2011
{
	unsigned long flags;
2012
	int cpu, success = 0;
P
Peter Zijlstra 已提交
2013

2014 2015 2016 2017 2018 2019 2020
	/*
	 * 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.
	 */
	smp_mb__before_spinlock();
2021
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
2022
	if (!(p->state & state))
L
Linus Torvalds 已提交
2023 2024
		goto out;

2025 2026
	trace_sched_waking(p);

2027
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
2028 2029
	cpu = task_cpu(p);

2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
	/*
	 * Ensure we load p->on_rq _after_ p->state, otherwise it would
	 * be possible to, falsely, observe p->on_rq == 0 and get stuck
	 * in smp_cond_load_acquire() below.
	 *
	 * sched_ttwu_pending()                 try_to_wake_up()
	 *   [S] p->on_rq = 1;                  [L] P->state
	 *       UNLOCK rq->lock  -----.
	 *                              \
	 *				 +---   RMB
	 * schedule()                   /
	 *       LOCK rq->lock    -----'
	 *       UNLOCK rq->lock
	 *
	 * [task p]
	 *   [S] p->state = UNINTERRUPTIBLE     [L] p->on_rq
	 *
	 * Pairs with the UNLOCK+LOCK on rq->lock from the
	 * last wakeup of our task and the schedule that got our task
	 * current.
	 */
	smp_rmb();
2052 2053
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2054 2055

#ifdef CONFIG_SMP
2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074
	/*
	 * Ensure we load p->on_cpu _after_ p->on_rq, otherwise it would be
	 * possible to, falsely, observe p->on_cpu == 0.
	 *
	 * One must be running (->on_cpu == 1) in order to remove oneself
	 * from the runqueue.
	 *
	 *  [S] ->on_cpu = 1;	[L] ->on_rq
	 *      UNLOCK rq->lock
	 *			RMB
	 *      LOCK   rq->lock
	 *  [S] ->on_rq = 0;    [L] ->on_cpu
	 *
	 * Pairs with the full barrier implied in the UNLOCK+LOCK on rq->lock
	 * from the consecutive calls to schedule(); the first switching to our
	 * task, the second putting it to sleep.
	 */
	smp_rmb();

P
Peter Zijlstra 已提交
2075
	/*
2076 2077
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
2078 2079 2080 2081 2082
	 *
	 * Pairs with the smp_store_release() in finish_lock_switch().
	 *
	 * This ensures that tasks getting woken will be fully ordered against
	 * their previous state and preserve Program Order.
2083
	 */
2084
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2085

2086
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2087
	p->state = TASK_WAKING;
2088

2089
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2090 2091
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2092
		set_task_cpu(p, cpu);
2093
	}
L
Linus Torvalds 已提交
2094 2095
#endif /* CONFIG_SMP */

2096
	ttwu_queue(p, cpu, wake_flags);
2097
stat:
2098
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2099
out:
2100
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2101 2102 2103 2104

	return success;
}

T
Tejun Heo 已提交
2105 2106 2107
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2108
 * @cookie: context's cookie for pinning
T
Tejun Heo 已提交
2109
 *
2110
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2111
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2112
 * the current task.
T
Tejun Heo 已提交
2113
 */
2114
static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie)
T
Tejun Heo 已提交
2115 2116 2117
{
	struct rq *rq = task_rq(p);

2118 2119 2120 2121
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2122 2123
	lockdep_assert_held(&rq->lock);

2124
	if (!raw_spin_trylock(&p->pi_lock)) {
2125 2126 2127 2128 2129 2130
		/*
		 * 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.
		 */
2131
		lockdep_unpin_lock(&rq->lock, cookie);
2132 2133 2134
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
2135
		lockdep_repin_lock(&rq->lock, cookie);
2136 2137
	}

T
Tejun Heo 已提交
2138
	if (!(p->state & TASK_NORMAL))
2139
		goto out;
T
Tejun Heo 已提交
2140

2141 2142
	trace_sched_waking(p);

2143
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
2144 2145
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

2146
	ttwu_do_wakeup(rq, p, 0, cookie);
2147
	ttwu_stat(p, smp_processor_id(), 0);
2148 2149
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2150 2151
}

2152 2153 2154 2155 2156
/**
 * 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
2157 2158 2159
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2160 2161 2162 2163
 *
 * 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.
 */
2164
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2165
{
2166
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2167 2168 2169
}
EXPORT_SYMBOL(wake_up_process);

2170
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2171 2172 2173 2174
{
	return try_to_wake_up(p, state, 0);
}

2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186
/*
 * This function clears the sched_dl_entity static params.
 */
void __dl_clear_params(struct task_struct *p)
{
	struct sched_dl_entity *dl_se = &p->dl;

	dl_se->dl_runtime = 0;
	dl_se->dl_deadline = 0;
	dl_se->dl_period = 0;
	dl_se->flags = 0;
	dl_se->dl_bw = 0;
2187 2188 2189

	dl_se->dl_throttled = 0;
	dl_se->dl_yielded = 0;
2190 2191
}

L
Linus Torvalds 已提交
2192 2193 2194
/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2195 2196 2197
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2198
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2199
{
P
Peter Zijlstra 已提交
2200 2201 2202
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2203 2204
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2205
	p->se.prev_sum_exec_runtime	= 0;
2206
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2207
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2208
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2209

2210 2211 2212 2213
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2214
#ifdef CONFIG_SCHEDSTATS
2215
	/* Even if schedstat is disabled, there should not be garbage */
2216
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2217
#endif
N
Nick Piggin 已提交
2218

2219
	RB_CLEAR_NODE(&p->dl.rb_node);
2220
	init_dl_task_timer(&p->dl);
2221
	__dl_clear_params(p);
2222

P
Peter Zijlstra 已提交
2223
	INIT_LIST_HEAD(&p->rt.run_list);
2224 2225 2226 2227
	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
2228

2229 2230 2231
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2232 2233 2234

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2235
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2236 2237 2238
		p->mm->numa_scan_seq = 0;
	}

2239 2240 2241 2242 2243
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2244 2245
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2246
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2247
	p->numa_work.next = &p->numa_work;
2248
	p->numa_faults = NULL;
2249 2250
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2251 2252

	p->numa_group = NULL;
2253
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2254 2255
}

2256 2257
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2258
#ifdef CONFIG_NUMA_BALANCING
2259

2260 2261 2262
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2263
		static_branch_enable(&sched_numa_balancing);
2264
	else
2265
		static_branch_disable(&sched_numa_balancing);
2266
}
2267 2268 2269 2270 2271 2272 2273

#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;
2274
	int state = static_branch_likely(&sched_numa_balancing);
2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289

	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
2290

2291 2292
#ifdef CONFIG_SCHEDSTATS

2293
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2294
static bool __initdata __sched_schedstats = false;
2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317

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;

2318 2319 2320 2321 2322
	/*
	 * 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.
	 */
2323
	if (!strcmp(str, "enable")) {
2324
		__sched_schedstats = true;
2325 2326
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2327
		__sched_schedstats = false;
2328 2329 2330 2331 2332 2333 2334 2335 2336 2337
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2338 2339 2340 2341 2342
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362
#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;
}
2363 2364 2365 2366
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2367 2368 2369 2370

/*
 * fork()/clone()-time setup:
 */
2371
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2372
{
2373
	unsigned long flags;
I
Ingo Molnar 已提交
2374 2375
	int cpu = get_cpu();

2376
	__sched_fork(clone_flags, p);
2377
	/*
2378
	 * We mark the process as NEW here. This guarantees that
2379 2380 2381
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2382
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2383

2384 2385 2386 2387 2388
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2389 2390 2391 2392
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2393
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2394
			p->policy = SCHED_NORMAL;
2395
			p->static_prio = NICE_TO_PRIO(0);
2396 2397 2398 2399 2400 2401
			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);
		set_load_weight(p);
2402

2403 2404 2405 2406 2407 2408
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2409

2410 2411 2412 2413 2414 2415
	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
2416
		p->sched_class = &fair_sched_class;
2417
	}
2418

2419
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2420

2421 2422 2423 2424 2425 2426 2427
	/*
	 * 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.
	 */
2428
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2429 2430 2431 2432 2433 2434 2435
	/*
	 * We're setting the cpu for the first time, we don't migrate,
	 * so use __set_task_cpu().
	 */
	__set_task_cpu(p, cpu);
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);
2436
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2437

2438
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2439
	if (likely(sched_info_on()))
2440
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2441
#endif
P
Peter Zijlstra 已提交
2442 2443
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2444
#endif
2445
	init_task_preempt_count(p);
2446
#ifdef CONFIG_SMP
2447
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2448
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2449
#endif
2450

N
Nick Piggin 已提交
2451
	put_cpu();
2452
	return 0;
L
Linus Torvalds 已提交
2453 2454
}

2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473
unsigned long to_ratio(u64 period, u64 runtime)
{
	if (runtime == RUNTIME_INF)
		return 1ULL << 20;

	/*
	 * 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;

	return div64_u64(runtime << 20, period);
}

#ifdef CONFIG_SMP
inline struct dl_bw *dl_bw_of(int i)
{
2474 2475
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2476 2477 2478
	return &cpu_rq(i)->rd->dl_bw;
}

2479
static inline int dl_bw_cpus(int i)
2480
{
2481 2482 2483
	struct root_domain *rd = cpu_rq(i)->rd;
	int cpus = 0;

2484 2485
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2486 2487 2488 2489
	for_each_cpu_and(i, rd->span, cpu_active_mask)
		cpus++;

	return cpus;
2490 2491 2492 2493 2494 2495 2496
}
#else
inline struct dl_bw *dl_bw_of(int i)
{
	return &cpu_rq(i)->dl.dl_bw;
}

2497
static inline int dl_bw_cpus(int i)
2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509
{
	return 1;
}
#endif

/*
 * We must be sure that accepting a new task (or allowing changing the
 * parameters of an existing one) is consistent with the bandwidth
 * constraints. If yes, this function also accordingly updates the currently
 * allocated bandwidth to reflect the new situation.
 *
 * This function is called while holding p's rq->lock.
2510 2511 2512
 *
 * XXX we should delay bw change until the task's 0-lag point, see
 * __setparam_dl().
2513 2514 2515 2516 2517 2518
 */
static int dl_overflow(struct task_struct *p, int policy,
		       const struct sched_attr *attr)
{

	struct dl_bw *dl_b = dl_bw_of(task_cpu(p));
2519
	u64 period = attr->sched_period ?: attr->sched_deadline;
2520 2521
	u64 runtime = attr->sched_runtime;
	u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2522
	int cpus, err = -1;
2523

2524 2525
	/* !deadline task may carry old deadline bandwidth */
	if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2526 2527 2528 2529 2530 2531 2532 2533
		return 0;

	/*
	 * Either if a task, enters, leave, or stays -deadline but changes
	 * its parameters, we may need to update accordingly the total
	 * allocated bandwidth of the container.
	 */
	raw_spin_lock(&dl_b->lock);
2534
	cpus = dl_bw_cpus(task_cpu(p));
2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554
	if (dl_policy(policy) && !task_has_dl_policy(p) &&
	    !__dl_overflow(dl_b, cpus, 0, new_bw)) {
		__dl_add(dl_b, new_bw);
		err = 0;
	} else if (dl_policy(policy) && task_has_dl_policy(p) &&
		   !__dl_overflow(dl_b, cpus, p->dl.dl_bw, new_bw)) {
		__dl_clear(dl_b, p->dl.dl_bw);
		__dl_add(dl_b, new_bw);
		err = 0;
	} else if (!dl_policy(policy) && task_has_dl_policy(p)) {
		__dl_clear(dl_b, p->dl.dl_bw);
		err = 0;
	}
	raw_spin_unlock(&dl_b->lock);

	return err;
}

extern void init_dl_bw(struct dl_bw *dl_b);

L
Linus Torvalds 已提交
2555 2556 2557 2558 2559 2560 2561
/*
 * 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.
 */
2562
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2563
{
2564
	struct rq_flags rf;
I
Ingo Molnar 已提交
2565
	struct rq *rq;
2566

2567
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2568
	p->state = TASK_RUNNING;
2569 2570 2571 2572 2573
#ifdef CONFIG_SMP
	/*
	 * Fork balancing, do it here and not earlier because:
	 *  - cpus_allowed can change in the fork path
	 *  - any previously selected cpu might disappear through hotplug
2574 2575 2576
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2577
	 */
2578
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2579
#endif
2580
	rq = __task_rq_lock(p, &rf);
2581
	post_init_entity_util_avg(&p->se);
2582

P
Peter Zijlstra 已提交
2583
	activate_task(rq, p, 0);
2584
	p->on_rq = TASK_ON_RQ_QUEUED;
2585
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2586
	check_preempt_curr(rq, p, WF_FORK);
2587
#ifdef CONFIG_SMP
2588 2589 2590 2591 2592
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2593
		lockdep_unpin_lock(&rq->lock, rf.cookie);
2594
		p->sched_class->task_woken(rq, p);
2595
		lockdep_repin_lock(&rq->lock, rf.cookie);
2596
	}
2597
#endif
2598
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2599 2600
}

2601 2602
#ifdef CONFIG_PREEMPT_NOTIFIERS

2603 2604
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616
void preempt_notifier_inc(void)
{
	static_key_slow_inc(&preempt_notifier_key);
}
EXPORT_SYMBOL_GPL(preempt_notifier_inc);

void preempt_notifier_dec(void)
{
	static_key_slow_dec(&preempt_notifier_key);
}
EXPORT_SYMBOL_GPL(preempt_notifier_dec);

2617
/**
2618
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2619
 * @notifier: notifier struct to register
2620 2621 2622
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2623 2624 2625
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2626 2627 2628 2629 2630 2631
	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
2632
 * @notifier: notifier struct to unregister
2633
 *
2634
 * This is *not* safe to call from within a preemption notifier.
2635 2636 2637 2638 2639 2640 2641
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2642
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2643 2644 2645
{
	struct preempt_notifier *notifier;

2646
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2647 2648 2649
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2650 2651 2652 2653 2654 2655
static __always_inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
{
	if (static_key_false(&preempt_notifier_key))
		__fire_sched_in_preempt_notifiers(curr);
}

2656
static void
2657 2658
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2659 2660 2661
{
	struct preempt_notifier *notifier;

2662
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2663 2664 2665
		notifier->ops->sched_out(notifier, next);
}

2666 2667 2668 2669 2670 2671 2672 2673
static __always_inline void
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
	if (static_key_false(&preempt_notifier_key))
		__fire_sched_out_preempt_notifiers(curr, next);
}

2674
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2675

2676
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2677 2678 2679
{
}

2680
static inline void
2681 2682 2683 2684 2685
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2686
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2687

2688 2689 2690
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2691
 * @prev: the current task that is being switched out
2692 2693 2694 2695 2696 2697 2698 2699 2700
 * @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.
 */
2701 2702 2703
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2704
{
2705
	sched_info_switch(rq, prev, next);
2706
	perf_event_task_sched_out(prev, next);
2707
	fire_sched_out_preempt_notifiers(prev, next);
2708 2709 2710 2711
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2712 2713 2714 2715
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2716 2717 2718 2719
 * 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
2720 2721
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2722
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2723 2724
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2725 2726 2727 2728 2729
 *
 * 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
2730
 */
2731
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2732 2733
	__releases(rq->lock)
{
2734
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2735
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2736
	long prev_state;
L
Linus Torvalds 已提交
2737

2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
	/*
	 * 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.
	 */
2749 2750 2751 2752
	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);
2753

L
Linus Torvalds 已提交
2754 2755 2756 2757
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2758
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2759 2760
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2761 2762 2763 2764 2765
	 *
	 * We must observe prev->state before clearing prev->on_cpu (in
	 * finish_lock_switch), otherwise a concurrent wakeup can get prev
	 * running on another CPU and we could rave with its RUNNING -> DEAD
	 * transition, resulting in a double drop.
L
Linus Torvalds 已提交
2766
	 */
O
Oleg Nesterov 已提交
2767
	prev_state = prev->state;
2768
	vtime_task_switch(prev);
2769
	perf_event_task_sched_in(prev, current);
2770
	finish_lock_switch(rq, prev);
2771
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2772

2773
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
2774 2775
	if (mm)
		mmdrop(mm);
2776
	if (unlikely(prev_state == TASK_DEAD)) {
2777 2778 2779
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2780 2781 2782
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2783
		 */
2784
		kprobe_flush_task(prev);
2785 2786 2787 2788

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

L
Linus Torvalds 已提交
2789
		put_task_struct(prev);
2790
	}
2791

2792
	tick_nohz_task_switch();
2793
	return rq;
L
Linus Torvalds 已提交
2794 2795
}

2796 2797 2798
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2799
static void __balance_callback(struct rq *rq)
2800
{
2801 2802 2803
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2804

2805 2806 2807 2808 2809 2810 2811 2812
	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;
2813

2814
		func(rq);
2815
	}
2816 2817 2818 2819 2820 2821 2822
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2823 2824 2825
}

#else
2826

2827
static inline void balance_callback(struct rq *rq)
2828
{
L
Linus Torvalds 已提交
2829 2830
}

2831 2832
#endif

L
Linus Torvalds 已提交
2833 2834 2835 2836
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2837
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2838 2839
	__releases(rq->lock)
{
2840
	struct rq *rq;
2841

2842 2843 2844 2845 2846 2847 2848 2849 2850
	/*
	 * 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).
	 */

2851
	rq = finish_task_switch(prev);
2852
	balance_callback(rq);
2853
	preempt_enable();
2854

L
Linus Torvalds 已提交
2855
	if (current->set_child_tid)
2856
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2857 2858 2859
}

/*
2860
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2861
 */
2862
static __always_inline struct rq *
2863
context_switch(struct rq *rq, struct task_struct *prev,
2864
	       struct task_struct *next, struct pin_cookie cookie)
L
Linus Torvalds 已提交
2865
{
I
Ingo Molnar 已提交
2866
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2867

2868
	prepare_task_switch(rq, prev, next);
2869

I
Ingo Molnar 已提交
2870 2871
	mm = next->mm;
	oldmm = prev->active_mm;
2872 2873 2874 2875 2876
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2877
	arch_start_context_switch(prev);
2878

2879
	if (!mm) {
L
Linus Torvalds 已提交
2880 2881 2882 2883
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
2884
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2885

2886
	if (!prev->mm) {
L
Linus Torvalds 已提交
2887 2888 2889
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2890 2891 2892 2893 2894 2895
	/*
	 * 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:
	 */
2896
	lockdep_unpin_lock(&rq->lock, cookie);
2897
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2898 2899 2900

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2901
	barrier();
2902 2903

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2904 2905 2906
}

/*
2907
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2908 2909
 *
 * externally visible scheduler statistics: current number of runnable
2910
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2911 2912 2913 2914 2915 2916 2917 2918 2919
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2920
}
L
Linus Torvalds 已提交
2921

