core.c 217.1 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>
78
#include <linux/mutex.h>
L
Linus Torvalds 已提交
79

80
#include <asm/switch_to.h>
81
#include <asm/tlb.h>
82
#include <asm/irq_regs.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
	lockdep_assert_held(&rq->lock);

105
	if (rq->clock_update_flags & RQCF_ACT_SKIP)
106
		return;
107

108 109 110
#ifdef CONFIG_SCHED_DEBUG
	rq->clock_update_flags |= RQCF_UPDATED;
#endif
111
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
112 113
	if (delta < 0)
		return;
114 115
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
116 117
}

I
Ingo Molnar 已提交
118 119 120
/*
 * Debugging: various feature bits
 */
P
Peter Zijlstra 已提交
121 122 123 124

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

I
Ingo Molnar 已提交
125
const_debug unsigned int sysctl_sched_features =
126
#include "features.h"
P
Peter Zijlstra 已提交
127 128 129 130
	0;

#undef SCHED_FEAT

131 132 133 134 135 136
/*
 * 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;

137 138 139 140 141 142 143 144
/*
 * 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
145
/*
P
Peter Zijlstra 已提交
146
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
147 148
 * default: 1s
 */
P
Peter Zijlstra 已提交
149
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
150

151
__read_mostly int scheduler_running;
152

P
Peter Zijlstra 已提交
153 154 155 156 157
/*
 * 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
158

159 160 161
/* cpus with isolated domains */
cpumask_var_t cpu_isolated_map;

L
Linus Torvalds 已提交
162
/*
163
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
164
 */
A
Alexey Dobriyan 已提交
165
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
166 167
	__acquires(rq->lock)
{
168
	struct rq *rq;
L
Linus Torvalds 已提交
169 170 171

	local_irq_disable();
	rq = this_rq();
172
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
173 174 175 176

	return rq;
}

177 178 179
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
180
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
181 182 183 184 185 186 187 188 189 190
	__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))) {
191
			rq_pin_lock(rq, rf);
192 193 194 195 196 197 198 199 200 201 202 203
			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.
 */
204
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
205 206 207 208 209 210
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
211
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230
		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))) {
231
			rq_pin_lock(rq, rf);
232 233 234
			return rq;
		}
		raw_spin_unlock(&rq->lock);
235
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
236 237 238 239 240 241

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

P
Peter Zijlstra 已提交
242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262
#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());

263
	raw_spin_lock(&rq->lock);
264
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
265
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
266
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
267 268 269 270

	return HRTIMER_NORESTART;
}

271
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
272

273
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
274 275 276
{
	struct hrtimer *timer = &rq->hrtick_timer;

277
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
278 279
}

280 281 282 283
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
284
{
285
	struct rq *rq = arg;
286

287
	raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
288
	__hrtick_restart(rq);
289
	rq->hrtick_csd_pending = 0;
290
	raw_spin_unlock(&rq->lock);
291 292
}

293 294 295 296 297
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
298
void hrtick_start(struct rq *rq, u64 delay)
299
{
300
	struct hrtimer *timer = &rq->hrtick_timer;
301 302 303 304 305 306 307 308 309
	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);
310

311
	hrtimer_set_expires(timer, time);
312 313

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
314
		__hrtick_restart(rq);
315
	} else if (!rq->hrtick_csd_pending) {
316
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
317 318
		rq->hrtick_csd_pending = 1;
	}
319 320
}

321 322 323 324 325 326
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
327
void hrtick_start(struct rq *rq, u64 delay)
328
{
W
Wanpeng Li 已提交
329 330 331 332 333
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
334 335
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
336 337
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
338

339
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
340
{
341 342
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
343

344 345 346 347
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
348

349 350
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
351
}
A
Andrew Morton 已提交
352
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
353 354 355 356 357 358 359
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}
A
Andrew Morton 已提交
360
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
361

362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379
/*
 * 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;								\
})

380
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
381 382 383 384 385 386 387 388 389 390
/*
 * 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);
}
391 392 393 394 395 396 397 398 399 400

/*
 * 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);
401
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
402 403 404 405 406 407 408 409 410 411 412 413 414 415

	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;
}

416 417 418 419 420 421
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
422 423 424 425 426 427 428

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
429 430
#endif

431 432 433 434 435 436 437 438 439 440
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
441
	 * barrier implied by the wakeup in wake_up_q().
442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476
	 */
	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
477
/*
478
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
479 480 481 482 483
 *
 * 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.
 */
484
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
485
{
486
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
487 488
	int cpu;

489
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
490

491
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
492 493
		return;

494
	cpu = cpu_of(rq);
495

496
	if (cpu == smp_processor_id()) {
497
		set_tsk_need_resched(curr);
498
		set_preempt_need_resched();
I
Ingo Molnar 已提交
499
		return;
500
	}
I
Ingo Molnar 已提交
501

502
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
503
		smp_send_reschedule(cpu);
504 505
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
506 507
}

508
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
509 510 511 512
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

513
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
514
		return;
515
	resched_curr(rq);
516
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
517
}
518

519
#ifdef CONFIG_SMP
520
#ifdef CONFIG_NO_HZ_COMMON
521 522 523 524 525 526 527 528
/*
 * 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).
 */
529
int get_nohz_timer_target(void)
530
{
531
	int i, cpu = smp_processor_id();
532 533
	struct sched_domain *sd;

534
	if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
535 536
		return cpu;

537
	rcu_read_lock();
538
	for_each_domain(cpu, sd) {
539
		for_each_cpu(i, sched_domain_span(sd)) {
540 541 542 543
			if (cpu == i)
				continue;

			if (!idle_cpu(i) && is_housekeeping_cpu(i)) {
544 545 546 547
				cpu = i;
				goto unlock;
			}
		}
548
	}
549 550 551

	if (!is_housekeeping_cpu(cpu))
		cpu = housekeeping_any_cpu();
552 553
unlock:
	rcu_read_unlock();
554 555
	return cpu;
}
556 557 558 559 560 561 562 563 564 565
/*
 * 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.
 */
566
static void wake_up_idle_cpu(int cpu)
567 568 569 570 571 572
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

573
	if (set_nr_and_not_polling(rq->idle))
574
		smp_send_reschedule(cpu);
575 576
	else
		trace_sched_wake_idle_without_ipi(cpu);
577 578
}

579
static bool wake_up_full_nohz_cpu(int cpu)
580
{
581 582 583 584 585 586
	/*
	 * 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.
	 */
587 588
	if (cpu_is_offline(cpu))
		return true;  /* Don't try to wake offline CPUs. */
589
	if (tick_nohz_full_cpu(cpu)) {
590 591
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
592
			tick_nohz_full_kick_cpu(cpu);
593 594 595 596 597 598
		return true;
	}

	return false;
}

599 600 601 602 603
/*
 * 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.
 */
604 605
void wake_up_nohz_cpu(int cpu)
{
606
	if (!wake_up_full_nohz_cpu(cpu))
607 608 609
		wake_up_idle_cpu(cpu);
}

610
static inline bool got_nohz_idle_kick(void)
611
{
612
	int cpu = smp_processor_id();
613 614 615 616 617 618 619 620 621 622 623 624 625

	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;
626 627
}

628
#else /* CONFIG_NO_HZ_COMMON */
629

630
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
631
{
632
	return false;
P
Peter Zijlstra 已提交
633 634
}

635
#endif /* CONFIG_NO_HZ_COMMON */
636

637
#ifdef CONFIG_NO_HZ_FULL
638
bool sched_can_stop_tick(struct rq *rq)
639
{
640 641 642 643 644 645
	int fifo_nr_running;