2922 2923
/*
 * Check if only the current task is running on the cpu.
2924 2925 2926 2927 2928 2929 2930 2931 2932 2933
 *
 * 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)
2934 2935 2936
 */
bool single_task_running(void)
{
2937
	return raw_rq()->nr_running == 1;
2938 2939 2940
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2941
unsigned long long nr_context_switches(void)
2942
{
2943 2944
	int i;
	unsigned long long sum = 0;
2945

2946
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2947
		sum += cpu_rq(i)->nr_switches;
2948

L
Linus Torvalds 已提交
2949 2950
	return sum;
}
2951

L
Linus Torvalds 已提交
2952 2953 2954
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2955

2956
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2957
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2958

L
Linus Torvalds 已提交
2959 2960
	return sum;
}
2961

2962
unsigned long nr_iowait_cpu(int cpu)
2963
{
2964
	struct rq *this = cpu_rq(cpu);
2965 2966
	return atomic_read(&this->nr_iowait);
}
2967

2968 2969
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2970 2971 2972
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2973 2974
}

I
Ingo Molnar 已提交
2975
#ifdef CONFIG_SMP
2976

2977
/*
P
Peter Zijlstra 已提交
2978 2979
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2980
 */
P
Peter Zijlstra 已提交
2981
void sched_exec(void)
2982
{
P
Peter Zijlstra 已提交
2983
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2984
	unsigned long flags;
2985
	int dest_cpu;
2986

2987
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2988
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2989 2990
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2991

2992
	if (likely(cpu_active(dest_cpu))) {
2993
		struct migration_arg arg = { p, dest_cpu };
2994

2995 2996
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2997 2998
		return;
	}
2999
unlock:
3000
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
3001
}
I
Ingo Molnar 已提交
3002

L
Linus Torvalds 已提交
3003 3004 3005
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
3006
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
3007 3008

EXPORT_PER_CPU_SYMBOL(kstat);
3009
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
3010

3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027
/*
 * 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);
}

3028 3029 3030 3031 3032 3033 3034
/*
 * 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)
{
3035
	struct rq_flags rf;
3036
	struct rq *rq;
3037
	u64 ns;
3038

3039 3040 3041 3042 3043 3044 3045 3046 3047
#if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
	/*
	 * 64-bit doesn't need locks to atomically read a 64bit value.
	 * So we have a optimization chance when the task's delta_exec is 0.
	 * Reading ->on_cpu is racy, but this is ok.
	 *
	 * 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
	 * indistinguishable from the read occurring a few cycles earlier.
3048 3049
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3050
	 */
3051
	if (!p->on_cpu || !task_on_rq_queued(p))
3052 3053 3054
		return p->se.sum_exec_runtime;
#endif

3055
	rq = task_rq_lock(p, &rf);
3056 3057 3058 3059 3060 3061
	/*
	 * 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)) {
3062
		prefetch_curr_exec_start(p);
3063 3064 3065 3066
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3067
	task_rq_unlock(rq, p, &rf);
3068 3069 3070

	return ns;
}
3071

3072 3073 3074 3075 3076 3077 3078 3079
/*
 * 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
3080
	struct task_struct *curr = rq->curr;
3081 3082

	sched_clock_tick();
I
Ingo Molnar 已提交
3083

3084
	raw_spin_lock(&rq->lock);
3085
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3086
	curr->sched_class->task_tick(rq, curr, 0);
3087
	cpu_load_update_active(rq);
3088
	calc_global_load_tick(rq);
3089
	raw_spin_unlock(&rq->lock);
3090

3091
	perf_event_task_tick();
3092

3093
#ifdef CONFIG_SMP
3094
	rq->idle_balance = idle_cpu(cpu);
3095
	trigger_load_balance(rq);
3096
#endif
3097
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
3098 3099
}

3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110
#ifdef CONFIG_NO_HZ_FULL
/**
 * scheduler_tick_max_deferment
 *
 * Keep at least one tick per second when a single
 * active task is running because the scheduler doesn't
 * yet completely support full dynticks environment.
 *
 * This makes sure that uptime, CFS vruntime, load
 * balancing, etc... continue to move forward, even
 * with a very low granularity.
3111 3112
 *
 * Return: Maximum deferment in nanoseconds.
3113 3114 3115 3116
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
3117
	unsigned long next, now = READ_ONCE(jiffies);
3118 3119 3120 3121 3122 3123

	next = rq->last_sched_tick + HZ;

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

3124
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
3125
}
3126
#endif
L
Linus Torvalds 已提交
3127

3128 3129
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143
/*
 * 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);
	}
}
3144

3145
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3146
{
3147
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3148 3149 3150
	/*
	 * Underflow?
	 */
3151 3152
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3153
#endif
3154
	__preempt_count_add(val);
3155
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3156 3157 3158
	/*
	 * Spinlock count overflowing soon?
	 */
3159 3160
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3161
#endif
3162
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3163
}
3164
EXPORT_SYMBOL(preempt_count_add);
3165
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3166

3167 3168 3169 3170 3171 3172 3173 3174 3175 3176
/*
 * 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());
}

3177
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3178
{
3179
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3180 3181 3182
	/*
	 * Underflow?
	 */
3183
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3184
		return;
L
Linus Torvalds 已提交
3185 3186 3187
	/*
	 * Is the spinlock portion underflowing?
	 */
3188 3189 3190
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3191
#endif
3192

3193
	preempt_latency_stop(val);
3194
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3195
}
3196
EXPORT_SYMBOL(preempt_count_sub);
3197
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3198

3199 3200 3201
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3202 3203 3204
#endif

/*
I
Ingo Molnar 已提交
3205
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3206
 */
I
Ingo Molnar 已提交
3207
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3208
{
3209 3210 3211
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3212 3213 3214
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3215 3216
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3217

I
Ingo Molnar 已提交
3218
	debug_show_held_locks(prev);
3219
	print_modules();
I
Ingo Molnar 已提交
3220 3221
	if (irqs_disabled())
		print_irqtrace_events(prev);
3222 3223
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3224
		pr_err("Preemption disabled at:");
3225
		print_ip_sym(preempt_disable_ip);
3226 3227
		pr_cont("\n");
	}
3228 3229 3230
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3231
	dump_stack();
3232
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3233
}
L
Linus Torvalds 已提交
3234

I
Ingo Molnar 已提交
3235 3236 3237 3238 3239
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3240
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3241 3242
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3243
#endif
3244

3245
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3246
		__schedule_bug(prev);
3247 3248
		preempt_count_set(PREEMPT_DISABLED);
	}
3249
	rcu_sleep_check();
I
Ingo Molnar 已提交
3250

L
Linus Torvalds 已提交
3251 3252
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3253
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3254 3255 3256 3257 3258 3259
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3260
pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
I
Ingo Molnar 已提交
3261
{
3262
	const struct sched_class *class = &fair_sched_class;
I
Ingo Molnar 已提交
3263
	struct task_struct *p;
L
Linus Torvalds 已提交
3264 3265

	/*
I
Ingo Molnar 已提交
3266 3267
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
3268
	 */
3269
	if (likely(prev->sched_class == class &&
3270
		   rq->nr_running == rq->cfs.h_nr_running)) {
3271
		p = fair_sched_class.pick_next_task(rq, prev, cookie);
3272 3273 3274 3275 3276
		if (unlikely(p == RETRY_TASK))
			goto again;

		/* assumes fair_sched_class->next == idle_sched_class */
		if (unlikely(!p))
3277
			p = idle_sched_class.pick_next_task(rq, prev, cookie);
3278 3279

		return p;
L
Linus Torvalds 已提交
3280 3281
	}

3282
again:
3283
	for_each_class(class) {
3284
		p = class->pick_next_task(rq, prev, cookie);
3285 3286 3287
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3288
			return p;
3289
		}
I
Ingo Molnar 已提交
3290
	}
3291 3292

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
3293
}
L
Linus Torvalds 已提交
3294

I
Ingo Molnar 已提交
3295
/*
3296
 * __schedule() is the main scheduler function.
3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330
 *
 * 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
3331
 *
3332
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3333
 */
3334
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3335 3336
{
	struct task_struct *prev, *next;
3337
	unsigned long *switch_count;
3338
	struct pin_cookie cookie;
I
Ingo Molnar 已提交
3339
	struct rq *rq;
3340
	int cpu;
I
Ingo Molnar 已提交
3341 3342 3343 3344 3345 3346

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3347

3348
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3349
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3350

3351 3352 3353
	local_irq_disable();
	rcu_note_context_switch();

3354 3355 3356 3357 3358 3359
	/*
	 * 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().
	 */
	smp_mb__before_spinlock();
3360
	raw_spin_lock(&rq->lock);
3361
	cookie = lockdep_pin_lock(&rq->lock);
L
Linus Torvalds 已提交
3362

3363 3364
	rq->clock_skip_update <<= 1; /* promote REQ to ACT */

3365
	switch_count = &prev->nivcsw;
3366
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3367
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3368
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3369
		} else {
3370 3371 3372
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
3373
			/*
3374 3375 3376
			 * 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
3377 3378 3379 3380
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3381
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3382
				if (to_wakeup)
3383
					try_to_wake_up_local(to_wakeup, cookie);
T
Tejun Heo 已提交
3384 3385
			}
		}
I
Ingo Molnar 已提交
3386
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3387 3388
	}

3389
	if (task_on_rq_queued(prev))
3390 3391
		update_rq_clock(rq);

3392
	next = pick_next_task(rq, prev, cookie);
3393
	clear_tsk_need_resched(prev);
3394
	clear_preempt_need_resched();
3395
	rq->clock_skip_update = 0;
L
Linus Torvalds 已提交
3396 3397 3398 3399 3400 3401

	if (likely(prev != next)) {
		rq->nr_switches++;
		rq->curr = next;
		++*switch_count;

3402
		trace_sched_switch(preempt, prev, next);
3403
		rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */
3404
	} else {
3405
		lockdep_unpin_lock(&rq->lock, cookie);
3406
		raw_spin_unlock_irq(&rq->lock);
3407
	}
L
Linus Torvalds 已提交
3408

3409
	balance_callback(rq);
L
Linus Torvalds 已提交
3410
}
3411

3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438
void __noreturn do_task_dead(void)
{
	/*
	 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
	 * when the following two conditions become true.
	 *   - There is race condition of mmap_sem (It is acquired by
	 *     exit_mm()), and
	 *   - SMI occurs before setting TASK_RUNINNG.
	 *     (or hypervisor of virtual machine switches to other guest)
	 *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
	 *
	 * To avoid it, we have to wait for releasing tsk->pi_lock which
	 * is held by try_to_wake_up()
	 */
	smp_mb();
	raw_spin_unlock_wait(&current->pi_lock);

	/* causes final put_task_struct in finish_task_switch(). */
	__set_current_state(TASK_DEAD);
	current->flags |= PF_NOFREEZE;	/* tell freezer to ignore us */
	__schedule(false);
	BUG();
	/* Avoid "noreturn function does return".  */
	for (;;)
		cpu_relax();	/* For when BUG is null */
}

3439 3440
static inline void sched_submit_work(struct task_struct *tsk)
{
3441
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3442 3443 3444 3445 3446 3447 3448 3449 3450
		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);
}

3451
asmlinkage __visible void __sched schedule(void)
3452
{
3453 3454 3455
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3456
	do {
3457
		preempt_disable();
3458
		__schedule(false);
3459
		sched_preempt_enable_no_resched();
3460
	} while (need_resched());
3461
}
L
Linus Torvalds 已提交
3462 3463
EXPORT_SYMBOL(schedule);

3464
#ifdef CONFIG_CONTEXT_TRACKING
3465
asmlinkage __visible void __sched schedule_user(void)
3466 3467 3468 3469 3470 3471
{
	/*
	 * 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.
3472 3473
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3474
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3475
	 * too frequently to make sense yet.
3476
	 */
3477
	enum ctx_state prev_state = exception_enter();
3478
	schedule();
3479
	exception_exit(prev_state);
3480 3481 3482
}
#endif

3483 3484 3485 3486 3487 3488 3489
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3490
	sched_preempt_enable_no_resched();
3491 3492 3493 3494
	schedule();
	preempt_disable();
}

3495
static void __sched notrace preempt_schedule_common(void)
3496 3497
{
	do {
3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510
		/*
		 * 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.
		 */
3511
		preempt_disable_notrace();
3512
		preempt_latency_start(1);
3513
		__schedule(true);
3514
		preempt_latency_stop(1);
3515
		preempt_enable_no_resched_notrace();
3516 3517 3518 3519 3520 3521 3522 3523

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

L
Linus Torvalds 已提交
3524 3525
#ifdef CONFIG_PREEMPT
/*
3526
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3527
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3528 3529
 * occur there and call schedule directly.
 */
3530
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3531 3532 3533
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3534
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3535
	 */
3536
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3537 3538
		return;

3539
	preempt_schedule_common();
L
Linus Torvalds 已提交
3540
}
3541
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3542
EXPORT_SYMBOL(preempt_schedule);
3543 3544

/**
3545
 * preempt_schedule_notrace - preempt_schedule called by tracing
3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557
 *
 * 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.
 */
3558
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3559 3560 3561 3562 3563 3564 3565
{
	enum ctx_state prev_ctx;

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

	do {
3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578
		/*
		 * 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.
		 */
3579
		preempt_disable_notrace();
3580
		preempt_latency_start(1);
3581 3582 3583 3584 3585 3586
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3587
		__schedule(true);
3588 3589
		exception_exit(prev_ctx);

3590
		preempt_latency_stop(1);
3591
		preempt_enable_no_resched_notrace();
3592 3593
	} while (need_resched());
}
3594
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3595

3596
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3597 3598

/*
3599
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3600 3601 3602 3603
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3604
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3605
{
3606
	enum ctx_state prev_state;
3607

3608
	/* Catch callers which need to be fixed */
3609
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3610

3611 3612
	prev_state = exception_enter();

3613
	do {
3614
		preempt_disable();
3615
		local_irq_enable();
3616
		__schedule(true);
3617
		local_irq_disable();
3618
		sched_preempt_enable_no_resched();
3619
	} while (need_resched());
3620 3621

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3622 3623
}

P
Peter Zijlstra 已提交
3624
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3625
			  void *key)
L
Linus Torvalds 已提交
3626
{
P
Peter Zijlstra 已提交
3627
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3628 3629 3630
}
EXPORT_SYMBOL(default_wake_function);

3631 3632 3633 3634 3635 3636 3637 3638 3639 3640
#ifdef CONFIG_RT_MUTEXES

/*
 * rt_mutex_setprio - set the current priority of a task
 * @p: task
 * @prio: prio value (kernel-internal form)
 *
 * This function changes the 'effective' priority of a task. It does
 * not touch ->normal_prio like __setscheduler().
 *
3641 3642
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3643
 */
3644
void rt_mutex_setprio(struct task_struct *p, int prio)
3645
{
3646
	int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE;
3647
	const struct sched_class *prev_class;
3648 3649
	struct rq_flags rf;
	struct rq *rq;
3650

3651
	BUG_ON(prio > MAX_PRIO);
3652

3653
	rq = __task_rq_lock(p, &rf);
3654

3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672
	/*
	 * 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;
	}

3673
	trace_sched_pi_setprio(p, prio);
3674
	oldprio = p->prio;
3675 3676 3677 3678

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

3679
	prev_class = p->sched_class;
3680
	queued = task_on_rq_queued(p);
3681
	running = task_current(rq, p);
3682
	if (queued)
3683
		dequeue_task(rq, p, queue_flag);
3684
	if (running)
3685
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3686

3687 3688 3689 3690 3691 3692 3693 3694 3695 3696
	/*
	 * 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)) {
3697 3698 3699
		struct task_struct *pi_task = rt_mutex_get_top_task(p);
		if (!dl_prio(p->normal_prio) ||
		    (pi_task && dl_entity_preempt(&pi_task->dl, &p->dl))) {
3700
			p->dl.dl_boosted = 1;
3701
			queue_flag |= ENQUEUE_REPLENISH;
3702 3703
		} else
			p->dl.dl_boosted = 0;
3704
		p->sched_class = &dl_sched_class;
3705 3706 3707 3708
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3709
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3710
		p->sched_class = &rt_sched_class;
3711 3712 3713
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3714 3715
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3716
		p->sched_class = &fair_sched_class;
3717
	}
I
Ingo Molnar 已提交
3718

3719 3720
	p->prio = prio;

3721
	if (queued)
3722
		enqueue_task(rq, p, queue_flag);
3723
	if (running)
3724
		set_curr_task(rq, p);
3725

P
Peter Zijlstra 已提交
3726
	check_class_changed(rq, p, prev_class, oldprio);
3727
out_unlock:
3728
	preempt_disable(); /* avoid rq from going away on us */
3729
	__task_rq_unlock(rq, &rf);
3730 3731 3732

	balance_callback(rq);
	preempt_enable();
3733 3734
}
#endif
3735

3736
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3737
{
P
Peter Zijlstra 已提交
3738 3739
	bool queued, running;
	int old_prio, delta;
3740
	struct rq_flags rf;
3741
	struct rq *rq;
L
Linus Torvalds 已提交
3742

3743
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3744 3745 3746 3747 3748
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3749
	rq = task_rq_lock(p, &rf);
L
Linus Torvalds 已提交
3750 3751 3752 3753
	/*
	 * 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
3754
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3755
	 */
3756
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3757 3758 3759
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3760
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3761
	running = task_current(rq, p);
3762
	if (queued)
3763
		dequeue_task(rq, p, DEQUEUE_SAVE);
P
Peter Zijlstra 已提交
3764 3765
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3766 3767

	p->static_prio = NICE_TO_PRIO(nice);
3768
	set_load_weight(p);
3769 3770 3771
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3772

3773
	if (queued) {
3774
		enqueue_task(rq, p, ENQUEUE_RESTORE);
L
Linus Torvalds 已提交
3775
		/*
3776 3777
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3778
		 */
3779
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3780
			resched_curr(rq);
L
Linus Torvalds 已提交
3781
	}
P
Peter Zijlstra 已提交
3782 3783
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3784
out_unlock:
3785
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3786 3787 3788
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3789 3790 3791 3792 3793
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3794
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3795
{
3796
	/* convert nice value [19,-20] to rlimit style value [1,40] */
3797
	int nice_rlim = nice_to_rlimit(nice);
3798

3799
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3800 3801 3802
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3803 3804 3805 3806 3807 3808 3809 3810 3811
#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.
 */
3812
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3813
{
3814
	long nice, retval;
L
Linus Torvalds 已提交
3815 3816 3817 3818 3819 3820

	/*
	 * 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.
	 */
3821
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3822
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3823

3824
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3825 3826 3827
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841
	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.
 *
3842
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3843 3844 3845
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3846
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3847 3848 3849 3850 3851 3852 3853
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
3854 3855
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3856 3857 3858
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872
	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
3873 3874 3875 3876 3877
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
3878 3879
 *
 * Return: The idle task for the cpu @cpu.
L
Linus Torvalds 已提交
3880
 */
3881
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3882 3883 3884 3885 3886 3887 3888
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3889 3890
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3891
 */
A
Alexey Dobriyan 已提交
3892
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3893
{
3894
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3895 3896
}

3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911
/*
 * This function initializes the sched_dl_entity of a newly becoming
 * SCHED_DEADLINE task.
 *
 * Only the static values are considered here, the actual runtime and the
 * absolute deadline will be properly calculated when the task is enqueued
 * for the first time with its new policy.
 */
static void
__setparam_dl(struct task_struct *p, const struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	dl_se->dl_runtime = attr->sched_runtime;
	dl_se->dl_deadline = attr->sched_deadline;
3912
	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
3913
	dl_se->flags = attr->sched_flags;
3914
	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934

	/*
	 * Changing the parameters of a task is 'tricky' and we're not doing
	 * the correct thing -- also see task_dead_dl() and switched_from_dl().
	 *
	 * What we SHOULD do is delay the bandwidth release until the 0-lag
	 * point. This would include retaining the task_struct until that time
	 * and change dl_overflow() to not immediately decrement the current
	 * amount.
	 *
	 * Instead we retain the current runtime/deadline and let the new
	 * parameters take effect after the current reservation period lapses.
	 * This is safe (albeit pessimistic) because the 0-lag point is always
	 * before the current scheduling deadline.
	 *
	 * We can still have temporary overloads because we do not delay the
	 * change in bandwidth until that time; so admission control is
	 * not on the safe side. It does however guarantee tasks will never
	 * consume more than promised.
	 */
3935 3936
}

3937 3938 3939 3940 3941 3942
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

3943 3944
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
3945
{
3946 3947
	int policy = attr->sched_policy;

3948
	if (policy == SETPARAM_POLICY)
3949 3950
		policy = p->policy;

L
Linus Torvalds 已提交
3951
	p->policy = policy;
3952

3953 3954
	if (dl_policy(policy))
		__setparam_dl(p, attr);
3955
	else if (fair_policy(policy))
3956 3957
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

3958 3959 3960 3961 3962 3963
	/*
	 * __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;
3964
	p->normal_prio = normal_prio(p);
3965 3966
	set_load_weight(p);
}
3967

3968 3969
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
3970
			   const struct sched_attr *attr, bool keep_boost)
3971 3972
{
	__setscheduler_params(p, attr);
3973

3974
	/*
3975 3976
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
3977
	 */
3978 3979 3980 3981
	if (keep_boost)
		p->prio = rt_mutex_get_effective_prio(p, normal_prio(p));
	else
		p->prio = normal_prio(p);
3982

3983 3984 3985
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
3986 3987 3988
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
3989
}
3990 3991 3992 3993 3994 3995 3996 3997 3998

static void
__getparam_dl(struct task_struct *p, struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	attr->sched_priority = p->rt_priority;
	attr->sched_runtime = dl_se->dl_runtime;
	attr->sched_deadline = dl_se->dl_deadline;
3999
	attr->sched_period = dl_se->dl_period;
4000 4001 4002 4003 4004 4005
	attr->sched_flags = dl_se->flags;
}

/*
 * This function validates the new parameters of a -deadline task.
 * We ask for the deadline not being zero, and greater or equal
4006
 * than the runtime, as well as the period of being zero or
4007
 * greater than deadline. Furthermore, we have to be sure that
4008 4009 4010 4011
 * user parameters are above the internal resolution of 1us (we
 * check sched_runtime only since it is always the smaller one) and
 * below 2^63 ns (we have to check both sched_deadline and
 * sched_period, as the latter can be zero).
4012 4013 4014 4015
 */
static bool
__checkparam_dl(const struct sched_attr *attr)
{
4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041
	/* deadline != 0 */
	if (attr->sched_deadline == 0)
		return false;

	/*
	 * Since we truncate DL_SCALE bits, make sure we're at least
	 * that big.
	 */
	if (attr->sched_runtime < (1ULL << DL_SCALE))
		return false;

	/*
	 * Since we use the MSB for wrap-around and sign issues, make
	 * sure it's not set (mind that period can be equal to zero).
	 */
	if (attr->sched_deadline & (1ULL << 63) ||
	    attr->sched_period & (1ULL << 63))
		return false;

	/* runtime <= deadline <= period (if period != 0) */
	if ((attr->sched_period != 0 &&
	     attr->sched_period < attr->sched_deadline) ||
	    attr->sched_deadline < attr->sched_runtime)
		return false;

	return true;
4042 4043
}

4044 4045 4046 4047 4048 4049 4050 4051 4052 4053
/*
 * check the target process has a UID that matches the current process's
 */
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);
4054 4055
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4056 4057 4058 4059
	rcu_read_unlock();
	return match;
}

4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073
static bool dl_param_changed(struct task_struct *p,
		const struct sched_attr *attr)
{
	struct sched_dl_entity *dl_se = &p->dl;

	if (dl_se->dl_runtime != attr->sched_runtime ||
		dl_se->dl_deadline != attr->sched_deadline ||
		dl_se->dl_period != attr->sched_period ||
		dl_se->flags != attr->sched_flags)
		return true;

	return false;
}

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

4088 4089
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
4090 4091
recheck:
	/* double check policy once rq lock held */
4092 4093
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4094
		policy = oldpolicy = p->policy;
4095
	} else {
4096
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4097

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

4102 4103 4104
	if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK))
		return -EINVAL;

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

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

4127
		if (rt_policy(policy)) {
4128 4129
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4130 4131 4132 4133 4134 4135

			/* can't set/change the rt policy */
			if (policy != p->policy && !rlim_rtprio)
				return -EPERM;

			/* can't increase priority */
4136 4137
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4138 4139
				return -EPERM;
		}
4140

4141 4142 4143 4144 4145 4146 4147 4148 4149
		 /*
		  * Can't set/change SCHED_DEADLINE policy at all for now
		  * (safest behavior); in the future we would like to allow
		  * unprivileged DL tasks to increase their relative deadline
		  * or reduce their runtime (both ways reducing utilization)
		  */
		if (dl_policy(policy))
			return -EPERM;

I
Ingo Molnar 已提交
4150
		/*
4151 4152
		 * Treat SCHED_IDLE as nice 20. Only allow a switch to
		 * SCHED_NORMAL if the RLIMIT_NICE would normally permit it.
I
Ingo Molnar 已提交
4153
		 */
4154
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4155
			if (!can_nice(p, task_nice(p)))
4156 4157
				return -EPERM;
		}
4158

4159
		/* can't change other user's priorities */
4160
		if (!check_same_owner(p))
4161
			return -EPERM;
4162 4163 4164 4165

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

4168
	if (user) {
4169
		retval = security_task_setscheduler(p);
4170 4171 4172 4173
		if (retval)
			return retval;
	}

4174 4175 4176
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
4177
	 *
L
Lucas De Marchi 已提交
4178
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4179 4180
	 * runqueue lock must be held.
	 */
4181
	rq = task_rq_lock(p, &rf);
4182

4183 4184 4185 4186
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
4187
		task_rq_unlock(rq, p, &rf);
4188 4189 4190
		return -EINVAL;
	}

4191
	/*
4192 4193
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4194
	 */
4195
	if (unlikely(policy == p->policy)) {
4196
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4197 4198 4199
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4200
		if (dl_policy(policy) && dl_param_changed(p, attr))
4201
			goto change;
4202

4203
		p->sched_reset_on_fork = reset_on_fork;
4204
		task_rq_unlock(rq, p, &rf);
4205 4206
		return 0;
	}
4207
change:
4208

4209
	if (user) {
4210
#ifdef CONFIG_RT_GROUP_SCHED
4211 4212 4213 4214 4215
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4216 4217
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4218
			task_rq_unlock(rq, p, &rf);
4219 4220 4221
			return -EPERM;
		}
#endif
4222 4223 4224 4225 4226 4227 4228 4229 4230
#ifdef CONFIG_SMP
		if (dl_bandwidth_enabled() && dl_policy(policy)) {
			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.
			 */
4231 4232
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4233
				task_rq_unlock(rq, p, &rf);
4234 4235 4236 4237 4238
				return -EPERM;
			}
		}
#endif
	}
4239

L
Linus Torvalds 已提交
4240 4241 4242
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4243
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4244 4245
		goto recheck;
	}
4246 4247 4248 4249 4250 4251

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4252
	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
4253
		task_rq_unlock(rq, p, &rf);
4254 4255 4256
		return -EBUSY;
	}

4257 4258 4259
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4260 4261 4262 4263 4264 4265 4266 4267 4268
	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.
		 */
		new_effective_prio = rt_mutex_get_effective_prio(p, newprio);
4269 4270
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4271 4272
	}

4273
	queued = task_on_rq_queued(p);
4274
	running = task_current(rq, p);
4275
	if (queued)
4276
		dequeue_task(rq, p, queue_flags);
4277
	if (running)
4278
		put_prev_task(rq, p);
4279

4280
	prev_class = p->sched_class;
4281
	__setscheduler(rq, p, attr, pi);
4282

4283
	if (queued) {
4284 4285 4286 4287
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4288 4289
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4290

4291
		enqueue_task(rq, p, queue_flags);
4292
	}
4293
	if (running)
4294
		set_curr_task(rq, p);
4295

P
Peter Zijlstra 已提交
4296
	check_class_changed(rq, p, prev_class, oldprio);
4297
	preempt_disable(); /* avoid rq from going away on us */
4298
	task_rq_unlock(rq, p, &rf);
4299

4300 4301
	if (pi)
		rt_mutex_adjust_pi(p);
4302

4303 4304 4305 4306 4307
	/*
	 * Run balance callbacks after we've adjusted the PI chain.
	 */
	balance_callback(rq);
	preempt_enable();
4308

L
Linus Torvalds 已提交
4309 4310
	return 0;
}
4311

4312 4313 4314 4315 4316 4317 4318 4319 4320
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),
	};

4321 4322
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4323 4324 4325 4326 4327
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4328
	return __sched_setscheduler(p, &attr, check, true);
4329
}
4330 4331 4332 4333 4334 4335
/**
 * 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.
 *
4336 4337
 * Return: 0 on success. An error code otherwise.
 *
4338 4339 4340
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4341
		       const struct sched_param *param)
4342
{
4343
	return _sched_setscheduler(p, policy, param, true);
4344
}
L
Linus Torvalds 已提交
4345 4346
EXPORT_SYMBOL_GPL(sched_setscheduler);

4347 4348
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4349
	return __sched_setscheduler(p, attr, true, true);
4350 4351 4352
}
EXPORT_SYMBOL_GPL(sched_setattr);

4353 4354 4355 4356 4357 4358 4359 4360 4361 4362
/**
 * 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.
4363 4364
 *
 * Return: 0 on success. An error code otherwise.
4365 4366
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4367
			       const struct sched_param *param)
4368
{
4369
	return _sched_setscheduler(p, policy, param, false);
4370
}
4371
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4372

I
Ingo Molnar 已提交
4373 4374
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4375 4376 4377
{
	struct sched_param lparam;
	struct task_struct *p;
4378
	int retval;
L
Linus Torvalds 已提交
4379 4380 4381 4382 4383

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4384 4385 4386

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4387
	p = find_process_by_pid(pid);
4388 4389 4390
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4391

L
Linus Torvalds 已提交
4392 4393 4394
	return retval;
}

4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456
/*
 * Mimics kernel/events/core.c perf_copy_attr().
 */
static int sched_copy_attr(struct sched_attr __user *uattr,
			   struct sched_attr *attr)
{
	u32 size;
	int ret;

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

	/*
	 * zero the full structure, so that a short copy will be nice.
	 */
	memset(attr, 0, sizeof(*attr));

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

	if (size > PAGE_SIZE)	/* silly large */
		goto err_size;

	if (!size)		/* abi compat */
		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;

	/*
	 * XXX: do we want to be lenient like existing syscalls; or do we want
	 * to be strict and return an error on out-of-bounds values?
	 */
4457
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4458

4459
	return 0;
4460 4461 4462

err_size:
	put_user(sizeof(*attr), &uattr->size);
4463
	return -E2BIG;
4464 4465
}

L
Linus Torvalds 已提交
4466 4467 4468 4469 4470
/**
 * 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.
4471 4472
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4473
 */
4474 4475
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
4476
{
4477 4478 4479 4480
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4481 4482 4483 4484 4485 4486 4487
	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.
4488 4489
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4490
 */
4491
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4492
{
4493
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4494 4495
}

4496 4497 4498
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4499
 * @uattr: structure containing the extended parameters.
4500
 * @flags: for future extension.
4501
 */
4502 4503
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4504 4505 4506 4507 4508
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4509
	if (!uattr || pid < 0 || flags)
4510 4511
		return -EINVAL;

4512 4513 4514
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4515

4516
	if ((int)attr.sched_policy < 0)
4517
		return -EINVAL;
4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528

	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
4529 4530 4531
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4532 4533 4534
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4535
 */
4536
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4537
{
4538
	struct task_struct *p;
4539
	int retval;
L
Linus Torvalds 已提交
4540 4541

	if (pid < 0)
4542
		return -EINVAL;
L
Linus Torvalds 已提交
4543 4544

	retval = -ESRCH;
4545
	rcu_read_lock();
L
Linus Torvalds 已提交
4546 4547 4548 4549
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4550 4551
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4552
	}
4553
	rcu_read_unlock();
L
Linus Torvalds 已提交
4554 4555 4556 4557
	return retval;
}

/**
4558
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4559 4560
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4561 4562 4563
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4564
 */
4565
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4566
{
4567
	struct sched_param lp = { .sched_priority = 0 };
4568
	struct task_struct *p;
4569
	int retval;
L
Linus Torvalds 已提交
4570 4571

	if (!param || pid < 0)
4572
		return -EINVAL;
L
Linus Torvalds 已提交
4573

4574
	rcu_read_lock();
L
Linus Torvalds 已提交
4575 4576 4577 4578 4579 4580 4581 4582 4583
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4584 4585
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4586
	rcu_read_unlock();
L
Linus Torvalds 已提交
4587 4588 4589 4590 4591 4592 4593 4594 4595

	/*
	 * 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:
4596
	rcu_read_unlock();
L
Linus Torvalds 已提交
4597 4598 4599
	return retval;
}

4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622
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)
4623
				return -EFBIG;
4624 4625 4626 4627 4628
		}

		attr->size = usize;
	}

4629
	ret = copy_to_user(uattr, attr, attr->size);
4630 4631 4632
	if (ret)
		return -EFAULT;

4633
	return 0;
4634 4635 4636
}

/**
4637
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4638
 * @pid: the pid in question.
J
Juri Lelli 已提交
4639
 * @uattr: structure containing the extended parameters.
4640
 * @size: sizeof(attr) for fwd/bwd comp.
4641
 * @flags: for future extension.
4642
 */
4643 4644
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4645 4646 4647 4648 4649 4650 4651 4652
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4653
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666
		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;
4667 4668
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4669 4670 4671
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4672 4673
		attr.sched_priority = p->rt_priority;
	else
4674
		attr.sched_nice = task_nice(p);
4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4686
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4687
{
4688
	cpumask_var_t cpus_allowed, new_mask;
4689 4690
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4691

4692
	rcu_read_lock();
L
Linus Torvalds 已提交
4693 4694 4695

	p = find_process_by_pid(pid);
	if (!p) {
4696
		rcu_read_unlock();
L
Linus Torvalds 已提交
4697 4698 4699
		return -ESRCH;
	}

4700
	/* Prevent p going away */
L
Linus Torvalds 已提交
4701
	get_task_struct(p);
4702
	rcu_read_unlock();
L
Linus Torvalds 已提交
4703

4704 4705 4706 4707
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4708 4709 4710 4711 4712 4713 4714 4715
	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
4716
	retval = -EPERM;
E
Eric W. Biederman 已提交
4717 4718 4719 4720
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4721
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4722 4723 4724
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4725

4726
	retval = security_task_setscheduler(p);
4727
	if (retval)
4728
		goto out_free_new_mask;
4729

4730 4731 4732 4733

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

4734 4735 4736 4737 4738 4739 4740
	/*
	 * 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
4741 4742 4743
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4744
			retval = -EBUSY;
4745
			rcu_read_unlock();
4746
			goto out_free_new_mask;
4747
		}
4748
		rcu_read_unlock();
4749 4750
	}
#endif
P
Peter Zijlstra 已提交
4751
again:
4752
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4753

P
Paul Menage 已提交
4754
	if (!retval) {
4755 4756
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4757 4758 4759 4760 4761
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4762
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4763 4764 4765
			goto again;
		}
	}
4766
out_free_new_mask:
4767 4768 4769 4770
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4771 4772 4773 4774 4775
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4776
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4777
{
4778 4779 4780 4781 4782
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4783 4784 4785 4786 4787 4788 4789 4790
	return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0;
}

/**
 * sys_sched_setaffinity - set the cpu affinity of a process
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
 * @user_mask_ptr: user-space pointer to the new cpu mask
4791 4792
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4793
 */
4794 4795
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4796
{
4797
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4798 4799
	int retval;

4800 4801
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4802

4803 4804 4805 4806 4807
	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
4808 4809
}

4810
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4811
{
4812
	struct task_struct *p;
4813
	unsigned long flags;
L
Linus Torvalds 已提交
4814 4815
	int retval;

4816
	rcu_read_lock();
L
Linus Torvalds 已提交
4817 4818 4819 4820 4821 4822

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

4823 4824 4825 4826
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4827
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4828
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4829
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4830 4831

out_unlock:
4832
	rcu_read_unlock();
L
Linus Torvalds 已提交
4833

4834
	return retval;
L
Linus Torvalds 已提交
4835 4836 4837 4838 4839 4840 4841
}

/**
 * sys_sched_getaffinity - get the cpu affinity of a process
 * @pid: pid of the process
 * @len: length in bytes of the bitmask pointed to by user_mask_ptr
 * @user_mask_ptr: user-space pointer to hold the current cpu mask
4842
 *
4843 4844
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4845
 */
4846 4847
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4848 4849
{
	int ret;
4850
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4851

A
Anton Blanchard 已提交
4852
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4853 4854
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4855 4856
		return -EINVAL;

4857 4858
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4859

4860 4861
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4862
		size_t retlen = min_t(size_t, len, cpumask_size());
4863 4864

		if (copy_to_user(user_mask_ptr, mask, retlen))
4865 4866
			ret = -EFAULT;
		else
4867
			ret = retlen;
4868 4869
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4870

4871
	return ret;
L
Linus Torvalds 已提交
4872 4873 4874 4875 4876
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4877 4878
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4879 4880
 *
 * Return: 0.
L
Linus Torvalds 已提交
4881
 */
4882
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4883
{
4884
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
4885

4886
	schedstat_inc(rq->yld_count);
4887
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4888 4889 4890 4891 4892 4893