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

646
	/*
647 648
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
649
	 */
650 651 652 653 654
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
655 656
	}

657 658 659 660 661 662 663 664 665 666 667 668 669 670
	/*
	 * 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)
671
		return false;
672

673
	return true;
674 675
}
#endif /* CONFIG_NO_HZ_FULL */
676

677
void sched_avg_update(struct rq *rq)
678
{
679 680
	s64 period = sched_avg_period();

681
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
682 683 684 685 686 687
		/*
		 * 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));
688 689 690
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
691 692
}

693
#endif /* CONFIG_SMP */
694

695 696
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
697
/*
698 699 700 701
 * 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.
702
 */
703
int walk_tg_tree_from(struct task_group *from,
704
			     tg_visitor down, tg_visitor up, void *data)
705 706
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
707
	int ret;
708

709 710
	parent = from;

711
down:
P
Peter Zijlstra 已提交
712 713
	ret = (*down)(parent, data);
	if (ret)
714
		goto out;
715 716 717 718 719 720 721
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
722
	ret = (*up)(parent, data);
723 724
	if (ret || parent == from)
		goto out;
725 726 727 728 729

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
730
out:
P
Peter Zijlstra 已提交
731
	return ret;
732 733
}

734
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
735
{
736
	return 0;
P
Peter Zijlstra 已提交
737
}
738 739
#endif

740 741
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
742 743 744
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
745 746 747
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
748
	if (idle_policy(p->policy)) {
749
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
750
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
751 752
		return;
	}
753

754 755
	load->weight = scale_load(sched_prio_to_weight[prio]);
	load->inv_weight = sched_prio_to_wmult[prio];
756 757
}

758
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
759
{
760
	update_rq_clock(rq);
761 762
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
763
	p->sched_class->enqueue_task(rq, p, flags);
764 765
}

766
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
767
{
768
	update_rq_clock(rq);
769 770
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
771
	p->sched_class->dequeue_task(rq, p, flags);
772 773
}

774
void activate_task(struct rq *rq, struct task_struct *p, int flags)
775 776 777 778
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

779
	enqueue_task(rq, p, flags);
780 781
}

782
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
783 784 785 786
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

787
	dequeue_task(rq, p, flags);
788 789
}

790
static void update_rq_clock_task(struct rq *rq, s64 delta)
791
{
792 793 794 795 796 797 798 799
/*
 * 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
800
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821

	/*
	 * 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;
822 823
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
824
	if (static_key_false((&paravirt_steal_rq_enabled))) {
825 826 827 828 829 830 831 832 833 834 835
		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

836 837
	rq->clock_task += delta;

838
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
839
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
840 841
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
842 843
}

844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873
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;
	}
}

874
/*
I
Ingo Molnar 已提交
875
 * __normal_prio - return the priority that is based on the static prio
876 877 878
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
879
	return p->static_prio;
880 881
}

882 883 884 885 886 887 888
/*
 * 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.
 */
889
static inline int normal_prio(struct task_struct *p)
890 891 892
{
	int prio;

893 894 895
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
896 897 898 899 900 901 902 903 904 905 906 907 908
		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.
 */
909
static int effective_prio(struct task_struct *p)
910 911 912 913 914 915 916 917 918 919 920 921
{
	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
922 923 924
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
925 926
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
927
 */
928
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
929 930 931 932
{
	return cpu_curr(task_cpu(p)) == p;
}

933
/*
934 935 936 937 938
 * 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().
939
 */
940 941
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
942
				       int oldprio)
943 944 945
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
946
			prev_class->switched_from(rq, p);
947

P
Peter Zijlstra 已提交
948
		p->sched_class->switched_to(rq, p);
949
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
950
		p->sched_class->prio_changed(rq, p, oldprio);
951 952
}

953
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
954 955 956 957 958 959 960 961 962 963
{
	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) {
964
				resched_curr(rq);
965 966 967 968 969 970 971 972 973
				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.
	 */
974
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
975
		rq_clock_skip_update(rq, true);
976 977
}

L
Linus Torvalds 已提交
978
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
/*
 * 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.
 */
998
static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
999 1000 1001 1002
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
1003
	dequeue_task(rq, p, 0);
P
Peter Zijlstra 已提交
1004 1005 1006 1007 1008 1009 1010 1011
	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);
1012
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031
	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.
 */
1032
static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
1033 1034
{
	if (unlikely(!cpu_active(dest_cpu)))
1035
		return rq;
P
Peter Zijlstra 已提交
1036 1037 1038

	/* Affinity changed (again). */
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
1039
		return rq;
P
Peter Zijlstra 已提交
1040

1041 1042 1043
	rq = move_queued_task(rq, p, dest_cpu);

	return rq;
P
Peter Zijlstra 已提交
1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
}

/*
 * 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;
1054 1055
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067

	/*
	 * 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();
1068 1069 1070 1071 1072 1073 1074 1075

	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.
	 */
1076 1077 1078 1079 1080 1081
	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;
	}
1082 1083 1084
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1085 1086 1087 1088
	local_irq_enable();
	return 0;
}

1089 1090 1091 1092 1093
/*
 * 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
1094 1095 1096 1097 1098
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1099 1100
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1101 1102 1103
	struct rq *rq = task_rq(p);
	bool queued, running;

1104
	lockdep_assert_held(&p->pi_lock);
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114

	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);
1115
		dequeue_task(rq, p, DEQUEUE_SAVE);
1116 1117 1118 1119
	}
	if (running)
		put_prev_task(rq, p);

1120
	p->sched_class->set_cpus_allowed(p, new_mask);
1121 1122

	if (queued)
1123
		enqueue_task(rq, p, ENQUEUE_RESTORE);
1124
	if (running)
1125
		set_curr_task(rq, p);
1126 1127
}

P
Peter Zijlstra 已提交
1128 1129 1130 1131 1132 1133 1134 1135 1136
/*
 * 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.
 */
1137 1138
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1139
{
1140
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1141
	unsigned int dest_cpu;
1142 1143
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1144 1145
	int ret = 0;

1146
	rq = task_rq_lock(p, &rf);
P
Peter Zijlstra 已提交
1147

1148 1149 1150 1151 1152 1153 1154
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1155 1156 1157 1158 1159 1160 1161 1162 1163
	/*
	 * 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
1164 1165 1166
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1167
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1168 1169 1170 1171 1172 1173
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1174 1175 1176 1177 1178 1179 1180 1181 1182 1183
	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
1184 1185 1186 1187
	/* 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;

1188
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1189 1190 1191
	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. */
1192
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1193 1194 1195
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1196 1197 1198 1199 1200
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1201
		rq_unpin_lock(rq, &rf);
1202
		rq = move_queued_task(rq, p, dest_cpu);
1203
		rq_repin_lock(rq, &rf);
1204
	}
P
Peter Zijlstra 已提交
1205
out:
1206
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1207 1208 1209

	return ret;
}
1210 1211 1212 1213 1214

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
1215 1216
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1217
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1218
{
1219 1220 1221 1222 1223
#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
1224
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1225
			!p->on_rq);
1226

1227 1228 1229 1230 1231 1232 1233 1234 1235
	/*
	 * 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)));

1236
#ifdef CONFIG_LOCKDEP
1237 1238 1239 1240 1241
	/*
	 * 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
1242
	 * see task_group().
1243 1244 1245 1246
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1247 1248 1249
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1250 1251
#endif

1252
	trace_sched_migrate_task(p, new_cpu);
1253

1254
	if (task_cpu(p) != new_cpu) {
1255
		if (p->sched_class->migrate_task_rq)
1256
			p->sched_class->migrate_task_rq(p);
1257
		p->se.nr_migrations++;
1258
		perf_event_task_migrate(p);
1259
	}
I
Ingo Molnar 已提交
1260 1261