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
4894
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4895
	do_raw_spin_unlock(&rq->lock);
4896
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4897 4898 4899 4900 4901 4902

	schedule();

	return 0;
}

4903
#ifndef CONFIG_PREEMPT
4904
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4905
{
4906
	if (should_resched(0)) {
4907
		preempt_schedule_common();
L
Linus Torvalds 已提交
4908 4909 4910 4911
		return 1;
	}
	return 0;
}
4912
EXPORT_SYMBOL(_cond_resched);
4913
#endif
L
Linus Torvalds 已提交
4914 4915

/*
4916
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4917 4918
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4919
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4920 4921 4922
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4923
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4924
{
4925
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4926 4927
	int ret = 0;

4928 4929
	lockdep_assert_held(lock);

4930
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4931
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4932
		if (resched)
4933
			preempt_schedule_common();
N
Nick Piggin 已提交
4934 4935
		else
			cpu_relax();
J
Jan Kara 已提交
4936
		ret = 1;
L
Linus Torvalds 已提交
4937 4938
		spin_lock(lock);
	}
J
Jan Kara 已提交
4939
	return ret;
L
Linus Torvalds 已提交
4940
}
4941
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4942

4943
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4944 4945 4946
{
	BUG_ON(!in_softirq());

4947
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4948
		local_bh_enable();
4949
		preempt_schedule_common();
L
Linus Torvalds 已提交
4950 4951 4952 4953 4954
		local_bh_disable();
		return 1;
	}
	return 0;
}
4955
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4956 4957 4958 4959

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977
 * 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)
 * 	yield();
 *
 * 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
4978 4979 4980 4981 4982 4983 4984 4985
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4986 4987 4988 4989
/**
 * 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
4990 4991
 * @p: target task
 * @preempt: whether task preemption is allowed or not
4992 4993 4994 4995
 *
 * 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.
 *
4996
 * Return:
4997 4998 4999
 *	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.
5000
 */
5001
int __sched yield_to(struct task_struct *p, bool preempt)
5002 5003 5004 5005
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5006
	int yielded = 0;
5007 5008 5009 5010 5011 5012

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5013 5014 5015 5016 5017 5018 5019 5020 5021
	/*
	 * 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;
	}

5022
	double_rq_lock(rq, p_rq);
5023
	if (task_rq(p) != p_rq) {
5024 5025 5026 5027 5028
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5029
		goto out_unlock;
5030 5031

	if (curr->sched_class != p->sched_class)
5032
		goto out_unlock;
5033 5034

	if (task_running(p_rq, p) || p->state)
5035
		goto out_unlock;
5036 5037

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5038
	if (yielded) {
5039
		schedstat_inc(rq->yld_count);
5040 5041 5042 5043 5044
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5045
			resched_curr(p_rq);
5046
	}
5047

5048
out_unlock:
5049
	double_rq_unlock(rq, p_rq);
5050
out_irq:
5051 5052
	local_irq_restore(flags);

5053
	if (yielded > 0)
5054 5055 5056 5057 5058 5059
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
5060
/*
I
Ingo Molnar 已提交
5061
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5062 5063 5064 5065
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5066 5067
	int old_iowait = current->in_iowait;
	struct rq *rq;
L
Linus Torvalds 已提交
5068 5069
	long ret;

5070
	current->in_iowait = 1;
5071
	blk_schedule_flush_plug(current);
5072

5073
	delayacct_blkio_start();
5074
	rq = raw_rq();
L
Linus Torvalds 已提交
5075 5076
	atomic_inc(&rq->nr_iowait);
	ret = schedule_timeout(timeout);
5077
	current->in_iowait = old_iowait;
L
Linus Torvalds 已提交
5078
	atomic_dec(&rq->nr_iowait);
5079
	delayacct_blkio_end();
5080

L
Linus Torvalds 已提交
5081 5082
	return ret;
}
5083
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5084 5085 5086 5087 5088

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5089 5090 5091
 * 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
5092
 */
5093
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5094 5095 5096 5097 5098 5099 5100 5101
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5102
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5103
	case SCHED_NORMAL:
5104
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5105
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5106 5107 5108 5109 5110 5111 5112 5113 5114 5115
		ret = 0;
		break;
	}
	return ret;
}

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

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5129
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5130
	case SCHED_NORMAL:
5131
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5132
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144
		ret = 0;
	}
	return ret;
}

/**
 * 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.
5145 5146 5147
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
L
Linus Torvalds 已提交
5148
 */
5149
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5150
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
5151
{
5152
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5153
	unsigned int time_slice;
5154 5155
	struct rq_flags rf;
	struct timespec t;
5156
	struct rq *rq;
5157
	int retval;
L
Linus Torvalds 已提交
5158 5159

	if (pid < 0)
5160
		return -EINVAL;
L
Linus Torvalds 已提交
5161 5162

	retval = -ESRCH;
5163
	rcu_read_lock();
L
Linus Torvalds 已提交
5164 5165 5166 5167 5168 5169 5170 5171
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5172
	rq = task_rq_lock(p, &rf);
5173 5174 5175
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5176
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5177

5178
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
5179
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
5180 5181
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
5182

L
Linus Torvalds 已提交
5183
out_unlock:
5184
	rcu_read_unlock();
L
Linus Torvalds 已提交
5185 5186 5187
	return retval;
}

5188
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5189

5190
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5191 5192
{
	unsigned long free = 0;
5193
	int ppid;
5194
	unsigned long state = p->state;
L
Linus Torvalds 已提交
5195

5196 5197
	if (state)
		state = __ffs(state) + 1;
5198
	printk(KERN_INFO "%-15.15s %c", p->comm,
5199
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5200
#if BITS_PER_LONG == 32
L
Linus Torvalds 已提交
5201
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5202
		printk(KERN_CONT " running  ");
L
Linus Torvalds 已提交
5203
	else
P
Peter Zijlstra 已提交
5204
		printk(KERN_CONT " %08lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5205 5206
#else
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5207
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5208
	else
P
Peter Zijlstra 已提交
5209
		printk(KERN_CONT " %016lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5210 5211
#endif
#ifdef CONFIG_DEBUG_STACK_USAGE
5212
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5213
#endif
5214
	ppid = 0;
5215
	rcu_read_lock();
5216 5217
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5218
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5219
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5220
		task_pid_nr(p), ppid,
5221
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5222

5223
	print_worker_info(KERN_INFO, p);
5224
	show_stack(p, NULL);
L
Linus Torvalds 已提交
5225 5226
}

I
Ingo Molnar 已提交
5227
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5228
{
5229
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5230

5231
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5232 5233
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5234
#else
P
Peter Zijlstra 已提交
5235 5236
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5237
#endif
5238
	rcu_read_lock();
5239
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5240 5241
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5242
		 * console might take a lot of time:
5243 5244 5245
		 * 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
5246 5247
		 */
		touch_nmi_watchdog();
5248
		touch_all_softlockup_watchdogs();
I
Ingo Molnar 已提交
5249
		if (!state_filter || (p->state & state_filter))
5250
			sched_show_task(p);
5251
	}
L
Linus Torvalds 已提交
5252

I
Ingo Molnar 已提交
5253
#ifdef CONFIG_SCHED_DEBUG
5254 5255
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5256
#endif
5257
	rcu_read_unlock();
I
Ingo Molnar 已提交
5258 5259 5260
	/*
	 * Only show locks if all tasks are dumped:
	 */
5261
	if (!state_filter)
I
Ingo Molnar 已提交
5262
		debug_show_all_locks();
L
Linus Torvalds 已提交
5263 5264
}

5265
void init_idle_bootup_task(struct task_struct *idle)
I
Ingo Molnar 已提交
5266
{
I
Ingo Molnar 已提交
5267
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
5268 5269
}

5270 5271 5272 5273 5274 5275 5276 5277
/**
 * init_idle - set up an idle thread for a given CPU
 * @idle: task in question
 * @cpu: cpu the idle task belongs to
 *
 * NOTE: this function does not set the idle thread's NEED_RESCHED
 * flag, to make booting more robust.
 */
5278
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5279
{
5280
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5281 5282
	unsigned long flags;

5283 5284
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5285

5286
	__sched_fork(0, idle);
5287
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5288 5289
	idle->se.exec_start = sched_clock();

5290 5291
	kasan_unpoison_task_stack(idle);

5292 5293 5294 5295 5296 5297 5298 5299 5300
#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
5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311
	/*
	 * We're having a chicken and egg problem, even though we are
	 * holding rq->lock, the cpu isn't yet set to this cpu so the
	 * 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
5312
	__set_task_cpu(idle, cpu);
5313
	rcu_read_unlock();
L
Linus Torvalds 已提交
5314 5315

	rq->curr = rq->idle = idle;
5316
	idle->on_rq = TASK_ON_RQ_QUEUED;
5317
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5318
	idle->on_cpu = 1;
5319
#endif
5320 5321
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5322 5323

	/* Set the preempt count _outside_ the spinlocks! */
5324
	init_idle_preempt_count(idle, cpu);
5325

I
Ingo Molnar 已提交
5326 5327 5328 5329
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5330
	ftrace_graph_init_idle_task(idle, cpu);
5331
	vtime_init_idle(idle, cpu);
5332
#ifdef CONFIG_SMP
5333 5334
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5335 5336
}

5337 5338 5339 5340 5341 5342 5343
int cpuset_cpumask_can_shrink(const struct cpumask *cur,
			      const struct cpumask *trial)
{
	int ret = 1, trial_cpus;
	struct dl_bw *cur_dl_b;
	unsigned long flags;

5344 5345 5346
	if (!cpumask_weight(cur))
		return ret;

5347
	rcu_read_lock_sched();
5348 5349 5350 5351 5352 5353 5354 5355
	cur_dl_b = dl_bw_of(cpumask_any(cur));
	trial_cpus = cpumask_weight(trial);

	raw_spin_lock_irqsave(&cur_dl_b->lock, flags);
	if (cur_dl_b->bw != -1 &&
	    cur_dl_b->bw * trial_cpus < cur_dl_b->total_bw)
		ret = 0;
	raw_spin_unlock_irqrestore(&cur_dl_b->lock, flags);
5356
	rcu_read_unlock_sched();
5357 5358 5359 5360

	return ret;
}

5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384
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
	 * to a new cpuset; we don't want to change their cpu
	 * 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;
	}

#ifdef CONFIG_SMP
	if (dl_task(p) && !cpumask_intersects(task_rq(p)->rd->span,
					      cs_cpus_allowed)) {
		unsigned int dest_cpu = cpumask_any_and(cpu_active_mask,
							cs_cpus_allowed);
5385
		struct dl_bw *dl_b;
5386 5387 5388 5389
		bool overflow;
		int cpus;
		unsigned long flags;

5390 5391
		rcu_read_lock_sched();
		dl_b = dl_bw_of(dest_cpu);
5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406
		raw_spin_lock_irqsave(&dl_b->lock, flags);
		cpus = dl_bw_cpus(dest_cpu);
		overflow = __dl_overflow(dl_b, cpus, 0, p->dl.dl_bw);
		if (overflow)
			ret = -EBUSY;
		else {
			/*
			 * We reserve space for this task in the destination
			 * root_domain, as we can't fail after this point.
			 * We will free resources in the source root_domain
			 * later on (see set_cpus_allowed_dl()).
			 */
			__dl_add(dl_b, p->dl.dl_bw);
		}
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
5407
		rcu_read_unlock_sched();
5408 5409 5410 5411 5412 5413 5414

	}
#endif
out:
	return ret;
}

L
Linus Torvalds 已提交
5415 5416
#ifdef CONFIG_SMP

5417 5418
static bool sched_smp_initialized __read_mostly;

5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433
#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;

	if (!cpumask_test_cpu(target_cpu, tsk_cpus_allowed(p)))
		return -EINVAL;

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

5434
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5435 5436
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5437 5438 5439 5440 5441 5442 5443

/*
 * 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)
{
5444
	bool queued, running;
5445 5446
	struct rq_flags rf;
	struct rq *rq;
5447

5448
	rq = task_rq_lock(p, &rf);
5449
	queued = task_on_rq_queued(p);
5450 5451
	running = task_current(rq, p);

5452
	if (queued)
5453
		dequeue_task(rq, p, DEQUEUE_SAVE);
5454
	if (running)
5455
		put_prev_task(rq, p);
5456 5457 5458

	p->numa_preferred_nid = nid;

5459
	if (queued)
5460
		enqueue_task(rq, p, ENQUEUE_RESTORE);
5461
	if (running)
5462
		set_curr_task(rq, p);
5463
	task_rq_unlock(rq, p, &rf);
5464
}
P
Peter Zijlstra 已提交
5465
#endif /* CONFIG_NUMA_BALANCING */
5466

L
Linus Torvalds 已提交
5467
#ifdef CONFIG_HOTPLUG_CPU
5468
/*
5469 5470
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
5471
 */
5472
void idle_task_exit(void)
L
Linus Torvalds 已提交
5473
{
5474
	struct mm_struct *mm = current->active_mm;
5475

5476
	BUG_ON(cpu_online(smp_processor_id()));
5477

5478
	if (mm != &init_mm) {
5479
		switch_mm_irqs_off(mm, &init_mm, current);
5480 5481
		finish_arch_post_lock_switch();
	}
5482
	mmdrop(mm);
L
Linus Torvalds 已提交
5483 5484 5485
}

/*
5486 5487
 * 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
5488 5489 5490
 * 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.
5491 5492
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5493
 */
5494
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5495
{
5496
	long delta = calc_load_fold_active(rq, 1);
5497 5498
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5499 5500
}

5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516
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,
};

5517
/*
5518 5519 5520 5521 5522 5523
 * 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
5524
 */
5525
static void migrate_tasks(struct rq *dead_rq)
L
Linus Torvalds 已提交
5526
{
5527
	struct rq *rq = dead_rq;
5528
	struct task_struct *next, *stop = rq->stop;
5529
	struct pin_cookie cookie;
5530
	int dest_cpu;
L
Linus Torvalds 已提交
5531 5532

	/*
5533 5534 5535 5536 5537 5538 5539
	 * 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
5540
	 */
5541
	rq->stop = NULL;
5542

5543 5544 5545 5546 5547 5548 5549
	/*
	 * 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);

5550
	for (;;) {
5551 5552 5553 5554 5555
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5556
			break;
5557

5558
		/*
W
Wanpeng Li 已提交
5559
		 * pick_next_task assumes pinned rq->lock.
5560
		 */
5561 5562
		cookie = lockdep_pin_lock(&rq->lock);
		next = pick_next_task(rq, &fake_task, cookie);
5563
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
5564
		next->sched_class->put_prev_task(rq, next);
5565

W
Wanpeng Li 已提交
5566 5567 5568 5569 5570 5571 5572 5573 5574
		/*
		 * 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.
		 */
5575
		lockdep_unpin_lock(&rq->lock, cookie);
W
Wanpeng Li 已提交
5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&next->pi_lock);
		raw_spin_lock(&rq->lock);

		/*
		 * 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;
		}

5590
		/* Find suitable destination for @next, with force if needed. */
5591
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5592

5593 5594 5595 5596 5597 5598
		rq = __migrate_task(rq, next, dest_cpu);
		if (rq != dead_rq) {
			raw_spin_unlock(&rq->lock);
			rq = dead_rq;
			raw_spin_lock(&rq->lock);
		}
W
Wanpeng Li 已提交
5599
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5600
	}
5601

5602
	rq->stop = stop;
5603
}
L
Linus Torvalds 已提交
5604 5605
#endif /* CONFIG_HOTPLUG_CPU */

5606 5607 5608 5609 5610
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

5611
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630
		rq->online = 1;

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

static void set_rq_offline(struct rq *rq)
{
	if (rq->online) {
		const struct sched_class *class;

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

5631
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5632 5633 5634 5635
		rq->online = 0;
	}
}

5636
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5637
{
5638
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5639

5640 5641 5642
	rq->age_stamp = sched_clock_cpu(cpu);
}

5643 5644
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

5645
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
5646

5647
static __read_mostly int sched_debug_enabled;
5648

5649
static int __init sched_debug_setup(char *str)
5650
{
5651
	sched_debug_enabled = 1;
5652 5653 5654

	return 0;
}
5655 5656 5657 5658 5659 5660
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
5661

5662
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
5663
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
5664
{
I
Ingo Molnar 已提交
5665
	struct sched_group *group = sd->groups;
L
Linus Torvalds 已提交
5666

5667
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
5668 5669 5670 5671

	printk(KERN_DEBUG "%*s domain %d: ", level, "", level);

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
5672
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
5673
		if (sd->parent)
P
Peter Zijlstra 已提交
5674 5675
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
5676
		return -1;
N
Nick Piggin 已提交
5677 5678
	}

5679 5680
	printk(KERN_CONT "span %*pbl level %s\n",
	       cpumask_pr_args(sched_domain_span(sd)), sd->name);
I
Ingo Molnar 已提交
5681

5682
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
5683 5684
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
5685
	}
5686
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5687 5688
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
5689
	}
L
Linus Torvalds 已提交
5690

I
Ingo Molnar 已提交
5691
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
5692
	do {
I
Ingo Molnar 已提交
5693
		if (!group) {
P
Peter Zijlstra 已提交
5694 5695
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
5696 5697 5698
			break;
		}

5699
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5700 5701
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
5702 5703
			break;
		}
L
Linus Torvalds 已提交
5704

5705 5706
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5707 5708
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
5709 5710
			break;
		}
L
Linus Torvalds 已提交
5711

5712
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
5713

5714 5715
		printk(KERN_CONT " %*pbl",
		       cpumask_pr_args(sched_group_cpus(group)));
5716
		if (group->sgc->capacity != SCHED_CAPACITY_SCALE) {
5717 5718
			printk(KERN_CONT " (cpu_capacity = %d)",
				group->sgc->capacity);
5719
		}
L
Linus Torvalds 已提交
5720

I
Ingo Molnar 已提交
5721 5722
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
5723
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
5724

5725
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
5726
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
5727

5728 5729
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
5730 5731
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
5732 5733
	return 0;
}
L
Linus Torvalds 已提交
5734

I
Ingo Molnar 已提交
5735 5736 5737
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
5738

5739
	if (!sched_debug_enabled)
5740 5741
		return;

I
Ingo Molnar 已提交
5742 5743 5744 5745
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
5746

I
Ingo Molnar 已提交
5747 5748 5749
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
5750
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
5751
			break;
L
Linus Torvalds 已提交
5752 5753
		level++;
		sd = sd->parent;
5754
		if (!sd)
I
Ingo Molnar 已提交
5755 5756
			break;
	}
L
Linus Torvalds 已提交
5757
}
5758
#else /* !CONFIG_SCHED_DEBUG */
5759 5760

# define sched_debug_enabled 0
5761
# define sched_domain_debug(sd, cpu) do { } while (0)
5762 5763 5764 5765
static inline bool sched_debug(void)
{
	return false;
}
5766
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
5767

5768
static int sd_degenerate(struct sched_domain *sd)
5769
{
5770
	if (cpumask_weight(sched_domain_span(sd)) == 1)
5771 5772 5773 5774 5775 5776
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
5777
			 SD_BALANCE_EXEC |
5778
			 SD_SHARE_CPUCAPACITY |
5779
			 SD_ASYM_CPUCAPACITY |
5780 5781
			 SD_SHARE_PKG_RESOURCES |
			 SD_SHARE_POWERDOMAIN)) {
5782 5783 5784 5785 5786
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
5787
	if (sd->flags & (SD_WAKE_AFFINE))
5788 5789 5790 5791 5792
		return 0;

	return 1;
}

5793 5794
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
5795 5796 5797 5798 5799 5800
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

5801
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
5802 5803 5804 5805 5806 5807 5808
		return 0;

	/* Flags needing groups don't count if only 1 group in parent */
	if (parent->groups == parent->groups->next) {
		pflags &= ~(SD_LOAD_BALANCE |
				SD_BALANCE_NEWIDLE |
				SD_BALANCE_FORK |
5809
				SD_BALANCE_EXEC |
5810
				SD_ASYM_CPUCAPACITY |
5811
				SD_SHARE_CPUCAPACITY |
5812
				SD_SHARE_PKG_RESOURCES |
5813 5814
				SD_PREFER_SIBLING |
				SD_SHARE_POWERDOMAIN);
5815 5816
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
5817 5818 5819 5820 5821 5822 5823
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

5824
static void free_rootdomain(struct rcu_head *rcu)
5825
{
5826
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
5827

5828
	cpupri_cleanup(&rd->cpupri);
5829
	cpudl_cleanup(&rd->cpudl);
5830
	free_cpumask_var(rd->dlo_mask);
5831 5832 5833 5834 5835 5836
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
5837 5838
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
5839
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
5840 5841
	unsigned long flags;

5842
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5843 5844

	if (rq->rd) {
I
Ingo Molnar 已提交
5845
		old_rd = rq->rd;
G
Gregory Haskins 已提交
5846

5847
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
5848
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5849

5850
		cpumask_clear_cpu(rq->cpu, old_rd->span);
5851

I
Ingo Molnar 已提交
5852
		/*
5853
		 * If we dont want to free the old_rd yet then
I
Ingo Molnar 已提交
5854 5855 5856 5857 5858
		 * set old_rd to NULL to skip the freeing later
		 * in this function:
		 */
		if (!atomic_dec_and_test(&old_rd->refcount))
			old_rd = NULL;
G
Gregory Haskins 已提交
5859 5860 5861 5862 5863
	}

	atomic_inc(&rd->refcount);
	rq->rd = rd;

5864
	cpumask_set_cpu(rq->cpu, rd->span);
5865
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
5866
		set_rq_online(rq);
G
Gregory Haskins 已提交
5867

5868
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
5869 5870

	if (old_rd)
5871
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
5872 5873
}

5874
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
5875 5876 5877
{
	memset(rd, 0, sizeof(*rd));

5878
	if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
5879
		goto out;
5880
	if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
5881
		goto free_span;
5882
	if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
5883
		goto free_online;
5884
	if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
5885
		goto free_dlo_mask;
5886

5887
	init_dl_bw(&rd->dl_bw);
5888 5889
	if (cpudl_init(&rd->cpudl) != 0)
		goto free_dlo_mask;
5890

5891
	if (cpupri_init(&rd->cpupri) != 0)
5892
		goto free_rto_mask;
5893
	return 0;
5894

5895 5896
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
5897 5898
free_dlo_mask:
	free_cpumask_var(rd->dlo_mask);
5899 5900 5901 5902
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
5903
out:
5904
	return -ENOMEM;
G
Gregory Haskins 已提交
5905 5906
}

5907 5908 5909 5910 5911 5912
/*
 * By default the system creates a single root-domain with all cpus as
 * members (mimicking the global state we have today).
 */
struct root_domain def_root_domain;

G
Gregory Haskins 已提交
5913 5914
static void init_defrootdomain(void)
{
5915
	init_rootdomain(&def_root_domain);
5916

G
Gregory Haskins 已提交
5917 5918 5919
	atomic_set(&def_root_domain.refcount, 1);
}

5920
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5921 5922 5923 5924 5925 5926 5927
{
	struct root_domain *rd;

	rd = kmalloc(sizeof(*rd), GFP_KERNEL);
	if (!rd)
		return NULL;

5928
	if (init_rootdomain(rd) != 0) {
5929 5930 5931
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5932 5933 5934 5935

	return rd;
}

5936
static void free_sched_groups(struct sched_group *sg, int free_sgc)
5937 5938 5939 5940 5941 5942 5943 5944 5945 5946
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

	first = sg;
	do {
		tmp = sg->next;

5947 5948
		if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
			kfree(sg->sgc);
5949 5950 5951 5952 5953 5954

		kfree(sg);
		sg = tmp;
	} while (sg != first);
}

5955
static void destroy_sched_domain(struct sched_domain *sd)
5956
{
5957 5958 5959 5960 5961 5962 5963
	/*
	 * If its an overlapping domain it has private groups, iterate and
	 * nuke them all.
	 */
	if (sd->flags & SD_OVERLAP) {
		free_sched_groups(sd->groups, 1);
	} else if (atomic_dec_and_test(&sd->groups->ref)) {
5964
		kfree(sd->groups->sgc);
5965
		kfree(sd->groups);
5966
	}
5967 5968
	if (sd->shared && atomic_dec_and_test(&sd->shared->ref))
		kfree(sd->shared);
5969 5970 5971
	kfree(sd);
}

5972
static void destroy_sched_domains_rcu(struct rcu_head *rcu)
5973
{
5974 5975 5976 5977 5978 5979 5980
	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);

	while (sd) {
		struct sched_domain *parent = sd->parent;
		destroy_sched_domain(sd);
		sd = parent;
	}
5981 5982
}

5983
static void destroy_sched_domains(struct sched_domain *sd)
5984
{
5985 5986
	if (sd)
		call_rcu(&sd->rcu, destroy_sched_domains_rcu);
5987 5988
}

5989 5990 5991 5992 5993 5994 5995
/*
 * Keep a special pointer to the highest sched_domain that has
 * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
 * allows us to avoid some pointer chasing select_idle_sibling().
 *
 * Also keep a unique ID per domain (we use the first cpu number in
 * the cpumask of the domain), this allows us to quickly tell if
5996
 * two cpus are in the same cache domain, see cpus_share_cache().
5997 5998
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
5999
DEFINE_PER_CPU(int, sd_llc_size);
6000
DEFINE_PER_CPU(int, sd_llc_id);
6001
DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
6002
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
6003
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
6004 6005 6006

static void update_top_cache_domain(int cpu)
{
6007
	struct sched_domain_shared *sds = NULL;
6008 6009
	struct sched_domain *sd;
	int id = cpu;
6010
	int size = 1;
6011 6012

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
6013
	if (sd) {
6014
		id = cpumask_first(sched_domain_span(sd));
6015
		size = cpumask_weight(sched_domain_span(sd));
6016
		sds = sd->shared;
6017
	}
6018 6019

	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
6020
	per_cpu(sd_llc_size, cpu) = size;
6021
	per_cpu(sd_llc_id, cpu) = id;
6022
	rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds);
6023 6024 6025

	sd = lowest_flag_domain(cpu, SD_NUMA);
	rcu_assign_pointer(per_cpu(sd_numa, cpu), sd);
6026 6027 6028

	sd = highest_flag_domain(cpu, SD_ASYM_PACKING);
	rcu_assign_pointer(per_cpu(sd_asym, cpu), sd);
6029 6030
}

L
Linus Torvalds 已提交
6031
/*
I
Ingo Molnar 已提交
6032
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
6033 6034
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
6035 6036
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
6037
{
6038
	struct rq *rq = cpu_rq(cpu);
6039 6040 6041
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
6042
	for (tmp = sd; tmp; ) {
6043 6044 6045
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
6046

6047
		if (sd_parent_degenerate(tmp, parent)) {
6048
			tmp->parent = parent->parent;
6049 6050
			if (parent->parent)
				parent->parent->child = tmp;
6051 6052 6053 6054 6055 6056 6057
			/*
			 * Transfer SD_PREFER_SIBLING down in case of a
			 * degenerate parent; the spans match for this
			 * so the property transfers.
			 */
			if (parent->flags & SD_PREFER_SIBLING)
				tmp->flags |= SD_PREFER_SIBLING;
6058
			destroy_sched_domain(parent);
6059 6060
		} else
			tmp = tmp->parent;
6061 6062
	}

6063
	if (sd && sd_degenerate(sd)) {
6064
		tmp = sd;
6065
		sd = sd->parent;
6066
		destroy_sched_domain(tmp);
6067 6068 6069
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
6070

6071
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
6072

G
Gregory Haskins 已提交
6073
	rq_attach_root(rq, rd);
6074
	tmp = rq->sd;
N
Nick Piggin 已提交
6075
	rcu_assign_pointer(rq->sd, sd);
6076
	destroy_sched_domains(tmp);
6077 6078

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
6079 6080 6081 6082 6083
}

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
6084 6085
	int ret;

R
Rusty Russell 已提交
6086
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
6087 6088 6089 6090 6091
	ret = cpulist_parse(str, cpu_isolated_map);
	if (ret) {
		pr_err("sched: Error, all isolcpus= values must be between 0 and %d\n", nr_cpu_ids);
		return 0;
	}
L
Linus Torvalds 已提交
6092 6093
	return 1;
}
I
Ingo Molnar 已提交
6094
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
6095

6096
struct s_data {
6097
	struct sched_domain ** __percpu sd;
6098 6099 6100
	struct root_domain	*rd;
};

6101 6102
enum s_alloc {
	sa_rootdomain,
6103
	sa_sd,
6104
	sa_sd_storage,
6105 6106 6107
	sa_none,
};

P
Peter Zijlstra 已提交
6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145
/*
 * Build an iteration mask that can exclude certain CPUs from the upwards
 * domain traversal.
 *
 * Asymmetric node setups can result in situations where the domain tree is of
 * unequal depth, make sure to skip domains that already cover the entire
 * range.
 *
 * In that case build_sched_domains() will have terminated the iteration early
 * and our sibling sd spans will be empty. Domains should always include the
 * cpu they're built on, so check that.
 *
 */
static void build_group_mask(struct sched_domain *sd, struct sched_group *sg)
{
	const struct cpumask *span = sched_domain_span(sd);
	struct sd_data *sdd = sd->private;
	struct sched_domain *sibling;
	int i;

	for_each_cpu(i, span) {
		sibling = *per_cpu_ptr(sdd->sd, i);
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
			continue;

		cpumask_set_cpu(i, sched_group_mask(sg));
	}
}

/*
 * Return the canonical balance cpu for this group, this is the first cpu
 * of this group that's also in the iteration mask.
 */
int group_balance_cpu(struct sched_group *sg)
{
	return cpumask_first_and(sched_group_cpus(sg), sched_group_mask(sg));
}

6146 6147 6148 6149 6150 6151 6152
static int
build_overlap_sched_groups(struct sched_domain *sd, int cpu)
{
	struct sched_group *first = NULL, *last = NULL, *groups = NULL, *sg;
	const struct cpumask *span = sched_domain_span(sd);
	struct cpumask *covered = sched_domains_tmpmask;
	struct sd_data *sdd = sd->private;
6153
	struct sched_domain *sibling;
6154 6155 6156 6157 6158 6159 6160 6161 6162 6163
	int i;

	cpumask_clear(covered);

	for_each_cpu(i, span) {
		struct cpumask *sg_span;

		if (cpumask_test_cpu(i, covered))
			continue;

6164
		sibling = *per_cpu_ptr(sdd->sd, i);
P
Peter Zijlstra 已提交
6165 6166

		/* See the comment near build_group_mask(). */
6167
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
P
Peter Zijlstra 已提交
6168 6169
			continue;

6170
		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
6171
				GFP_KERNEL, cpu_to_node(cpu));
6172 6173 6174 6175 6176

		if (!sg)
			goto fail;

		sg_span = sched_group_cpus(sg);
6177 6178 6179
		if (sibling->child)
			cpumask_copy(sg_span, sched_domain_span(sibling->child));
		else
6180 6181 6182 6183
			cpumask_set_cpu(i, sg_span);

		cpumask_or(covered, covered, sg_span);

6184 6185
		sg->sgc = *per_cpu_ptr(sdd->sgc, i);
		if (atomic_inc_return(&sg->sgc->ref) == 1)
P
Peter Zijlstra 已提交
6186 6187
			build_group_mask(sd, sg);

6188
		/*
6189
		 * Initialize sgc->capacity such that even if we mess up the
6190 6191 6192
		 * domains and no possible iteration will get us here, we won't
		 * die on a /0 trap.
		 */
6193
		sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span);
6194

P
Peter Zijlstra 已提交
6195 6196 6197 6198 6199
		/*
		 * Make sure the first group of this domain contains the
		 * canonical balance cpu. Otherwise the sched_domain iteration
		 * breaks. See update_sg_lb_stats().
		 */
P
Peter Zijlstra 已提交
6200
		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
P
Peter Zijlstra 已提交
6201
		    group_balance_cpu(sg) == cpu)
6202 6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220
			groups = sg;

		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
		last->next = first;
	}
	sd->groups = groups;

	return 0;

fail:
	free_sched_groups(first, 0);

	return -ENOMEM;
}

6221
static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
L
Linus Torvalds 已提交
6222
{
6223 6224
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	struct sched_domain *child = sd->child;
L
Linus Torvalds 已提交
6225

6226 6227
	if (child)
		cpu = cpumask_first(sched_domain_span(child));
6228

6229
	if (sg) {
6230
		*sg = *per_cpu_ptr(sdd->sg, cpu);
6231 6232
		(*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu);
		atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */
6233
	}
6234 6235

	return cpu;
6236 6237
}

6238
/*
6239 6240
 * build_sched_groups will build a circular linked list of the groups
 * covered by the given span, and will set each group's ->cpumask correctly,
6241
 * and ->cpu_capacity to 0.
6242 6243
 *
 * Assumes the sched_domain tree is fully constructed
6244
 */
6245 6246
static int
build_sched_groups(struct sched_domain *sd, int cpu)
L
Linus Torvalds 已提交
6247
{
6248 6249 6250
	struct sched_group *first = NULL, *last = NULL;
	struct sd_data *sdd = sd->private;
	const struct cpumask *span = sched_domain_span(sd);
6251
	struct cpumask *covered;
6252
	int i;
6253

6254 6255 6256
	get_group(cpu, sdd, &sd->groups);
	atomic_inc(&sd->groups->ref);

6257
	if (cpu != cpumask_first(span))
6258 6259
		return 0;

6260 6261 6262
	lockdep_assert_held(&sched_domains_mutex);
	covered = sched_domains_tmpmask;

6263
	cpumask_clear(covered);
6264

6265 6266
	for_each_cpu(i, span) {
		struct sched_group *sg;
6267
		int group, j;
6268

6269 6270
		if (cpumask_test_cpu(i, covered))
			continue;
6271

6272
		group = get_group(i, sdd, &sg);
P
Peter Zijlstra 已提交
6273
		cpumask_setall(sched_group_mask(sg));
6274

6275 6276 6277
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
				continue;
6278

6279 6280 6281
			cpumask_set_cpu(j, covered);
			cpumask_set_cpu(j, sched_group_cpus(sg));
		}
6282

6283 6284 6285 6286 6287 6288 6289
		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
	}
	last->next = first;
6290 6291

	return 0;
6292
}
6293

6294
/*
6295
 * Initialize sched groups cpu_capacity.
6296
 *
6297
 * cpu_capacity indicates the capacity of sched group, which is used while
6298
 * distributing the load between different sched groups in a sched domain.
6299 6300 6301 6302
 * Typically cpu_capacity for all the groups in a sched domain will be same
 * unless there are asymmetries in the topology. If there are asymmetries,
 * group having more cpu_capacity will pickup more load compared to the
 * group having less cpu_capacity.
6303
 */
6304
static void init_sched_groups_capacity(int cpu, struct sched_domain *sd)
6305
{
6306
	struct sched_group *sg = sd->groups;
6307

6308
	WARN_ON(!sg);
6309 6310 6311 6312 6313

	do {
		sg->group_weight = cpumask_weight(sched_group_cpus(sg));
		sg = sg->next;
	} while (sg != sd->groups);
6314

P
Peter Zijlstra 已提交
6315
	if (cpu != group_balance_cpu(sg))
6316
		return;
6317

6318
	update_group_capacity(sd, cpu);
6319 6320
}

6321 6322 6323 6324 6325
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

6326
static int default_relax_domain_level = -1;
6327
int sched_domain_level_max;
6328 6329 6330

static int __init setup_relax_domain_level(char *str)
{
6331 6332
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
6333

6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351
	return 1;
}
__setup("relax_domain_level=", setup_relax_domain_level);

static void set_domain_attribute(struct sched_domain *sd,
				 struct sched_domain_attr *attr)
{
	int request;

	if (!attr || attr->relax_domain_level < 0) {
		if (default_relax_domain_level < 0)
			return;
		else
			request = default_relax_domain_level;
	} else
		request = attr->relax_domain_level;
	if (request < sd->level) {
		/* turn off idle balance on this domain */
6352
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6353 6354
	} else {
		/* turn on idle balance on this domain */
6355
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6356 6357 6358
	}
}

6359 6360 6361
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

6362 6363 6364 6365 6366
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
6367 6368
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
6369 6370
	case sa_sd:
		free_percpu(d->sd); /* fall through */
6371
	case sa_sd_storage:
6372
		__sdt_free(cpu_map); /* fall through */
6373 6374 6375 6376
	case sa_none:
		break;
	}
}
6377

6378 6379 6380
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
6381 6382
	memset(d, 0, sizeof(*d));

6383 6384
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
6385 6386 6387
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
6388
	d->rd = alloc_rootdomain();
6389
	if (!d->rd)
6390
		return sa_sd;
6391 6392
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
6393

6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405
/*
 * NULL the sd_data elements we've used to build the sched_domain and
 * sched_group structure so that the subsequent __free_domain_allocs()
 * will not free the data we're using.
 */
static void claim_allocations(int cpu, struct sched_domain *sd)
{
	struct sd_data *sdd = sd->private;

	WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd);
	*per_cpu_ptr(sdd->sd, cpu) = NULL;

6406 6407 6408
	if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref))
		*per_cpu_ptr(sdd->sds, cpu) = NULL;

6409
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
6410
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
6411

6412 6413
	if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
		*per_cpu_ptr(sdd->sgc, cpu) = NULL;
6414 6415
}

6416 6417
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
6418
enum numa_topology_type sched_numa_topology_type;
6419
static int *sched_domains_numa_distance;
6420
int sched_max_numa_distance;
6421 6422
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
6423
#endif
6424