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1262 1263
}

1264 1265
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1266
	if (task_on_rq_queued(p)) {
1267 1268 1269 1270 1271
		struct rq *src_rq, *dst_rq;

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

1272
		p->on_rq = TASK_ON_RQ_MIGRATING;
1273 1274 1275
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1276
		p->on_rq = TASK_ON_RQ_QUEUED;
1277 1278 1279 1280 1281
		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
1282
		 * previous cpu our target instead of where it really is.
1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298
		 */
		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;

1299 1300 1301
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1302 1303 1304
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1305 1306
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1307
	double_rq_lock(src_rq, dst_rq);
1308

1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327
	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);
1328 1329
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351

	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;

1352 1353 1354 1355
	/*
	 * 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.
	 */
1356 1357 1358 1359 1360 1361 1362 1363 1364
	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;

1365
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1366 1367 1368 1369 1370 1371
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1372 1373 1374
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1375 1376 1377 1378 1379 1380 1381
 * 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
1382 1383 1384 1385 1386 1387
 * 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
1388
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1389
{
1390
	int running, queued;
1391
	struct rq_flags rf;
R
Roland McGrath 已提交
1392
	unsigned long ncsw;
1393
	struct rq *rq;
L
Linus Torvalds 已提交
1394

1395 1396 1397 1398 1399 1400 1401 1402
	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);
1403

1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414
		/*
		 * 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
1415 1416 1417
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1418
			cpu_relax();
R
Roland McGrath 已提交
1419
		}
1420

1421 1422 1423 1424 1425
		/*
		 * 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.
		 */
1426
		rq = task_rq_lock(p, &rf);
1427
		trace_sched_wait_task(p);
1428
		running = task_running(rq, p);
1429
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1430
		ncsw = 0;
1431
		if (!match_state || p->state == match_state)
1432
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1433
		task_rq_unlock(rq, p, &rf);
1434

R
Roland McGrath 已提交
1435 1436 1437 1438 1439 1440
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
		/*
		 * 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;
		}
1451

1452 1453 1454 1455 1456
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1457
		 * So if it was still runnable (but just not actively
1458 1459 1460
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1461
		if (unlikely(queued)) {
T
Thomas Gleixner 已提交
1462
			ktime_t to = NSEC_PER_SEC / HZ;
1463 1464 1465

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1466 1467
			continue;
		}
1468

1469 1470 1471 1472 1473 1474 1475
		/*
		 * 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
1476 1477

	return ncsw;
L
Linus Torvalds 已提交
1478 1479 1480 1481 1482 1483 1484 1485 1486
}

/***
 * 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
1487
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1488 1489 1490 1491 1492
 * 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.
 */
1493
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1494 1495 1496 1497 1498 1499 1500 1501 1502
{
	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
1503
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1504

1505
/*
1506
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525
 *
 * 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.
1526
 */
1527 1528
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1529 1530
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1531 1532
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1533

1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548
	/*
	 * 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;
		}
1549
	}
1550

1551 1552
	for (;;) {
		/* Any allowed, online CPU? */
1553
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1554 1555 1556
			if (!(p->flags & PF_KTHREAD) && !cpu_active(dest_cpu))
				continue;
			if (!cpu_online(dest_cpu))
1557 1558 1559
				continue;
			goto out;
		}
1560

1561
		/* No more Mr. Nice Guy. */
1562 1563
		switch (state) {
		case cpuset:
1564 1565 1566 1567 1568 1569
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
			/* fall-through */
1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588
		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()) {
1589
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1590 1591
					task_pid_nr(p), p->comm, cpu);
		}
1592 1593 1594 1595 1596
	}

	return dest_cpu;
}

1597
/*
1598
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1599
 */
1600
static inline
1601
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1602
{
1603 1604
	lockdep_assert_held(&p->pi_lock);

1605
	if (tsk_nr_cpus_allowed(p) > 1)
1606
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1607 1608
	else
		cpu = cpumask_any(tsk_cpus_allowed(p));
1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619

	/*
	 * 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 ]
	 */
1620
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1621
		     !cpu_online(cpu)))
1622
		cpu = select_fallback_rq(task_cpu(p), p);
1623 1624

	return cpu;
1625
}
1626 1627 1628 1629 1630 1631

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1632 1633 1634 1635 1636 1637 1638 1639 1640

#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
1641
#endif /* CONFIG_SMP */
1642

P
Peter Zijlstra 已提交
1643
static void
1644
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1645
{
1646
	struct rq *rq;
1647

1648 1649 1650 1651
	if (!schedstat_enabled())
		return;

	rq = this_rq();
P
Peter Zijlstra 已提交
1652

1653 1654
#ifdef CONFIG_SMP
	if (cpu == rq->cpu) {
1655 1656
		schedstat_inc(rq->ttwu_local);
		schedstat_inc(p->se.statistics.nr_wakeups_local);
P
Peter Zijlstra 已提交
1657 1658 1659
	} else {
		struct sched_domain *sd;

1660
		schedstat_inc(p->se.statistics.nr_wakeups_remote);
1661
		rcu_read_lock();
1662
		for_each_domain(rq->cpu, sd) {
P
Peter Zijlstra 已提交
1663
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
1664
				schedstat_inc(sd->ttwu_wake_remote);
P
Peter Zijlstra 已提交
1665 1666 1667
				break;
			}
		}
1668
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1669
	}
1670 1671

	if (wake_flags & WF_MIGRATED)
1672
		schedstat_inc(p->se.statistics.nr_wakeups_migrate);
P
Peter Zijlstra 已提交
1673 1674
#endif /* CONFIG_SMP */

1675 1676
	schedstat_inc(rq->ttwu_count);
	schedstat_inc(p->se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1677 1678

	if (wake_flags & WF_SYNC)
1679
		schedstat_inc(p->se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1680 1681
}

1682
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1683
{
T
Tejun Heo 已提交
1684
	activate_task(rq, p, en_flags);
1685
	p->on_rq = TASK_ON_RQ_QUEUED;
1686 1687 1688 1689

	/* 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
1690 1691
}

1692 1693 1694
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1695
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
1696
			   struct rq_flags *rf)
T
Tejun Heo 已提交
1697 1698 1699
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1700 1701
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1702
#ifdef CONFIG_SMP
1703 1704
	if (p->sched_class->task_woken) {
		/*
1705 1706
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1707
		 */
1708
		rq_unpin_lock(rq, rf);
T
Tejun Heo 已提交
1709
		p->sched_class->task_woken(rq, p);
1710
		rq_repin_lock(rq, rf);
1711
	}
T
Tejun Heo 已提交
1712

1713
	if (rq->idle_stamp) {
1714
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1715
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1716

1717 1718 1719
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1720
			rq->avg_idle = max;
1721

T
Tejun Heo 已提交
1722 1723 1724 1725 1726
		rq->idle_stamp = 0;
	}
#endif
}

1727
static void
1728
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
1729
		 struct rq_flags *rf)
1730
{
1731 1732
	int en_flags = ENQUEUE_WAKEUP;

1733 1734
	lockdep_assert_held(&rq->lock);

1735 1736 1737
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1738 1739

	if (wake_flags & WF_MIGRATED)
1740
		en_flags |= ENQUEUE_MIGRATED;
1741 1742
#endif

1743
	ttwu_activate(rq, p, en_flags);
1744
	ttwu_do_wakeup(rq, p, wake_flags, rf);
1745 1746 1747 1748 1749 1750 1751 1752 1753 1754
}

/*
 * 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)
{
1755
	struct rq_flags rf;
1756 1757 1758
	struct rq *rq;
	int ret = 0;

1759
	rq = __task_rq_lock(p, &rf);
1760
	if (task_on_rq_queued(p)) {
1761 1762
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1763
		ttwu_do_wakeup(rq, p, wake_flags, &rf);
1764 1765
		ret = 1;
	}
1766
	__task_rq_unlock(rq, &rf);
1767 1768 1769 1770

	return ret;
}

1771
#ifdef CONFIG_SMP
1772
void sched_ttwu_pending(void)
1773 1774
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1775 1776
	struct llist_node *llist = llist_del_all(&rq->wake_list);
	struct task_struct *p;
1777
	unsigned long flags;
1778
	struct rq_flags rf;
1779

1780 1781 1782 1783
	if (!llist)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
1784
	rq_pin_lock(rq, &rf);
1785

P
Peter Zijlstra 已提交
1786
	while (llist) {
P
Peter Zijlstra 已提交
1787 1788
		int wake_flags = 0;

P
Peter Zijlstra 已提交
1789 1790
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
P
Peter Zijlstra 已提交
1791 1792 1793 1794

		if (p->sched_remote_wakeup)
			wake_flags = WF_MIGRATED;

1795
		ttwu_do_activate(rq, p, wake_flags, &rf);
1796 1797
	}

1798
	rq_unpin_lock(rq, &rf);
1799
	raw_spin_unlock_irqrestore(&rq->lock, flags);
1800 1801 1802 1803
}

void scheduler_ipi(void)
{
1804 1805 1806 1807 1808
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1809
	preempt_fold_need_resched();
1810

1811
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827
		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
1828
	sched_ttwu_pending();
1829 1830 1831 1832

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1833
	if (unlikely(got_nohz_idle_kick())) {
1834
		this_rq()->idle_balance = 1;
1835
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1836
	}
1837
	irq_exit();
1838 1839
}

P
Peter Zijlstra 已提交
1840
static void ttwu_queue_remote(struct task_struct *p, int cpu, int wake_flags)
1841
{
1842 1843
	struct rq *rq = cpu_rq(cpu);

P
Peter Zijlstra 已提交
1844 1845
	p->sched_remote_wakeup = !!(wake_flags & WF_MIGRATED);

1846 1847 1848 1849 1850 1851
	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);
	}
1852
}
1853

1854 1855 1856 1857 1858
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

1859 1860 1861 1862
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872

	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);
	}
1873 1874 1875

out:
	rcu_read_unlock();
1876 1877
}

1878
bool cpus_share_cache(int this_cpu, int that_cpu)
1879 1880 1881
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1882
#endif /* CONFIG_SMP */
1883

1884
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1885 1886
{
	struct rq *rq = cpu_rq(cpu);
1887
	struct rq_flags rf;
1888

1889
#if defined(CONFIG_SMP)
1890
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1891
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
P
Peter Zijlstra 已提交
1892
		ttwu_queue_remote(p, cpu, wake_flags);
1893 1894 1895 1896
		return;
	}
#endif

1897
	raw_spin_lock(&rq->lock);
1898 1899 1900
	rq_pin_lock(rq, &rf);
	ttwu_do_activate(rq, p, wake_flags, &rf);
	rq_unpin_lock(rq, &rf);
1901
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
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 1951 1952 1953
/*
 * 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)
1954
 *   2) smp_cond_load_acquire(!X->on_cpu)
1955 1956 1957 1958 1959 1960 1961 1962 1963 1964
 *
 * 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);
 *
1965
 *                    smp_cond_load_acquire(&X->on_cpu, !VAL);
1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
 *                    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,
1991
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_load_acquire).
1992 1993 1994
 *
 */

T
Tejun Heo 已提交
1995
/**
L
Linus Torvalds 已提交
1996
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1997
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1998
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1999
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
2000
 *
2001
 * If (@state & @p->state) @p->state = TASK_RUNNING.
L
Linus Torvalds 已提交
2002
 *
2003 2004 2005 2006 2007 2008 2009
 * 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
2010
 */
2011 2012
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
2013 2014
{
	unsigned long flags;
2015
	int cpu, success = 0;
P
Peter Zijlstra 已提交
2016

2017 2018 2019 2020 2021 2022 2023
	/*
	 * 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();
2024
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
2025
	if (!(p->state & state))
L
Linus Torvalds 已提交
2026 2027
		goto out;

2028 2029
	trace_sched_waking(p);

2030
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
2031 2032
	cpu = task_cpu(p);

2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054
	/*
	 * 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();
2055 2056
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2057 2058

#ifdef CONFIG_SMP
2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077
	/*
	 * 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
2078
	/*
2079 2080
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
2081 2082 2083 2084 2085
	 *
	 * 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.
2086
	 */
2087
	smp_cond_load_acquire(&p->on_cpu, !VAL);
L
Linus Torvalds 已提交
2088

2089
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2090
	p->state = TASK_WAKING;
2091

2092
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2093 2094
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2095
		set_task_cpu(p, cpu);
2096
	}
L
Linus Torvalds 已提交
2097 2098
#endif /* CONFIG_SMP */

2099
	ttwu_queue(p, cpu, wake_flags);
2100
stat:
2101
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2102
out:
2103
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2104 2105 2106 2107

	return success;
}

T
Tejun Heo 已提交
2108 2109 2110
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
2111
 * @cookie: context's cookie for pinning
T
Tejun Heo 已提交
2112
 *
2113
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2114
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2115
 * the current task.
T
Tejun Heo 已提交
2116
 */
2117
static void try_to_wake_up_local(struct task_struct *p, struct rq_flags *rf)
T
Tejun Heo 已提交
2118 2119 2120
{
	struct rq *rq = task_rq(p);

2121 2122 2123 2124
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2125 2126
	lockdep_assert_held(&rq->lock);

2127
	if (!raw_spin_trylock(&p->pi_lock)) {
2128 2129 2130 2131 2132 2133
		/*
		 * 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.
		 */
2134
		rq_unpin_lock(rq, rf);
2135 2136 2137
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
2138
		rq_repin_lock(rq, rf);
2139 2140
	}

T
Tejun Heo 已提交
2141
	if (!(p->state & TASK_NORMAL))
2142
		goto out;
T
Tejun Heo 已提交
2143

2144 2145
	trace_sched_waking(p);

2146
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
2147 2148
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

2149
	ttwu_do_wakeup(rq, p, 0, rf);
2150
	ttwu_stat(p, smp_processor_id(), 0);
2151 2152
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2153 2154
}

2155 2156 2157 2158 2159
/**
 * 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
2160 2161 2162
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2163 2164 2165 2166
 *
 * 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.
 */
2167
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2168
{
2169
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2170 2171 2172
}
EXPORT_SYMBOL(wake_up_process);

2173
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2174 2175 2176 2177
{
	return try_to_wake_up(p, state, 0);
}

2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189
/*
 * 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;
2190 2191 2192

	dl_se->dl_throttled = 0;
	dl_se->dl_yielded = 0;
2193 2194
}

L
Linus Torvalds 已提交
2195 2196 2197
/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2198 2199 2200
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2201
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2202
{
P
Peter Zijlstra 已提交
2203 2204 2205
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2206 2207
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2208
	p->se.prev_sum_exec_runtime	= 0;
2209
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2210
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2211
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2212

2213 2214 2215 2216
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2217
#ifdef CONFIG_SCHEDSTATS
2218
	/* Even if schedstat is disabled, there should not be garbage */
2219
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2220
#endif
N
Nick Piggin 已提交
2221