6425 6426 6427
/*
 * SD_flags allowed in topology descriptions.
 *
6428 6429 6430
 * These flags are purely descriptive of the topology and do not prescribe
 * behaviour. Behaviour is artificial and mapped in the below sd_init()
 * function:
6431
 *
6432 6433 6434 6435
 *   SD_SHARE_CPUCAPACITY   - describes SMT topologies
 *   SD_SHARE_PKG_RESOURCES - describes shared caches
 *   SD_NUMA                - describes NUMA topologies
 *   SD_SHARE_POWERDOMAIN   - describes shared power domain
6436
 *   SD_ASYM_CPUCAPACITY    - describes mixed capacity topologies
6437 6438 6439
 *
 * Odd one out, which beside describing the topology has a quirk also
 * prescribes the desired behaviour that goes along with it:
6440
 *
6441
 *   SD_ASYM_PACKING        - describes SMT quirks
6442 6443
 */
#define TOPOLOGY_SD_FLAGS		\
6444
	(SD_SHARE_CPUCAPACITY |		\
6445 6446
	 SD_SHARE_PKG_RESOURCES |	\
	 SD_NUMA |			\
6447
	 SD_ASYM_PACKING |		\
6448
	 SD_ASYM_CPUCAPACITY |		\
6449
	 SD_SHARE_POWERDOMAIN)
6450 6451

static struct sched_domain *
6452
sd_init(struct sched_domain_topology_level *tl,
6453
	const struct cpumask *cpu_map,
6454
	struct sched_domain *child, int cpu)
6455
{
6456 6457 6458
	struct sd_data *sdd = &tl->data;
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	int sd_id, sd_weight, sd_flags = 0;
6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473

#ifdef CONFIG_NUMA
	/*
	 * Ugly hack to pass state to sd_numa_mask()...
	 */
	sched_domains_curr_level = tl->numa_level;
#endif

	sd_weight = cpumask_weight(tl->mask(cpu));

	if (tl->sd_flags)
		sd_flags = (*tl->sd_flags)();
	if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS,
			"wrong sd_flags in topology description\n"))
		sd_flags &= ~TOPOLOGY_SD_FLAGS;
6474 6475 6476 6477 6478

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
6479
		.imbalance_pct		= 125,
6480 6481 6482 6483

		.cache_nice_tries	= 0,
		.busy_idx		= 0,
		.idle_idx		= 0,
6484 6485 6486 6487 6488 6489
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
6490 6491
					| 1*SD_BALANCE_EXEC
					| 1*SD_BALANCE_FORK
6492
					| 0*SD_BALANCE_WAKE
6493
					| 1*SD_WAKE_AFFINE
6494
					| 0*SD_SHARE_CPUCAPACITY
6495
					| 0*SD_SHARE_PKG_RESOURCES
6496
					| 0*SD_SERIALIZE
6497
					| 0*SD_PREFER_SIBLING
6498 6499
					| 0*SD_NUMA
					| sd_flags
6500
					,
6501

6502 6503
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
6504
		.smt_gain		= 0,
6505 6506
		.max_newidle_lb_cost	= 0,
		.next_decay_max_lb_cost	= jiffies,
6507
		.child			= child,
6508 6509 6510
#ifdef CONFIG_SCHED_DEBUG
		.name			= tl->name,
#endif
6511 6512
	};

6513 6514 6515
	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
	sd_id = cpumask_first(sched_domain_span(sd));

6516
	/*
6517
	 * Convert topological properties into behaviour.
6518
	 */
6519

6520 6521 6522 6523 6524 6525 6526
	if (sd->flags & SD_ASYM_CPUCAPACITY) {
		struct sched_domain *t = sd;

		for_each_lower_domain(t)
			t->flags |= SD_BALANCE_WAKE;
	}

6527
	if (sd->flags & SD_SHARE_CPUCAPACITY) {
6528
		sd->flags |= SD_PREFER_SIBLING;
6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557
		sd->imbalance_pct = 110;
		sd->smt_gain = 1178; /* ~15% */

	} else if (sd->flags & SD_SHARE_PKG_RESOURCES) {
		sd->imbalance_pct = 117;
		sd->cache_nice_tries = 1;
		sd->busy_idx = 2;

#ifdef CONFIG_NUMA
	} else if (sd->flags & SD_NUMA) {
		sd->cache_nice_tries = 2;
		sd->busy_idx = 3;
		sd->idle_idx = 2;

		sd->flags |= SD_SERIALIZE;
		if (sched_domains_numa_distance[tl->numa_level] > RECLAIM_DISTANCE) {
			sd->flags &= ~(SD_BALANCE_EXEC |
				       SD_BALANCE_FORK |
				       SD_WAKE_AFFINE);
		}

#endif
	} else {
		sd->flags |= SD_PREFER_SIBLING;
		sd->cache_nice_tries = 1;
		sd->busy_idx = 2;
		sd->idle_idx = 1;
	}

6558 6559 6560 6561 6562 6563 6564
	/*
	 * For all levels sharing cache; connect a sched_domain_shared
	 * instance.
	 */
	if (sd->flags & SD_SHARE_PKG_RESOURCES) {
		sd->shared = *per_cpu_ptr(sdd->sds, sd_id);
		atomic_inc(&sd->shared->ref);
6565
		atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
6566 6567 6568
	}

	sd->private = sdd;
6569 6570 6571 6572

	return sd;
}

6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586
/*
 * Topology list, bottom-up.
 */
static struct sched_domain_topology_level default_topology[] = {
#ifdef CONFIG_SCHED_SMT
	{ cpu_smt_mask, cpu_smt_flags, SD_INIT_NAME(SMT) },
#endif
#ifdef CONFIG_SCHED_MC
	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
#endif
	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
	{ NULL, },
};

6587 6588
static struct sched_domain_topology_level *sched_domain_topology =
	default_topology;
6589 6590 6591 6592 6593 6594

#define for_each_sd_topology(tl)			\
	for (tl = sched_domain_topology; tl->mask; tl++)

void set_sched_topology(struct sched_domain_topology_level *tl)
{
6595 6596 6597
	if (WARN_ON_ONCE(sched_smp_initialized))
		return;

6598 6599 6600 6601 6602
	sched_domain_topology = tl;
}

#ifdef CONFIG_NUMA

6603 6604 6605 6606 6607
static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628
static void sched_numa_warn(const char *str)
{
	static int done = false;
	int i,j;

	if (done)
		return;

	done = true;

	printk(KERN_WARNING "ERROR: %s\n\n", str);

	for (i = 0; i < nr_node_ids; i++) {
		printk(KERN_WARNING "  ");
		for (j = 0; j < nr_node_ids; j++)
			printk(KERN_CONT "%02d ", node_distance(i,j));
		printk(KERN_CONT "\n");
	}
	printk(KERN_WARNING "\n");
}

6629
bool find_numa_distance(int distance)
6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643
{
	int i;

	if (distance == node_distance(0, 0))
		return true;

	for (i = 0; i < sched_domains_numa_levels; i++) {
		if (sched_domains_numa_distance[i] == distance)
			return true;
	}

	return false;
}

6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668
/*
 * A system can have three types of NUMA topology:
 * NUMA_DIRECT: all nodes are directly connected, or not a NUMA system
 * NUMA_GLUELESS_MESH: some nodes reachable through intermediary nodes
 * NUMA_BACKPLANE: nodes can reach other nodes through a backplane
 *
 * The difference between a glueless mesh topology and a backplane
 * topology lies in whether communication between not directly
 * connected nodes goes through intermediary nodes (where programs
 * could run), or through backplane controllers. This affects
 * placement of programs.
 *
 * The type of topology can be discerned with the following tests:
 * - If the maximum distance between any nodes is 1 hop, the system
 *   is directly connected.
 * - If for two nodes A and B, located N > 1 hops away from each other,
 *   there is an intermediary node C, which is < N hops away from both
 *   nodes A and B, the system is a glueless mesh.
 */
static void init_numa_topology_type(void)
{
	int a, b, c, n;

	n = sched_max_numa_distance;

6669
	if (sched_domains_numa_levels <= 1) {
6670
		sched_numa_topology_type = NUMA_DIRECT;
6671 6672
		return;
	}
6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695

	for_each_online_node(a) {
		for_each_online_node(b) {
			/* Find two nodes furthest removed from each other. */
			if (node_distance(a, b) < n)
				continue;

			/* Is there an intermediary node between a and b? */
			for_each_online_node(c) {
				if (node_distance(a, c) < n &&
				    node_distance(b, c) < n) {
					sched_numa_topology_type =
							NUMA_GLUELESS_MESH;
					return;
				}
			}

			sched_numa_topology_type = NUMA_BACKPLANE;
			return;
		}
	}
}

6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716
static void sched_init_numa(void)
{
	int next_distance, curr_distance = node_distance(0, 0);
	struct sched_domain_topology_level *tl;
	int level = 0;
	int i, j, k;

	sched_domains_numa_distance = kzalloc(sizeof(int) * nr_node_ids, GFP_KERNEL);
	if (!sched_domains_numa_distance)
		return;

	/*
	 * O(nr_nodes^2) deduplicating selection sort -- in order to find the
	 * unique distances in the node_distance() table.
	 *
	 * Assumes node_distance(0,j) includes all distances in
	 * node_distance(i,j) in order to avoid cubic time.
	 */
	next_distance = curr_distance;
	for (i = 0; i < nr_node_ids; i++) {
		for (j = 0; j < nr_node_ids; j++) {
6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740
			for (k = 0; k < nr_node_ids; k++) {
				int distance = node_distance(i, k);

				if (distance > curr_distance &&
				    (distance < next_distance ||
				     next_distance == curr_distance))
					next_distance = distance;

				/*
				 * While not a strong assumption it would be nice to know
				 * about cases where if node A is connected to B, B is not
				 * equally connected to A.
				 */
				if (sched_debug() && node_distance(k, i) != distance)
					sched_numa_warn("Node-distance not symmetric");

				if (sched_debug() && i && !find_numa_distance(distance))
					sched_numa_warn("Node-0 not representative");
			}
			if (next_distance != curr_distance) {
				sched_domains_numa_distance[level++] = next_distance;
				sched_domains_numa_levels = level;
				curr_distance = next_distance;
			} else break;
6741
		}
6742 6743 6744 6745 6746 6747

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
6748
	}
6749 6750 6751 6752

	if (!level)
		return;

6753 6754 6755 6756
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
6757
	 * The sched_domains_numa_distance[] array includes the actual distance
6758 6759 6760
	 * numbers.
	 */

6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771
	/*
	 * Here, we should temporarily reset sched_domains_numa_levels to 0.
	 * If it fails to allocate memory for array sched_domains_numa_masks[][],
	 * the array will contain less then 'level' members. This could be
	 * dangerous when we use it to iterate array sched_domains_numa_masks[][]
	 * in other functions.
	 *
	 * We reset it to 'level' at the end of this function.
	 */
	sched_domains_numa_levels = 0;

6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786
	sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL);
	if (!sched_domains_numa_masks)
		return;

	/*
	 * Now for each level, construct a mask per node which contains all
	 * cpus of nodes that are that many hops away from us.
	 */
	for (i = 0; i < level; i++) {
		sched_domains_numa_masks[i] =
			kzalloc(nr_node_ids * sizeof(void *), GFP_KERNEL);
		if (!sched_domains_numa_masks[i])
			return;

		for (j = 0; j < nr_node_ids; j++) {
6787
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
6788 6789 6790 6791 6792
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

6793
			for_each_node(k) {
6794
				if (node_distance(j, k) > sched_domains_numa_distance[i])
6795 6796 6797 6798 6799 6800 6801
					continue;

				cpumask_or(mask, mask, cpumask_of_node(k));
			}
		}
	}

6802 6803 6804
	/* Compute default topology size */
	for (i = 0; sched_domain_topology[i].mask; i++);

6805
	tl = kzalloc((i + level + 1) *
6806 6807 6808 6809 6810 6811 6812
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
6813 6814
	for (i = 0; sched_domain_topology[i].mask; i++)
		tl[i] = sched_domain_topology[i];
6815 6816 6817 6818 6819 6820 6821

	/*
	 * .. and append 'j' levels of NUMA goodness.
	 */
	for (j = 0; j < level; i++, j++) {
		tl[i] = (struct sched_domain_topology_level){
			.mask = sd_numa_mask,
6822
			.sd_flags = cpu_numa_flags,
6823 6824
			.flags = SDTL_OVERLAP,
			.numa_level = j,
6825
			SD_INIT_NAME(NUMA)
6826 6827 6828 6829
		};
	}

	sched_domain_topology = tl;
6830 6831

	sched_domains_numa_levels = level;
6832
	sched_max_numa_distance = sched_domains_numa_distance[level - 1];
6833 6834

	init_numa_topology_type();
6835
}
6836

6837
static void sched_domains_numa_masks_set(unsigned int cpu)
6838 6839
{
	int node = cpu_to_node(cpu);
6840
	int i, j;
6841 6842 6843 6844 6845 6846 6847 6848 6849

	for (i = 0; i < sched_domains_numa_levels; i++) {
		for (j = 0; j < nr_node_ids; j++) {
			if (node_distance(j, node) <= sched_domains_numa_distance[i])
				cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
		}
	}
}

6850
static void sched_domains_numa_masks_clear(unsigned int cpu)
6851 6852
{
	int i, j;
6853

6854 6855 6856 6857 6858 6859
	for (i = 0; i < sched_domains_numa_levels; i++) {
		for (j = 0; j < nr_node_ids; j++)
			cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]);
	}
}

6860
#else
6861 6862 6863
static inline void sched_init_numa(void) { }
static void sched_domains_numa_masks_set(unsigned int cpu) { }
static void sched_domains_numa_masks_clear(unsigned int cpu) { }
6864 6865
#endif /* CONFIG_NUMA */

6866 6867 6868 6869 6870
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6871
	for_each_sd_topology(tl) {
6872 6873 6874 6875 6876 6877
		struct sd_data *sdd = &tl->data;

		sdd->sd = alloc_percpu(struct sched_domain *);
		if (!sdd->sd)
			return -ENOMEM;

6878 6879 6880 6881
		sdd->sds = alloc_percpu(struct sched_domain_shared *);
		if (!sdd->sds)
			return -ENOMEM;

6882 6883 6884 6885
		sdd->sg = alloc_percpu(struct sched_group *);
		if (!sdd->sg)
			return -ENOMEM;

6886 6887
		sdd->sgc = alloc_percpu(struct sched_group_capacity *);
		if (!sdd->sgc)
6888 6889
			return -ENOMEM;

6890 6891
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
6892
			struct sched_domain_shared *sds;
6893
			struct sched_group *sg;
6894
			struct sched_group_capacity *sgc;
6895

P
Peter Zijlstra 已提交
6896
			sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6897 6898 6899 6900 6901 6902
					GFP_KERNEL, cpu_to_node(j));
			if (!sd)
				return -ENOMEM;

			*per_cpu_ptr(sdd->sd, j) = sd;

6903 6904 6905 6906 6907 6908 6909
			sds = kzalloc_node(sizeof(struct sched_domain_shared),
					GFP_KERNEL, cpu_to_node(j));
			if (!sds)
				return -ENOMEM;

			*per_cpu_ptr(sdd->sds, j) = sds;

6910 6911 6912 6913 6914
			sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
					GFP_KERNEL, cpu_to_node(j));
			if (!sg)
				return -ENOMEM;

6915 6916
			sg->next = sg;

6917
			*per_cpu_ptr(sdd->sg, j) = sg;
6918

6919
			sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
6920
					GFP_KERNEL, cpu_to_node(j));
6921
			if (!sgc)
6922 6923
				return -ENOMEM;

6924
			*per_cpu_ptr(sdd->sgc, j) = sgc;
6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935
		}
	}

	return 0;
}

static void __sdt_free(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6936
	for_each_sd_topology(tl) {
6937 6938 6939
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
6940 6941 6942 6943 6944 6945 6946 6947 6948
			struct sched_domain *sd;

			if (sdd->sd) {
				sd = *per_cpu_ptr(sdd->sd, j);
				if (sd && (sd->flags & SD_OVERLAP))
					free_sched_groups(sd->groups, 0);
				kfree(*per_cpu_ptr(sdd->sd, j));
			}

6949 6950
			if (sdd->sds)
				kfree(*per_cpu_ptr(sdd->sds, j));
6951 6952
			if (sdd->sg)
				kfree(*per_cpu_ptr(sdd->sg, j));
6953 6954
			if (sdd->sgc)
				kfree(*per_cpu_ptr(sdd->sgc, j));
6955 6956
		}
		free_percpu(sdd->sd);
6957
		sdd->sd = NULL;
6958 6959
		free_percpu(sdd->sds);
		sdd->sds = NULL;
6960
		free_percpu(sdd->sg);
6961
		sdd->sg = NULL;
6962 6963
		free_percpu(sdd->sgc);
		sdd->sgc = NULL;
6964 6965 6966
	}
}

6967
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6968 6969
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
6970
{
6971
	struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu);
6972

6973 6974 6975
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
6976
		child->parent = sd;
P
Peter Zijlstra 已提交
6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990

		if (!cpumask_subset(sched_domain_span(child),
				    sched_domain_span(sd))) {
			pr_err("BUG: arch topology borken\n");
#ifdef CONFIG_SCHED_DEBUG
			pr_err("     the %s domain not a subset of the %s domain\n",
					child->name, sd->name);
#endif
			/* Fixup, ensure @sd has at least @child cpus. */
			cpumask_or(sched_domain_span(sd),
				   sched_domain_span(sd),
				   sched_domain_span(child));
		}

6991
	}
6992
	set_domain_attribute(sd, attr);
6993 6994 6995 6996

	return sd;
}

6997 6998 6999 7000
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
7001 7002
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
7003
{
7004
	enum s_alloc alloc_state;
7005
	struct sched_domain *sd;
7006
	struct s_data d;
7007
	struct rq *rq = NULL;
7008
	int i, ret = -ENOMEM;
7009

7010 7011 7012
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
7013

7014
	/* Set up domains for cpus specified by the cpu_map. */
7015
	for_each_cpu(i, cpu_map) {
7016 7017
		struct sched_domain_topology_level *tl;

7018
		sd = NULL;
7019
		for_each_sd_topology(tl) {
7020
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
7021 7022
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
7023 7024
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
7025 7026
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
7027
		}
7028 7029 7030 7031 7032 7033
	}

	/* Build the groups for the domains */
	for_each_cpu(i, cpu_map) {
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			sd->span_weight = cpumask_weight(sched_domain_span(sd));
7034 7035 7036 7037 7038 7039 7040
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
7041
		}
7042
	}
7043

7044
	/* Calculate CPU capacity for physical packages and nodes */
7045 7046 7047
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
7048

7049 7050
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
7051
			init_sched_groups_capacity(i, sd);
7052
		}
7053
	}
7054

L
Linus Torvalds 已提交
7055
	/* Attach the domains */
7056
	rcu_read_lock();
7057
	for_each_cpu(i, cpu_map) {
7058
		rq = cpu_rq(i);
7059
		sd = *per_cpu_ptr(d.sd, i);
7060 7061 7062 7063 7064

		/* Use READ_ONCE()/WRITE_ONCE() to avoid load/store tearing: */
		if (rq->cpu_capacity_orig > READ_ONCE(d.rd->max_cpu_capacity))
			WRITE_ONCE(d.rd->max_cpu_capacity, rq->cpu_capacity_orig);

7065
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
7066
	}
7067
	rcu_read_unlock();
7068

7069
	if (rq && sched_debug_enabled) {
7070 7071 7072 7073
		pr_info("span: %*pbl (max cpu_capacity = %lu)\n",
			cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity);
	}

7074
	ret = 0;
7075
error:
7076
	__free_domain_allocs(&d, alloc_state, cpu_map);
7077
	return ret;
L
Linus Torvalds 已提交
7078
}
P
Paul Jackson 已提交
7079

7080
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
7081
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
7082 7083
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
7084 7085 7086

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
7087 7088
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
7089
 */
7090
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
7091

7092 7093 7094 7095 7096
/*
 * arch_update_cpu_topology lets virtualized architectures update the
 * cpu core maps. It is supposed to return 1 if the topology changed
 * or 0 if it stayed the same.
 */
7097
int __weak arch_update_cpu_topology(void)
7098
{
7099
	return 0;
7100 7101
}

7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126
cpumask_var_t *alloc_sched_domains(unsigned int ndoms)
{
	int i;
	cpumask_var_t *doms;

	doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL);
	if (!doms)
		return NULL;
	for (i = 0; i < ndoms; i++) {
		if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) {
			free_sched_domains(doms, i);
			return NULL;
		}
	}
	return doms;
}

void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms)
{
	unsigned int i;
	for (i = 0; i < ndoms; i++)
		free_cpumask_var(doms[i]);
	kfree(doms);
}

7127
/*
I
Ingo Molnar 已提交
7128
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
7129 7130
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
7131
 */
7132
static int init_sched_domains(const struct cpumask *cpu_map)
7133
{
7134 7135
	int err;

7136
	arch_update_cpu_topology();
P
Paul Jackson 已提交
7137
	ndoms_cur = 1;
7138
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
7139
	if (!doms_cur)
7140 7141
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
7142
	err = build_sched_domains(doms_cur[0], NULL);
7143
	register_sched_domain_sysctl();
7144 7145

	return err;
7146 7147 7148 7149 7150 7151
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
7152
static void detach_destroy_domains(const struct cpumask *cpu_map)
7153 7154 7155
{
	int i;

7156
	rcu_read_lock();
7157
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
7158
		cpu_attach_domain(NULL, &def_root_domain, i);
7159
	rcu_read_unlock();
7160 7161
}

7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177
/* handle null as "default" */
static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,
			struct sched_domain_attr *new, int idx_new)
{
	struct sched_domain_attr tmp;

	/* fast path */
	if (!new && !cur)
		return 1;

	tmp = SD_ATTR_INIT;
	return !memcmp(cur ? (cur + idx_cur) : &tmp,
			new ? (new + idx_new) : &tmp,
			sizeof(struct sched_domain_attr));
}

P
Paul Jackson 已提交
7178 7179
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
7180
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
7181 7182 7183
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
7184
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
7185 7186 7187
 * The masks don't intersect (don't overlap.) We should setup one
 * sched domain for each mask. CPUs not in any of the cpumasks will
 * not be load balanced. If the same cpumask appears both in the
P
Paul Jackson 已提交
7188 7189 7190
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
7191 7192 7193 7194 7195 7196
 * The passed in 'doms_new' should be allocated using
 * alloc_sched_domains.  This routine takes ownership of it and will
 * free_sched_domains it when done with it. If the caller failed the
 * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1,
 * and partition_sched_domains() will fallback to the single partition
 * 'fallback_doms', it also forces the domains to be rebuilt.
P
Paul Jackson 已提交
7197
 *
7198
 * If doms_new == NULL it will be replaced with cpu_online_mask.
7199 7200
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
7201
 *
P
Paul Jackson 已提交
7202 7203
 * Call with hotplug lock held
 */
7204
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
7205
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
7206
{
7207
	int i, j, n;
7208
	int new_topology;
P
Paul Jackson 已提交
7209

7210
	mutex_lock(&sched_domains_mutex);
7211

7212 7213 7214
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

7215 7216 7217
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

7218
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
7219 7220 7221

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
7222
		for (j = 0; j < n && !new_topology; j++) {
7223
			if (cpumask_equal(doms_cur[i], doms_new[j])
7224
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
7225 7226 7227
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
7228
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
7229 7230 7231 7232
match1:
		;
	}

7233
	n = ndoms_cur;
7234
	if (doms_new == NULL) {
7235
		n = 0;
7236
		doms_new = &fallback_doms;
7237
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
7238
		WARN_ON_ONCE(dattr_new);
7239 7240
	}

P
Paul Jackson 已提交
7241 7242
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
7243
		for (j = 0; j < n && !new_topology; j++) {
7244
			if (cpumask_equal(doms_new[i], doms_cur[j])
7245
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
7246 7247 7248
				goto match2;
		}
		/* no match - add a new doms_new */
7249
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
7250 7251 7252 7253 7254
match2:
		;
	}

	/* Remember the new sched domains */
7255 7256
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
7257
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
7258
	doms_cur = doms_new;
7259
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
7260
	ndoms_cur = ndoms_new;
7261 7262

	register_sched_domain_sysctl();
7263

7264
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
7265 7266
}

7267 7268
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
7269
/*
7270 7271 7272
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
7273 7274 7275
 *
 * 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
7276
 */
7277
static void cpuset_cpu_active(void)
7278
{
7279
	if (cpuhp_tasks_frozen) {
7280 7281 7282 7283 7284 7285 7286 7287 7288
		/*
		 * 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.
		 */
		num_cpus_frozen--;
		if (likely(num_cpus_frozen)) {
			partition_sched_domains(1, NULL, NULL);
7289
			return;
7290 7291 7292 7293 7294 7295
		}
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
7296
	}
7297
	cpuset_update_active_cpus(true);
7298
}
7299

7300
static int cpuset_cpu_inactive(unsigned int cpu)
7301
{
7302 7303
	unsigned long flags;
	struct dl_bw *dl_b;
7304 7305
	bool overflow;
	int cpus;
7306

7307
	if (!cpuhp_tasks_frozen) {
7308 7309
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
7310

7311 7312 7313 7314
		raw_spin_lock_irqsave(&dl_b->lock, flags);
		cpus = dl_bw_cpus(cpu);
		overflow = __dl_overflow(dl_b, cpus, 0, 0);
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
7315

7316
		rcu_read_unlock_sched();
7317

7318
		if (overflow)
7319
			return -EBUSY;
7320
		cpuset_update_active_cpus(false);
7321
	} else {
7322 7323
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
7324
	}
7325
	return 0;
7326 7327
}

7328
int sched_cpu_activate(unsigned int cpu)
7329
{
7330 7331 7332
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

7333
	set_cpu_active(cpu, true);
7334

7335
	if (sched_smp_initialized) {
7336
		sched_domains_numa_masks_set(cpu);
7337
		cpuset_cpu_active();
7338
	}
7339 7340 7341 7342 7343 7344 7345 7346 7347 7348 7349 7350 7351 7352 7353 7354 7355 7356 7357

	/*
	 * Put the rq online, if not already. This happens:
	 *
	 * 1) In the early boot process, because we build the real domains
	 *    after all cpus have been brought up.
	 *
	 * 2) At runtime, if cpuset_cpu_active() fails to rebuild the
	 *    domains.
	 */
	raw_spin_lock_irqsave(&rq->lock, flags);
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_online(rq);
	}
	raw_spin_unlock_irqrestore(&rq->lock, flags);

	update_max_interval();

7358
	return 0;
7359 7360
}

7361
int sched_cpu_deactivate(unsigned int cpu)
7362 7363 7364
{
	int ret;

7365
	set_cpu_active(cpu, false);
7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379
	/*
	 * 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.
	 *
	 * For CONFIG_PREEMPT we have preemptible RCU and its sync_rcu() might
	 * not imply sync_sched(), so wait for both.
	 *
	 * Do sync before park smpboot threads to take care the rcu boost case.
	 */
	if (IS_ENABLED(CONFIG_PREEMPT))
		synchronize_rcu_mult(call_rcu, call_rcu_sched);
	else
		synchronize_rcu();
7380 7381 7382 7383 7384 7385 7386 7387

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
7388
	}
7389 7390
	sched_domains_numa_masks_clear(cpu);
	return 0;
7391 7392
}

7393 7394 7395 7396 7397 7398 7399 7400
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();
}

7401 7402 7403
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
7404
	sched_rq_cpu_starting(cpu);
7405
	return 0;
7406 7407
}

7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423 7424 7425
#ifdef CONFIG_HOTPLUG_CPU
int sched_cpu_dying(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

	/* Handle pending wakeups and then migrate everything off */
	sched_ttwu_pending();
	raw_spin_lock_irqsave(&rq->lock, flags);
	if (rq->rd) {
		BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
		set_rq_offline(rq);
	}
	migrate_tasks(rq);
	BUG_ON(rq->nr_running != 1);
	raw_spin_unlock_irqrestore(&rq->lock, flags);
	calc_load_migrate(rq);
	update_max_interval();
7426
	nohz_balance_exit_idle(cpu);
7427
	hrtick_clear(rq);
7428 7429 7430 7431
	return 0;
}
#endif

P
Peter Zijlstra 已提交
7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447
#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
7448 7449
void __init sched_init_smp(void)
{
7450 7451 7452
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7453
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
7454

7455 7456
	sched_init_numa();

7457 7458 7459 7460 7461
	/*
	 * There's no userspace yet to cause hotplug operations; hence all the
	 * cpu masks are stable and all blatant races in the below code cannot
	 * happen.
	 */
7462
	mutex_lock(&sched_domains_mutex);
7463
	init_sched_domains(cpu_active_mask);
7464 7465 7466
	cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
	if (cpumask_empty(non_isolated_cpus))
		cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);
7467
	mutex_unlock(&sched_domains_mutex);
7468

7469
	/* Move init over to a non-isolated CPU */
7470
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
7471
		BUG();
I
Ingo Molnar 已提交
7472
	sched_init_granularity();
7473
	free_cpumask_var(non_isolated_cpus);
7474

7475
	init_sched_rt_class();
7476
	init_sched_dl_class();
P
Peter Zijlstra 已提交
7477 7478 7479

	sched_init_smt();

7480
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
7481
}
7482 7483 7484

static int __init migration_init(void)
{
7485
	sched_rq_cpu_starting(smp_processor_id());
7486
	return 0;
L
Linus Torvalds 已提交
7487
}
7488 7489
early_initcall(migration_init);

L
Linus Torvalds 已提交
7490 7491 7492
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
7493
	sched_init_granularity();
L
Linus Torvalds 已提交
7494 7495 7496 7497 7498 7499 7500 7501 7502 7503
}
#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);
}

7504
#ifdef CONFIG_CGROUP_SCHED
7505 7506 7507 7508
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
7509
struct task_group root_task_group;
7510
LIST_HEAD(task_groups);
7511 7512 7513

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
7514
#endif
P
Peter Zijlstra 已提交
7515

7516
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
7517
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
7518

L
Linus Torvalds 已提交
7519 7520
void __init sched_init(void)
{
I
Ingo Molnar 已提交
7521
	int i, j;
7522 7523 7524 7525 7526 7527 7528 7529 7530
	unsigned long alloc_size = 0, ptr;

#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) {
7531
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
7532 7533

#ifdef CONFIG_FAIR_GROUP_SCHED
7534
		root_task_group.se = (struct sched_entity **)ptr;
7535 7536
		ptr += nr_cpu_ids * sizeof(void **);

7537
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
7538
		ptr += nr_cpu_ids * sizeof(void **);
7539

7540
#endif /* CONFIG_FAIR_GROUP_SCHED */
7541
#ifdef CONFIG_RT_GROUP_SCHED
7542
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
7543 7544
		ptr += nr_cpu_ids * sizeof(void **);

7545
		root_task_group.rt_rq = (struct rt_rq **)ptr;
7546 7547
		ptr += nr_cpu_ids * sizeof(void **);

7548
#endif /* CONFIG_RT_GROUP_SCHED */
7549
	}
7550
#ifdef CONFIG_CPUMASK_OFFSTACK
7551 7552 7553
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7554 7555
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7556
	}
7557
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
7558

7559 7560 7561
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth,
7562
			global_rt_period(), global_rt_runtime());
7563

G
Gregory Haskins 已提交
7564 7565 7566 7567
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

7568
#ifdef CONFIG_RT_GROUP_SCHED
7569
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
7570
			global_rt_period(), global_rt_runtime());
7571
#endif /* CONFIG_RT_GROUP_SCHED */
7572

D
Dhaval Giani 已提交
7573
#ifdef CONFIG_CGROUP_SCHED
7574 7575
	task_group_cache = KMEM_CACHE(task_group, 0);

7576 7577
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
7578
	INIT_LIST_HEAD(&root_task_group.siblings);
7579
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
7580
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
7581

7582
	for_each_possible_cpu(i) {
7583
		struct rq *rq;
L
Linus Torvalds 已提交
7584 7585

		rq = cpu_rq(i);
7586
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
7587
		rq->nr_running = 0;
7588 7589
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
7590
		init_cfs_rq(&rq->cfs);
7591 7592
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
7593
#ifdef CONFIG_FAIR_GROUP_SCHED
7594
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
7595
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
D
Dhaval Giani 已提交
7596
		/*
7597
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
7598 7599 7600 7601
		 *
		 * In case of task-groups formed thr' the cgroup filesystem, it
		 * gets 100% of the cpu resources in the system. This overall
		 * system cpu resource is divided among the tasks of
7602
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
7603 7604 7605
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
7606
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
7607 7608 7609
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
7610
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
7611
		 *
7612 7613
		 * 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
7614
		 */
7615
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
7616
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
7617 7618 7619
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
7620
#ifdef CONFIG_RT_GROUP_SCHED
7621
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
7622
#endif
L
Linus Torvalds 已提交
7623

I
Ingo Molnar 已提交
7624 7625
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
7626

L
Linus Torvalds 已提交
7627
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
7628
		rq->sd = NULL;
G
Gregory Haskins 已提交
7629
		rq->rd = NULL;
7630
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
7631
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
7632
		rq->active_balance = 0;
I
Ingo Molnar 已提交
7633
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
7634
		rq->push_cpu = 0;
7635
		rq->cpu = i;
7636
		rq->online = 0;
7637 7638
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
7639
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
7640 7641 7642

		INIT_LIST_HEAD(&rq->cfs_tasks);

7643
		rq_attach_root(rq, &def_root_domain);
7644
#ifdef CONFIG_NO_HZ_COMMON
7645
		rq->last_load_update_tick = jiffies;
7646
		rq->nohz_flags = 0;
7647
#endif
7648 7649 7650
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
7651
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
7652
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
7653 7654 7655
		atomic_set(&rq->nr_iowait, 0);
	}

7656
	set_load_weight(&init_task);
7657

L
Linus Torvalds 已提交
7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
	atomic_inc(&init_mm.mm_count);
	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());
7671 7672 7673

	calc_load_update = jiffies + LOAD_FREQ;

7674
#ifdef CONFIG_SMP
7675
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
7676 7677 7678
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
7679
	idle_thread_set_boot_cpu();
7680
	set_cpu_rq_start_time(smp_processor_id());
7681 7682
#endif
	init_sched_fair_class();
7683

7684 7685
	init_schedstats();

7686
	scheduler_running = 1;
L
Linus Torvalds 已提交
7687 7688
}

7689
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
7690 7691
static inline int preempt_count_equals(int preempt_offset)
{
7692
	int nested = preempt_count() + rcu_preempt_depth();
7693

A
Arnd Bergmann 已提交
7694
	return (nested == preempt_offset);
7695 7696
}

7697
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7698
{
P
Peter Zijlstra 已提交
7699 7700 7701 7702 7703
	/*
	 * 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.
	 */
7704
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
7705 7706 7707 7708
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
7709
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
7710

7711 7712 7713 7714 7715
	___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
7716 7717
{
	static unsigned long prev_jiffy;	/* ratelimiting */
7718
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
7719

7720
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
7721 7722
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
7723
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
7724 7725 7726 7727 7728
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

7729 7730 7731
	/* Save this before calling printk(), since that will clobber it */
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
7732 7733 7734 7735 7736 7737 7738
	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
7739

7740 7741 7742
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
7743 7744 7745
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
7746 7747
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
7748
		pr_err("Preemption disabled at:");
7749
		print_ip_sym(preempt_disable_ip);
7750 7751
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
7752
	dump_stack();
7753
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
7754
}
7755
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
7756 7757 7758
#endif

#ifdef CONFIG_MAGIC_SYSRQ
7759
void normalize_rt_tasks(void)
7760
{
7761
	struct task_struct *g, *p;
7762 7763 7764
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
7765

7766
	read_lock(&tasklist_lock);
7767
	for_each_process_thread(g, p) {
7768 7769 7770
		/*
		 * Only normalize user tasks:
		 */
7771
		if (p->flags & PF_KTHREAD)
7772 7773
			continue;

7774 7775 7776 7777
		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
7778

7779
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
7780 7781 7782 7783
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
7784
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
7785
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
7786
			continue;
I
Ingo Molnar 已提交
7787
		}
L
Linus Torvalds 已提交
7788

7789
		__sched_setscheduler(p, &attr, false, false);
7790
	}
7791
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
7792 7793 7794
}

#endif /* CONFIG_MAGIC_SYSRQ */
7795

7796
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
7797
/*
7798
 * These functions are only useful for the IA64 MCA handling, or kdb.
7799 7800 7801 7802 7803 7804 7805 7806 7807 7808 7809 7810 7811
 *
 * 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.
 */

/**
 * curr_task - return the current task for a given cpu.
 * @cpu: the processor in question.
 *
 * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED!
7812 7813
 *
 * Return: The current task for @cpu.
7814
 */
7815
struct task_struct *curr_task(int cpu)
7816 7817 7818 7819
{
	return cpu_curr(cpu);
}

7820 7821 7822
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
7823 7824 7825 7826 7827 7828
/**
 * set_curr_task - set the current task for a given cpu.
 * @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
7829 7830
 * are serviced on a separate stack. It allows the architecture to switch the
 * notion of the current task on a cpu in a non-blocking manner. This function
7831 7832 7833 7834 7835 7836 7837
 * 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!
 */
7838
void ia64_set_curr_task(int cpu, struct task_struct *p)
7839 7840 7841 7842 7843
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
7844

D
Dhaval Giani 已提交
7845
#ifdef CONFIG_CGROUP_SCHED
7846 7847 7848
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

7849
static void sched_free_group(struct task_group *tg)
7850 7851 7852
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
7853
	autogroup_free(tg);
7854
	kmem_cache_free(task_group_cache, tg);
7855 7856 7857
}

/* allocate runqueue etc for a new task group */
7858
struct task_group *sched_create_group(struct task_group *parent)
7859 7860 7861
{
	struct task_group *tg;