2222
	RB_CLEAR_NODE(&p->dl.rb_node);
2223
	init_dl_task_timer(&p->dl);
2224
	__dl_clear_params(p);
2225

P
Peter Zijlstra 已提交
2226
	INIT_LIST_HEAD(&p->rt.run_list);
2227 2228 2229 2230
	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
2231

2232 2233 2234
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2235 2236 2237

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2238
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2239 2240 2241
		p->mm->numa_scan_seq = 0;
	}

2242 2243 2244 2245 2246
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2247 2248
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2249
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2250
	p->numa_work.next = &p->numa_work;
2251
	p->numa_faults = NULL;
2252 2253
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2254 2255

	p->numa_group = NULL;
2256
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2257 2258
}

2259 2260
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2261
#ifdef CONFIG_NUMA_BALANCING
2262

2263 2264 2265
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2266
		static_branch_enable(&sched_numa_balancing);
2267
	else
2268
		static_branch_disable(&sched_numa_balancing);
2269
}
2270 2271 2272 2273 2274 2275 2276

#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;
2277
	int state = static_branch_likely(&sched_numa_balancing);
2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292

	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
2293

2294 2295
#ifdef CONFIG_SCHEDSTATS

2296
DEFINE_STATIC_KEY_FALSE(sched_schedstats);
2297
static bool __initdata __sched_schedstats = false;
2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320

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;

2321 2322 2323 2324 2325
	/*
	 * 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.
	 */
2326
	if (!strcmp(str, "enable")) {
2327
		__sched_schedstats = true;
2328 2329
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2330
		__sched_schedstats = false;
2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

2341 2342 2343 2344 2345
static void __init init_schedstats(void)
{
	set_schedstats(__sched_schedstats);
}

2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365
#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;
}
2366 2367 2368 2369
#endif /* CONFIG_PROC_SYSCTL */
#else  /* !CONFIG_SCHEDSTATS */
static inline void init_schedstats(void) {}
#endif /* CONFIG_SCHEDSTATS */
I
Ingo Molnar 已提交
2370 2371 2372 2373

/*
 * fork()/clone()-time setup:
 */
2374
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2375
{
2376
	unsigned long flags;
I
Ingo Molnar 已提交
2377 2378
	int cpu = get_cpu();

2379
	__sched_fork(clone_flags, p);
2380
	/*
2381
	 * We mark the process as NEW here. This guarantees that
2382 2383 2384
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2385
	p->state = TASK_NEW;
I
Ingo Molnar 已提交
2386

2387 2388 2389 2390 2391
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2392 2393 2394 2395
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2396
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2397
			p->policy = SCHED_NORMAL;
2398
			p->static_prio = NICE_TO_PRIO(0);
2399 2400 2401 2402 2403 2404
			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);
2405

2406 2407 2408 2409 2410 2411
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2412

2413 2414 2415 2416 2417 2418
	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
2419
		p->sched_class = &fair_sched_class;
2420
	}
2421

2422
	init_entity_runnable_average(&p->se);
P
Peter Zijlstra 已提交
2423

2424 2425 2426 2427 2428 2429 2430
	/*
	 * 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.
	 */
2431
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2432 2433 2434 2435 2436 2437 2438
	/*
	 * 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);
2439
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2440

2441
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2442
	if (likely(sched_info_on()))
2443
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2444
#endif
P
Peter Zijlstra 已提交
2445 2446
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2447
#endif
2448
	init_task_preempt_count(p);
2449
#ifdef CONFIG_SMP
2450
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2451
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2452
#endif
2453

N
Nick Piggin 已提交
2454
	put_cpu();
2455
	return 0;
L
Linus Torvalds 已提交
2456 2457
}

2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476
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)
{
2477 2478
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2479 2480 2481
	return &cpu_rq(i)->rd->dl_bw;
}

2482
static inline int dl_bw_cpus(int i)
2483
{
2484 2485 2486
	struct root_domain *rd = cpu_rq(i)->rd;
	int cpus = 0;

2487 2488
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2489 2490 2491 2492
	for_each_cpu_and(i, rd->span, cpu_active_mask)
		cpus++;

	return cpus;
2493 2494 2495 2496 2497 2498 2499
}
#else
inline struct dl_bw *dl_bw_of(int i)
{
	return &cpu_rq(i)->dl.dl_bw;
}

2500
static inline int dl_bw_cpus(int i)
2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512
{
	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.
2513 2514 2515
 *
 * XXX we should delay bw change until the task's 0-lag point, see
 * __setparam_dl().
2516 2517 2518 2519 2520 2521
 */
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));
2522
	u64 period = attr->sched_period ?: attr->sched_deadline;
2523 2524
	u64 runtime = attr->sched_runtime;
	u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2525
	int cpus, err = -1;
2526

2527 2528
	/* !deadline task may carry old deadline bandwidth */
	if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2529 2530 2531 2532 2533 2534 2535 2536
		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);
2537
	cpus = dl_bw_cpus(task_cpu(p));
2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557
	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
2558 2559 2560 2561 2562 2563 2564
/*
 * 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.
 */
2565
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2566
{
2567
	struct rq_flags rf;
I
Ingo Molnar 已提交
2568
	struct rq *rq;
2569

2570
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2571
	p->state = TASK_RUNNING;
2572 2573 2574 2575 2576
#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
2577 2578 2579
	 *
	 * Use __set_task_cpu() to avoid calling sched_class::migrate_task_rq,
	 * as we're not fully set-up yet.
2580
	 */
2581
	__set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2582
#endif
2583
	rq = __task_rq_lock(p, &rf);
2584
	update_rq_clock(rq);
2585
	post_init_entity_util_avg(&p->se);
2586

P
Peter Zijlstra 已提交
2587
	activate_task(rq, p, 0);
2588
	p->on_rq = TASK_ON_RQ_QUEUED;
2589
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2590
	check_preempt_curr(rq, p, WF_FORK);
2591
#ifdef CONFIG_SMP
2592 2593 2594 2595 2596
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2597
		rq_unpin_lock(rq, &rf);
2598
		p->sched_class->task_woken(rq, p);
2599
		rq_repin_lock(rq, &rf);
2600
	}
2601
#endif
2602
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2603 2604
}

2605 2606
#ifdef CONFIG_PREEMPT_NOTIFIERS

2607 2608
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620
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);

2621
/**
2622
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2623
 * @notifier: notifier struct to register
2624 2625 2626
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2627 2628 2629
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2630 2631 2632 2633 2634 2635
	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
2636
 * @notifier: notifier struct to unregister
2637
 *
2638
 * This is *not* safe to call from within a preemption notifier.
2639 2640 2641 2642 2643 2644 2645
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2646
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2647 2648 2649
{
	struct preempt_notifier *notifier;

2650
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2651 2652 2653
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2654 2655 2656 2657 2658 2659
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);
}

2660
static void
2661 2662
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2663 2664 2665
{
	struct preempt_notifier *notifier;

2666
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2667 2668 2669
		notifier->ops->sched_out(notifier, next);
}

2670 2671 2672 2673 2674 2675 2676 2677
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);
}

2678
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2679

2680
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2681 2682 2683
{
}

2684
static inline void
2685 2686 2687 2688 2689
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2690
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2691

2692 2693 2694
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2695
 * @prev: the current task that is being switched out
2696 2697 2698 2699 2700 2701 2702 2703 2704
 * @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.
 */
2705 2706 2707
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2708
{
2709
	sched_info_switch(rq, prev, next);
2710
	perf_event_task_sched_out(prev, next);
2711
	fire_sched_out_preempt_notifiers(prev, next);
2712 2713 2714 2715
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2716 2717 2718 2719
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2720 2721 2722 2723
 * 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
2724 2725
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2726
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2727 2728
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2729 2730 2731 2732 2733
 *
 * 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
2734
 */
2735
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2736 2737
	__releases(rq->lock)
{
2738
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2739
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2740
	long prev_state;
L
Linus Torvalds 已提交
2741

2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
	/*
	 * 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.
	 */
2753 2754 2755 2756
	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);
2757

L
Linus Torvalds 已提交
2758 2759 2760 2761
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2762
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2763 2764
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2765 2766 2767 2768 2769
	 *
	 * 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
2770
	 */
O
Oleg Nesterov 已提交
2771
	prev_state = prev->state;
2772
	vtime_task_switch(prev);
2773
	perf_event_task_sched_in(prev, current);
2774
	finish_lock_switch(rq, prev);
2775
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2776

2777
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
2778 2779
	if (mm)
		mmdrop(mm);
2780
	if (unlikely(prev_state == TASK_DEAD)) {
2781 2782 2783
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2784 2785 2786
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2787
		 */
2788
		kprobe_flush_task(prev);
2789 2790 2791 2792

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

L
Linus Torvalds 已提交
2793
		put_task_struct(prev);
2794
	}
2795

2796
	tick_nohz_task_switch();
2797
	return rq;
L
Linus Torvalds 已提交
2798 2799
}

2800 2801 2802
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2803
static void __balance_callback(struct rq *rq)
2804
{
2805 2806 2807
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2808

2809 2810 2811 2812 2813 2814 2815 2816
	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;
2817

2818
		func(rq);
2819
	}
2820 2821 2822 2823 2824 2825 2826
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2827 2828 2829
}

#else
2830

2831
static inline void balance_callback(struct rq *rq)
2832
{
L
Linus Torvalds 已提交
2833 2834
}

2835 2836
#endif

L
Linus Torvalds 已提交
2837 2838 2839 2840
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2841
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2842 2843
	__releases(rq->lock)
{
2844
	struct rq *rq;
2845

2846 2847 2848 2849 2850 2851 2852 2853 2854
	/*
	 * 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).
	 */

2855
	rq = finish_task_switch(prev);
2856
	balance_callback(rq);
2857
	preempt_enable();
2858

L
Linus Torvalds 已提交
2859
	if (current->set_child_tid)
2860
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2861 2862 2863
}

/*
2864
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2865
 */
2866
static __always_inline struct rq *
2867
context_switch(struct rq *rq, struct task_struct *prev,
2868
	       struct task_struct *next, struct rq_flags *rf)
L
Linus Torvalds 已提交
2869
{
I
Ingo Molnar 已提交
2870
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2871

2872
	prepare_task_switch(rq, prev, next);
2873

I
Ingo Molnar 已提交
2874 2875
	mm = next->mm;
	oldmm = prev->active_mm;
2876 2877 2878 2879 2880
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2881
	arch_start_context_switch(prev);
2882

2883
	if (!mm) {
L
Linus Torvalds 已提交
2884 2885 2886 2887
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
2888
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2889

2890
	if (!prev->mm) {
L
Linus Torvalds 已提交
2891 2892 2893
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2894

2895
	rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
2896

2897 2898 2899 2900 2901 2902
	/*
	 * 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:
	 */
2903
	rq_unpin_lock(rq, rf);
2904
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2905 2906 2907

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2908
	barrier();
2909 2910

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2911 2912 2913
}

/*
2914
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2915 2916
 *
 * externally visible scheduler statistics: current number of runnable
2917
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2918 2919 2920 2921 2922 2923 2924 2925 2926
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2927
}
L
Linus Torvalds 已提交
2928

2929 2930
/*
 * Check if only the current task is running on the cpu.
2931 2932 2933 2934 2935 2936 2937 2938 2939 2940
 *
 * 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)
2941 2942 2943
 */
bool single_task_running(void)
{
2944
	return raw_rq()->nr_running == 1;
2945 2946 2947
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2948
unsigned long long nr_context_switches(void)
2949
{
2950 2951
	int i;
	unsigned long long sum = 0;
2952

2953
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2954
		sum += cpu_rq(i)->nr_switches;
2955

L
Linus Torvalds 已提交
2956 2957
	return sum;
}
2958

L
Linus Torvalds 已提交
2959 2960 2961
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2962

2963
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2964
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2965

L
Linus Torvalds 已提交
2966 2967
	return sum;
}
2968

2969
unsigned long nr_iowait_cpu(int cpu)
2970
{
2971
	struct rq *this = cpu_rq(cpu);
2972 2973
	return atomic_read(&this->nr_iowait);
}
2974

2975 2976
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2977 2978 2979
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2980 2981
}

I
Ingo Molnar 已提交
2982
#ifdef CONFIG_SMP
2983

2984
/*
P
Peter Zijlstra 已提交
2985 2986
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2987
 */
P
Peter Zijlstra 已提交
2988
void sched_exec(void)
2989
{
P
Peter Zijlstra 已提交
2990
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2991
	unsigned long flags;
2992
	int dest_cpu;
2993

2994
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2995
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2996 2997
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2998

2999
	if (likely(cpu_active(dest_cpu))) {
3000
		struct migration_arg arg = { p, dest_cpu };
3001

3002 3003
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
3004 3005
		return;
	}
3006
unlock:
3007
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
3008
}
I
Ingo Molnar 已提交
3009

L
Linus Torvalds 已提交
3010 3011 3012
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
3013
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
3014 3015

EXPORT_PER_CPU_SYMBOL(kstat);
3016
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
3017

3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034
/*
 * 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);
}

3035 3036 3037 3038 3039 3040 3041
/*
 * 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)
{
3042
	struct rq_flags rf;
3043
	struct rq *rq;
3044
	u64 ns;
3045

3046 3047 3048 3049 3050 3051 3052 3053 3054
#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.
3055 3056
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
3057
	 */
3058
	if (!p->on_cpu || !task_on_rq_queued(p))
3059 3060 3061
		return p->se.sum_exec_runtime;
#endif

3062
	rq = task_rq_lock(p, &rf);
3063 3064 3065 3066 3067 3068
	/*
	 * 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)) {
3069
		prefetch_curr_exec_start(p);
3070 3071 3072 3073
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
3074
	task_rq_unlock(rq, p, &rf);
3075 3076 3077

	return ns;
}
3078

3079 3080 3081 3082 3083 3084 3085 3086
/*
 * 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
3087
	struct task_struct *curr = rq->curr;
3088 3089

	sched_clock_tick();
I
Ingo Molnar 已提交
3090

3091
	raw_spin_lock(&rq->lock);
3092
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3093
	curr->sched_class->task_tick(rq, curr, 0);
3094
	cpu_load_update_active(rq);
3095
	calc_global_load_tick(rq);
3096
	raw_spin_unlock(&rq->lock);
3097

3098
	perf_event_task_tick();
3099

3100
#ifdef CONFIG_SMP
3101
	rq->idle_balance = idle_cpu(cpu);
3102
	trigger_load_balance(rq);
3103
#endif
3104
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
3105 3106
}

3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117
#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.
3118 3119
 *
 * Return: Maximum deferment in nanoseconds.
3120 3121 3122 3123
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
3124
	unsigned long next, now = READ_ONCE(jiffies);
3125 3126 3127 3128 3129 3130

	next = rq->last_sched_tick + HZ;

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

3131
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
3132
}
3133
#endif
L
Linus Torvalds 已提交
3134

3135 3136
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150
/*
 * 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);
	}
}
3151