7862
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
7863 7864 7865
	if (!tg)
		return ERR_PTR(-ENOMEM);

7866
	if (!alloc_fair_sched_group(tg, parent))
7867 7868
		goto err;

7869
	if (!alloc_rt_sched_group(tg, parent))
7870 7871
		goto err;

7872 7873 7874
	return tg;

err:
7875
	sched_free_group(tg);
7876 7877 7878 7879 7880 7881 7882
	return ERR_PTR(-ENOMEM);
}

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

7883
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7884
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
7885 7886 7887 7888 7889

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
7890
	list_add_rcu(&tg->siblings, &parent->children);
7891
	spin_unlock_irqrestore(&task_group_lock, flags);
7892 7893

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
7894 7895
}

7896
/* rcu callback to free various structures associated with a task group */
7897
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
7898 7899
{
	/* now it should be safe to free those cfs_rqs */
7900
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
7901 7902
}

7903
void sched_destroy_group(struct task_group *tg)
7904 7905
{
	/* wait for possible concurrent references to cfs_rqs complete */
7906
	call_rcu(&tg->rcu, sched_free_group_rcu);
7907 7908 7909
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
7910
{
7911
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
7912

7913
	/* end participation in shares distribution */
7914
	unregister_fair_sched_group(tg);
7915 7916

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7917
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
7918
	list_del_rcu(&tg->siblings);
7919
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7920 7921
}

7922
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
7923
{
P
Peter Zijlstra 已提交
7924
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
7925

7926 7927 7928 7929 7930 7931
	/*
	 * 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
7932 7933 7934 7935
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
7936
#ifdef CONFIG_FAIR_GROUP_SCHED
7937 7938
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
7939
	else
P
Peter Zijlstra 已提交
7940
#endif
7941
		set_task_rq(tsk, task_cpu(tsk));
7942 7943 7944 7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955 7956 7957 7958 7959 7960 7961 7962 7963 7964 7965 7966 7967
}

/*
 * 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)
{
	int queued, running;
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(tsk, &rf);

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

	if (queued)
		dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE);
	if (unlikely(running))
		put_prev_task(rq, tsk);

	sched_change_group(tsk, TASK_MOVE_GROUP);
P
Peter Zijlstra 已提交
7968

7969
	if (queued)
7970
		enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE);
7971
	if (unlikely(running))
7972
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
7973

7974
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
7975
}
D
Dhaval Giani 已提交
7976
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
7977

7978 7979 7980 7981 7982
#ifdef CONFIG_RT_GROUP_SCHED
/*
 * Ensure that the real time constraints are schedulable.
 */
static DEFINE_MUTEX(rt_constraints_mutex);
P
Peter Zijlstra 已提交
7983

P
Peter Zijlstra 已提交
7984 7985
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
7986
{
P
Peter Zijlstra 已提交
7987
	struct task_struct *g, *p;
7988

7989 7990 7991 7992 7993 7994
	/*
	 * Autogroups do not have RT tasks; see autogroup_create().
	 */
	if (task_group_is_autogroup(tg))
		return 0;

7995
	for_each_process_thread(g, p) {
7996
		if (rt_task(p) && task_group(p) == tg)
P
Peter Zijlstra 已提交
7997
			return 1;
7998
	}
7999

P
Peter Zijlstra 已提交
8000 8001
	return 0;
}
8002

P
Peter Zijlstra 已提交
8003 8004 8005 8006 8007
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
8008

8009
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
8010 8011 8012 8013 8014
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
8015

P
Peter Zijlstra 已提交
8016 8017
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
8018

P
Peter Zijlstra 已提交
8019 8020 8021
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
8022 8023
	}

8024 8025 8026 8027 8028
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
8029

8030 8031 8032
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
8033 8034
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
8035

P
Peter Zijlstra 已提交
8036
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
8037

8038 8039 8040 8041 8042
	/*
	 * Nobody can have more than the global setting allows.
	 */
	if (total > to_ratio(global_rt_period(), global_rt_runtime()))
		return -EINVAL;
P
Peter Zijlstra 已提交
8043

8044 8045 8046
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
8047 8048 8049
	list_for_each_entry_rcu(child, &tg->children, siblings) {
		period = ktime_to_ns(child->rt_bandwidth.rt_period);
		runtime = child->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
8050

P
Peter Zijlstra 已提交
8051 8052 8053 8054
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
8055

P
Peter Zijlstra 已提交
8056
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
8057
	}
P
Peter Zijlstra 已提交
8058

P
Peter Zijlstra 已提交
8059 8060 8061 8062
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
8063 8064
}

P
Peter Zijlstra 已提交
8065
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8066
{
8067 8068
	int ret;

P
Peter Zijlstra 已提交
8069 8070 8071 8072 8073 8074
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

8075 8076 8077 8078 8079
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8080 8081
}

8082
static int tg_set_rt_bandwidth(struct task_group *tg,
8083
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
8084
{
P
Peter Zijlstra 已提交
8085
	int i, err = 0;
P
Peter Zijlstra 已提交
8086

8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097
	/*
	 * Disallowing the root group RT runtime is BAD, it would disallow the
	 * kernel creating (and or operating) RT threads.
	 */
	if (tg == &root_task_group && rt_runtime == 0)
		return -EINVAL;

	/* No period doesn't make any sense. */
	if (rt_period == 0)
		return -EINVAL;

P
Peter Zijlstra 已提交
8098
	mutex_lock(&rt_constraints_mutex);
8099
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
8100 8101
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
8102
		goto unlock;
P
Peter Zijlstra 已提交
8103

8104
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8105 8106
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
8107 8108 8109 8110

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

8111
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8112
		rt_rq->rt_runtime = rt_runtime;
8113
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8114
	}
8115
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
8116
unlock:
8117
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
8118 8119 8120
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
8121 8122
}

8123
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8124 8125 8126 8127 8128 8129 8130 8131
{
	u64 rt_runtime, rt_period;

	rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
	if (rt_runtime_us < 0)
		rt_runtime = RUNTIME_INF;

8132
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8133 8134
}

8135
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
8136 8137 8138
{
	u64 rt_runtime_us;

8139
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
8140 8141
		return -1;

8142
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
8143 8144 8145
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
8146

8147
static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
8148 8149 8150
{
	u64 rt_runtime, rt_period;

8151
	rt_period = rt_period_us * NSEC_PER_USEC;
8152 8153
	rt_runtime = tg->rt_bandwidth.rt_runtime;

8154
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8155 8156
}

8157
static long sched_group_rt_period(struct task_group *tg)
8158 8159 8160 8161 8162 8163 8164
{
	u64 rt_period_us;

	rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
	do_div(rt_period_us, NSEC_PER_USEC);
	return rt_period_us;
}
8165
#endif /* CONFIG_RT_GROUP_SCHED */
8166

8167
#ifdef CONFIG_RT_GROUP_SCHED
8168 8169 8170 8171 8172
static int sched_rt_global_constraints(void)
{
	int ret = 0;

	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
8173
	read_lock(&tasklist_lock);
8174
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
8175
	read_unlock(&tasklist_lock);
8176 8177 8178 8179
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
8180

8181
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8182 8183 8184 8185 8186 8187 8188 8189
{
	/* Don't accept realtime tasks when there is no way for them to run */
	if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0)
		return 0;

	return 1;
}

8190
#else /* !CONFIG_RT_GROUP_SCHED */
8191 8192
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
8193
	unsigned long flags;
8194
	int i;
8195

8196
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8197 8198 8199
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

8200
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8201
		rt_rq->rt_runtime = global_rt_runtime();
8202
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8203
	}
8204
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8205

8206
	return 0;
8207
}
8208
#endif /* CONFIG_RT_GROUP_SCHED */
8209

8210
static int sched_dl_global_validate(void)
8211
{
8212 8213
	u64 runtime = global_rt_runtime();
	u64 period = global_rt_period();
8214
	u64 new_bw = to_ratio(period, runtime);
8215
	struct dl_bw *dl_b;
8216
	int cpu, ret = 0;
8217
	unsigned long flags;
8218 8219 8220 8221 8222 8223 8224 8225 8226 8227

	/*
	 * Here we want to check the bandwidth not being set to some
	 * value smaller than the currently allocated bandwidth in
	 * any of the root_domains.
	 *
	 * FIXME: Cycling on all the CPUs is overdoing, but simpler than
	 * cycling on root_domains... Discussion on different/better
	 * solutions is welcome!
	 */
8228
	for_each_possible_cpu(cpu) {
8229 8230
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8231

8232
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8233 8234
		if (new_bw < dl_b->total_bw)
			ret = -EBUSY;
8235
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8236

8237 8238
		rcu_read_unlock_sched();

8239 8240
		if (ret)
			break;
8241 8242
	}

8243
	return ret;
8244 8245
}

8246
static void sched_dl_do_global(void)
8247
{
8248
	u64 new_bw = -1;
8249
	struct dl_bw *dl_b;
8250
	int cpu;
8251
	unsigned long flags;
8252

8253 8254 8255 8256 8257 8258 8259 8260 8261 8262
	def_dl_bandwidth.dl_period = global_rt_period();
	def_dl_bandwidth.dl_runtime = global_rt_runtime();

	if (global_rt_runtime() != RUNTIME_INF)
		new_bw = to_ratio(global_rt_period(), global_rt_runtime());

	/*
	 * FIXME: As above...
	 */
	for_each_possible_cpu(cpu) {
8263 8264
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8265

8266
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8267
		dl_b->bw = new_bw;
8268
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8269 8270

		rcu_read_unlock_sched();
8271
	}
8272 8273 8274 8275 8276 8277 8278
}

static int sched_rt_global_validate(void)
{
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

8279 8280
	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
8281 8282 8283 8284 8285 8286 8287 8288 8289
		return -EINVAL;

	return 0;
}

static void sched_rt_do_global(void)
{
	def_rt_bandwidth.rt_runtime = global_rt_runtime();
	def_rt_bandwidth.rt_period = ns_to_ktime(global_rt_period());
8290 8291
}

8292
int sched_rt_handler(struct ctl_table *table, int write,
8293
		void __user *buffer, size_t *lenp,
8294 8295 8296 8297
		loff_t *ppos)
{
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);
8298
	int ret;
8299 8300 8301 8302 8303

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

8304
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8305 8306

	if (!ret && write) {
8307 8308 8309 8310
		ret = sched_rt_global_validate();
		if (ret)
			goto undo;

8311
		ret = sched_dl_global_validate();
8312 8313 8314
		if (ret)
			goto undo;

8315
		ret = sched_rt_global_constraints();
8316 8317 8318 8319 8320 8321 8322 8323 8324 8325
		if (ret)
			goto undo;

		sched_rt_do_global();
		sched_dl_do_global();
	}
	if (0) {
undo:
		sysctl_sched_rt_period = old_period;
		sysctl_sched_rt_runtime = old_runtime;
8326 8327 8328 8329 8330
	}
	mutex_unlock(&mutex);

	return ret;
}
8331

8332
int sched_rr_handler(struct ctl_table *table, int write,
8333 8334 8335 8336 8337 8338 8339 8340
		void __user *buffer, size_t *lenp,
		loff_t *ppos)
{
	int ret;
	static DEFINE_MUTEX(mutex);

	mutex_lock(&mutex);
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8341 8342
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
8343
	if (!ret && write) {
8344 8345
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
8346 8347 8348 8349 8350
	}
	mutex_unlock(&mutex);
	return ret;
}

8351
#ifdef CONFIG_CGROUP_SCHED
8352

8353
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
8354
{
8355
	return css ? container_of(css, struct task_group, css) : NULL;
8356 8357
}

8358 8359
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
8360
{
8361 8362
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
8363

8364
	if (!parent) {
8365
		/* This is early initialization for the top cgroup */
8366
		return &root_task_group.css;
8367 8368
	}

8369
	tg = sched_create_group(parent);
8370 8371 8372
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

8373 8374
	sched_online_group(tg, parent);

8375 8376 8377
	return &tg->css;
}

8378
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
8379
{
8380
	struct task_group *tg = css_tg(css);
8381

8382
	sched_offline_group(tg);
8383 8384
}

8385
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
8386
{
8387
	struct task_group *tg = css_tg(css);
8388

8389 8390 8391 8392
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
8393 8394
}

8395 8396 8397 8398
/*
 * 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.
 */
8399
static void cpu_cgroup_fork(struct task_struct *task)
8400
{
8401 8402 8403 8404 8405 8406 8407 8408
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
8409 8410
}

8411
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
8412
{
8413
	struct task_struct *task;
8414
	struct cgroup_subsys_state *css;
8415
	int ret = 0;
8416

8417
	cgroup_taskset_for_each(task, css, tset) {
8418
#ifdef CONFIG_RT_GROUP_SCHED
8419
		if (!sched_rt_can_attach(css_tg(css), task))
8420
			return -EINVAL;
8421
#else
8422 8423 8424
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
8425
#endif
8426 8427 8428 8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441
		/*
		 * 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;
8442
	}
8443
	return ret;
8444
}
8445

8446
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
8447
{
8448
	struct task_struct *task;
8449
	struct cgroup_subsys_state *css;
8450

8451
	cgroup_taskset_for_each(task, css, tset)
8452
		sched_move_task(task);
8453 8454
}

8455
#ifdef CONFIG_FAIR_GROUP_SCHED
8456 8457
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
8458
{
8459
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
8460 8461
}

8462 8463
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
8464
{
8465
	struct task_group *tg = css_tg(css);
8466

8467
	return (u64) scale_load_down(tg->shares);
8468
}
8469 8470

#ifdef CONFIG_CFS_BANDWIDTH
8471 8472
static DEFINE_MUTEX(cfs_constraints_mutex);

8473 8474 8475
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

8476 8477
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

8478 8479
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
8480
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
8481
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8482 8483 8484 8485 8486 8487 8488 8489 8490 8491 8492 8493 8494 8495 8496 8497 8498 8499 8500 8501

	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;

8502 8503 8504 8505 8506
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
8507 8508 8509 8510 8511
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

8512
	runtime_enabled = quota != RUNTIME_INF;
8513
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
8514 8515 8516 8517 8518 8519
	/*
	 * 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();
8520 8521 8522
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
8523

P
Paul Turner 已提交
8524
	__refill_cfs_bandwidth_runtime(cfs_b);
8525
	/* restart the period timer (if active) to handle new period expiry */
P
Peter Zijlstra 已提交
8526 8527
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
8528 8529
	raw_spin_unlock_irq(&cfs_b->lock);

8530
	for_each_online_cpu(i) {
8531
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
8532
		struct rq *rq = cfs_rq->rq;
8533 8534

		raw_spin_lock_irq(&rq->lock);
8535
		cfs_rq->runtime_enabled = runtime_enabled;
8536
		cfs_rq->runtime_remaining = 0;
8537

8538
		if (cfs_rq->throttled)
8539
			unthrottle_cfs_rq(cfs_rq);
8540 8541
		raw_spin_unlock_irq(&rq->lock);
	}
8542 8543
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
8544 8545
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
8546
	put_online_cpus();
8547

8548
	return ret;
8549 8550 8551 8552 8553 8554
}

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

8555
	period = ktime_to_ns(tg->cfs_bandwidth.period);
8556 8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567
	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;

8568
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
8569 8570
		return -1;

8571
	quota_us = tg->cfs_bandwidth.quota;
8572 8573 8574 8575 8576 8577 8578 8579 8580 8581
	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;
8582
	quota = tg->cfs_bandwidth.quota;
8583 8584 8585 8586 8587 8588 8589 8590

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

8591
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
8592 8593 8594 8595 8596
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

8597 8598
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
8599
{
8600
	return tg_get_cfs_quota(css_tg(css));
8601 8602
}

8603 8604
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
8605
{
8606
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
8607 8608
}

8609 8610
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8611
{
8612
	return tg_get_cfs_period(css_tg(css));
8613 8614
}

8615 8616
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
8617
{
8618
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
8619 8620
}

8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652
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;
8653
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8654 8655 8656 8657 8658
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
8659
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
8660 8661

		quota = normalize_cfs_quota(tg, d);
8662
		parent_quota = parent_b->hierarchical_quota;
8663 8664 8665 8666 8667 8668 8669 8670 8671 8672

		/*
		 * ensure max(child_quota) <= parent_quota, inherit when no
		 * limit is set
		 */
		if (quota == RUNTIME_INF)
			quota = parent_quota;
		else if (parent_quota != RUNTIME_INF && quota > parent_quota)
			return -EINVAL;
	}
8673
	cfs_b->hierarchical_quota = quota;
8674 8675 8676 8677 8678 8679

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
8680
	int ret;
8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691
	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);
	}

8692 8693 8694 8695 8696
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8697
}
8698

8699
static int cpu_stats_show(struct seq_file *sf, void *v)
8700
{
8701
	struct task_group *tg = css_tg(seq_css(sf));
8702
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8703

8704 8705 8706
	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);
8707 8708 8709

	return 0;
}
8710
#endif /* CONFIG_CFS_BANDWIDTH */
8711
#endif /* CONFIG_FAIR_GROUP_SCHED */
8712

8713
#ifdef CONFIG_RT_GROUP_SCHED
8714 8715
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
8716
{
8717
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
8718 8719
}

8720 8721
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
8722
{
8723
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
8724
}
8725

8726 8727
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
8728
{
8729
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
8730 8731
}

8732 8733
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8734
{
8735
	return sched_group_rt_period(css_tg(css));
8736
}
8737
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
8738

8739
static struct cftype cpu_files[] = {
8740
#ifdef CONFIG_FAIR_GROUP_SCHED
8741 8742
	{
		.name = "shares",
8743 8744
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
8745
	},
8746
#endif
8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757
#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,
	},
8758 8759
	{
		.name = "stat",
8760
		.seq_show = cpu_stats_show,
8761
	},
8762
#endif
8763
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
8764
	{
P
Peter Zijlstra 已提交
8765
		.name = "rt_runtime_us",
8766 8767
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
8768
	},
8769 8770
	{
		.name = "rt_period_us",
8771 8772
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
8773
	},
8774
#endif
8775
	{ }	/* terminate */
8776 8777
};

8778
struct cgroup_subsys cpu_cgrp_subsys = {
8779
	.css_alloc	= cpu_cgroup_css_alloc,
8780
	.css_released	= cpu_cgroup_css_released,
8781
	.css_free	= cpu_cgroup_css_free,
8782
	.fork		= cpu_cgroup_fork,
8783 8784
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
8785
	.legacy_cftypes	= cpu_files,
8786
	.early_init	= true,
8787 8788
};

8789
#endif	/* CONFIG_CGROUP_SCHED */
8790

8791 8792 8793 8794 8795
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
8796 8797 8798 8799 8800 8801 8802 8803 8804 8805 8806 8807 8808 8809 8810 8811 8812 8813 8814 8815 8816 8817 8818 8819 8820 8821 8822 8823 8824 8825 8826 8827 8828 8829 8830 8831 8832 8833 8834 8835 8836

/*
 * 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,
};