3152
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3153
{
3154
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3155 3156 3157
	/*
	 * Underflow?
	 */
3158 3159
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3160
#endif
3161
	__preempt_count_add(val);
3162
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3163 3164 3165
	/*
	 * Spinlock count overflowing soon?
	 */
3166 3167
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3168
#endif
3169
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3170
}
3171
EXPORT_SYMBOL(preempt_count_add);
3172
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3173

3174 3175 3176 3177 3178 3179 3180 3181 3182 3183
/*
 * 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());
}

3184
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3185
{
3186
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3187 3188 3189
	/*
	 * Underflow?
	 */
3190
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3191
		return;
L
Linus Torvalds 已提交
3192 3193 3194
	/*
	 * Is the spinlock portion underflowing?
	 */
3195 3196 3197
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3198
#endif
3199

3200
	preempt_latency_stop(val);
3201
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3202
}
3203
EXPORT_SYMBOL(preempt_count_sub);
3204
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3205

3206 3207 3208
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3209 3210 3211
#endif

/*
I
Ingo Molnar 已提交
3212
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3213
 */
I
Ingo Molnar 已提交
3214
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3215
{
3216 3217 3218
	/* Save this before calling printk(), since that will clobber it */
	unsigned long preempt_disable_ip = get_preempt_disable_ip(current);

3219 3220 3221
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3222 3223
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3224

I
Ingo Molnar 已提交
3225
	debug_show_held_locks(prev);
3226
	print_modules();
I
Ingo Molnar 已提交
3227 3228
	if (irqs_disabled())
		print_irqtrace_events(prev);
3229 3230
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && in_atomic_preempt_off()) {
3231
		pr_err("Preemption disabled at:");
3232
		print_ip_sym(preempt_disable_ip);
3233 3234
		pr_cont("\n");
	}
3235 3236 3237
	if (panic_on_warn)
		panic("scheduling while atomic\n");

3238
	dump_stack();
3239
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3240
}
L
Linus Torvalds 已提交
3241

I
Ingo Molnar 已提交
3242 3243 3244 3245 3246
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3247
#ifdef CONFIG_SCHED_STACK_END_CHECK
J
Jann Horn 已提交
3248 3249
	if (task_stack_end_corrupted(prev))
		panic("corrupted stack end detected inside scheduler\n");
3250
#endif
3251

3252
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3253
		__schedule_bug(prev);
3254 3255
		preempt_count_set(PREEMPT_DISABLED);
	}
3256
	rcu_sleep_check();
I
Ingo Molnar 已提交
3257

L
Linus Torvalds 已提交
3258 3259
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3260
	schedstat_inc(this_rq()->sched_count);
I
Ingo Molnar 已提交
3261 3262 3263 3264 3265 3266
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3267
pick_next_task(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
I
Ingo Molnar 已提交
3268
{
3269
	const struct sched_class *class = &fair_sched_class;
I
Ingo Molnar 已提交
3270
	struct task_struct *p;
L
Linus Torvalds 已提交
3271 3272

	/*
I
Ingo Molnar 已提交
3273 3274
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
3275
	 */
3276
	if (likely(prev->sched_class == class &&
3277
		   rq->nr_running == rq->cfs.h_nr_running)) {
3278
		p = fair_sched_class.pick_next_task(rq, prev, rf);
3279 3280 3281 3282 3283
		if (unlikely(p == RETRY_TASK))
			goto again;

		/* assumes fair_sched_class->next == idle_sched_class */
		if (unlikely(!p))
3284
			p = idle_sched_class.pick_next_task(rq, prev, rf);
3285 3286

		return p;
L
Linus Torvalds 已提交
3287 3288
	}

3289
again:
3290
	for_each_class(class) {
3291
		p = class->pick_next_task(rq, prev, rf);
3292 3293 3294
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3295
			return p;
3296
		}
I
Ingo Molnar 已提交
3297
	}
3298 3299

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
3300
}
L
Linus Torvalds 已提交
3301

I
Ingo Molnar 已提交
3302
/*
3303
 * __schedule() is the main scheduler function.
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 3331 3332 3333 3334 3335 3336 3337
 *
 * 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
3338
 *
3339
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3340
 */
3341
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3342 3343
{
	struct task_struct *prev, *next;
3344
	unsigned long *switch_count;
3345
	struct rq_flags rf;
I
Ingo Molnar 已提交
3346
	struct rq *rq;
3347
	int cpu;
I
Ingo Molnar 已提交
3348 3349 3350 3351 3352 3353

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

	schedule_debug(prev);
L
Linus Torvalds 已提交
3354

3355
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3356
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3357

3358 3359 3360
	local_irq_disable();
	rcu_note_context_switch();

3361 3362 3363 3364 3365 3366
	/*
	 * 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();
3367
	raw_spin_lock(&rq->lock);
3368
	rq_pin_lock(rq, &rf);
L
Linus Torvalds 已提交
3369

3370
	rq->clock_update_flags <<= 1; /* promote REQ to ACT */
3371

3372
	switch_count = &prev->nivcsw;
3373
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3374
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3375
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3376
		} else {
3377 3378 3379
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
3380
			/*
3381 3382 3383
			 * 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
3384 3385 3386 3387
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3388
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3389
				if (to_wakeup)
3390
					try_to_wake_up_local(to_wakeup, &rf);
T
Tejun Heo 已提交
3391 3392
			}
		}
I
Ingo Molnar 已提交
3393
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3394 3395
	}

3396
	if (task_on_rq_queued(prev))
3397 3398
		update_rq_clock(rq);

3399
	next = pick_next_task(rq, prev, &rf);
3400
	clear_tsk_need_resched(prev);
3401
	clear_preempt_need_resched();
L
Linus Torvalds 已提交
3402 3403 3404 3405 3406 3407

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

3408
		trace_sched_switch(preempt, prev, next);
3409
		rq = context_switch(rq, prev, next, &rf); /* unlocks the rq */
3410
	} else {
3411
		rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
3412
		rq_unpin_lock(rq, &rf);
3413
		raw_spin_unlock_irq(&rq->lock);
3414
	}
L
Linus Torvalds 已提交
3415

3416
	balance_callback(rq);
L
Linus Torvalds 已提交
3417
}
3418

3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445
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 */
}

3446 3447
static inline void sched_submit_work(struct task_struct *tsk)
{
3448
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3449 3450 3451 3452 3453 3454 3455 3456 3457
		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);
}

3458
asmlinkage __visible void __sched schedule(void)
3459
{
3460 3461 3462
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3463
	do {
3464
		preempt_disable();
3465
		__schedule(false);
3466
		sched_preempt_enable_no_resched();
3467
	} while (need_resched());
3468
}
L
Linus Torvalds 已提交
3469 3470
EXPORT_SYMBOL(schedule);

3471
#ifdef CONFIG_CONTEXT_TRACKING
3472
asmlinkage __visible void __sched schedule_user(void)
3473 3474 3475 3476 3477 3478
{
	/*
	 * 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.
3479 3480
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3481
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3482
	 * too frequently to make sense yet.
3483
	 */
3484
	enum ctx_state prev_state = exception_enter();
3485
	schedule();
3486
	exception_exit(prev_state);
3487 3488 3489
}
#endif

3490 3491 3492 3493 3494 3495 3496
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3497
	sched_preempt_enable_no_resched();
3498 3499 3500 3501
	schedule();
	preempt_disable();
}

3502
static void __sched notrace preempt_schedule_common(void)
3503 3504
{
	do {
3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517
		/*
		 * 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.
		 */
3518
		preempt_disable_notrace();
3519
		preempt_latency_start(1);
3520
		__schedule(true);
3521
		preempt_latency_stop(1);
3522
		preempt_enable_no_resched_notrace();
3523 3524 3525 3526 3527 3528 3529 3530

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

L
Linus Torvalds 已提交
3531 3532
#ifdef CONFIG_PREEMPT
/*
3533
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3534
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3535 3536
 * occur there and call schedule directly.
 */
3537
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3538 3539 3540
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3541
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3542
	 */
3543
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3544 3545
		return;

3546
	preempt_schedule_common();
L
Linus Torvalds 已提交
3547
}
3548
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3549
EXPORT_SYMBOL(preempt_schedule);
3550 3551

/**
3552
 * preempt_schedule_notrace - preempt_schedule called by tracing
3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564
 *
 * 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.
 */
3565
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3566 3567 3568 3569 3570 3571 3572
{
	enum ctx_state prev_ctx;

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

	do {
3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585
		/*
		 * 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.
		 */
3586
		preempt_disable_notrace();
3587
		preempt_latency_start(1);
3588 3589 3590 3591 3592 3593
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3594
		__schedule(true);
3595 3596
		exception_exit(prev_ctx);

3597
		preempt_latency_stop(1);
3598
		preempt_enable_no_resched_notrace();
3599 3600
	} while (need_resched());
}
3601
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3602

3603
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3604 3605

/*
3606
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3607 3608 3609 3610
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3611
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3612
{
3613
	enum ctx_state prev_state;
3614

3615
	/* Catch callers which need to be fixed */
3616
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3617

3618 3619
	prev_state = exception_enter();

3620
	do {
3621
		preempt_disable();
3622
		local_irq_enable();
3623
		__schedule(true);
3624
		local_irq_disable();
3625
		sched_preempt_enable_no_resched();
3626
	} while (need_resched());
3627 3628

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3629 3630
}

P
Peter Zijlstra 已提交
3631
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3632
			  void *key)
L
Linus Torvalds 已提交
3633
{
P
Peter Zijlstra 已提交
3634
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3635 3636 3637
}
EXPORT_SYMBOL(default_wake_function);

3638 3639 3640 3641 3642 3643 3644 3645 3646 3647
#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().
 *
3648 3649
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3650
 */
3651
void rt_mutex_setprio(struct task_struct *p, int prio)
3652
{
3653
	int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE;
3654
	const struct sched_class *prev_class;
3655 3656
	struct rq_flags rf;
	struct rq *rq;
3657

3658
	BUG_ON(prio > MAX_PRIO);
3659

3660
	rq = __task_rq_lock(p, &rf);
3661
	update_rq_clock(rq);
3662

3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680
	/*
	 * 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;
	}

3681
	trace_sched_pi_setprio(p, prio);
3682
	oldprio = p->prio;
3683 3684 3685 3686

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

3687
	prev_class = p->sched_class;
3688
	queued = task_on_rq_queued(p);
3689
	running = task_current(rq, p);
3690
	if (queued)
3691
		dequeue_task(rq, p, queue_flag);
3692
	if (running)
3693
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3694

3695 3696 3697 3698 3699 3700 3701 3702 3703 3704
	/*
	 * 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)) {
3705 3706 3707
		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))) {
3708
			p->dl.dl_boosted = 1;
3709
			queue_flag |= ENQUEUE_REPLENISH;
3710 3711
		} else
			p->dl.dl_boosted = 0;
3712
		p->sched_class = &dl_sched_class;
3713 3714 3715 3716
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3717
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3718
		p->sched_class = &rt_sched_class;
3719 3720 3721
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3722 3723
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3724
		p->sched_class = &fair_sched_class;
3725
	}
I
Ingo Molnar 已提交
3726

3727 3728
	p->prio = prio;

3729
	if (queued)
3730
		enqueue_task(rq, p, queue_flag);
3731
	if (running)
3732
		set_curr_task(rq, p);
3733

P
Peter Zijlstra 已提交
3734
	check_class_changed(rq, p, prev_class, oldprio);
3735
out_unlock:
3736
	preempt_disable(); /* avoid rq from going away on us */
3737
	__task_rq_unlock(rq, &rf);
3738 3739 3740

	balance_callback(rq);
	preempt_enable();
3741 3742
}
#endif
3743

3744
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3745
{
P
Peter Zijlstra 已提交
3746 3747
	bool queued, running;
	int old_prio, delta;
3748
	struct rq_flags rf;
3749
	struct rq *rq;
L
Linus Torvalds 已提交
3750

3751
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3752 3753 3754 3755 3756
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3757
	rq = task_rq_lock(p, &rf);
3758 3759
	update_rq_clock(rq);

L
Linus Torvalds 已提交
3760 3761 3762 3763
	/*
	 * 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
3764
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3765
	 */
3766
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3767 3768 3769
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3770
	queued = task_on_rq_queued(p);
P
Peter Zijlstra 已提交
3771
	running = task_current(rq, p);
3772
	if (queued)
3773
		dequeue_task(rq, p, DEQUEUE_SAVE);
P
Peter Zijlstra 已提交
3774 3775
	if (running)
		put_prev_task(rq, p);
L
Linus Torvalds 已提交
3776 3777

	p->static_prio = NICE_TO_PRIO(nice);
3778
	set_load_weight(p);
3779 3780 3781
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3782

3783
	if (queued) {
3784
		enqueue_task(rq, p, ENQUEUE_RESTORE);
L
Linus Torvalds 已提交
3785
		/*
3786 3787
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3788
		 */
3789
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3790
			resched_curr(rq);
L
Linus Torvalds 已提交
3791
	}
P
Peter Zijlstra 已提交
3792 3793
	if (running)
		set_curr_task(rq, p);
L
Linus Torvalds 已提交
3794
out_unlock:
3795
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3796 3797 3798
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3799 3800 3801 3802 3803
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3804
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3805
{
3806
	/* convert nice value [19,-20] to rlimit style value [1,40] */
3807
	int nice_rlim = nice_to_rlimit(nice);
3808

3809
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3810 3811 3812
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3813 3814 3815 3816 3817 3818 3819 3820 3821
#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.
 */
3822
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3823
{
3824
	long nice, retval;
L
Linus Torvalds 已提交
3825 3826 3827 3828 3829 3830

	/*
	 * 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.
	 */
3831
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3832
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3833

3834
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3835 3836 3837
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851
	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.
 *
3852
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3853 3854 3855
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3856
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3857 3858 3859 3860 3861 3862 3863
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
3864 3865
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3866 3867 3868
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882
	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
3883 3884 3885 3886 3887
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
3888 3889
 *
 * Return: The idle task for the cpu @cpu.
L
Linus Torvalds 已提交
3890
 */
3891
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3892 3893 3894 3895 3896 3897 3898
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3899 3900
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3901
 */
A
Alexey Dobriyan 已提交
3902
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3903
{
3904
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3905 3906
}

3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921
/*
 * 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;
3922
	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
3923
	dl_se->flags = attr->sched_flags;
3924
	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944

	/*
	 * 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.
	 */
3945 3946
}

3947 3948 3949 3950 3951 3952
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

3953 3954
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
3955
{
3956 3957
	int policy = attr->sched_policy;

3958
	if (policy == SETPARAM_POLICY)
3959 3960
		policy = p->policy;

L
Linus Torvalds 已提交
3961
	p->policy = policy;
3962

3963 3964
	if (dl_policy(policy))
		__setparam_dl(p, attr);
3965
	else if (fair_policy(policy))
3966 3967
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

3968 3969 3970 3971 3972 3973
	/*
	 * __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;
3974
	p->normal_prio = normal_prio(p);
3975 3976
	set_load_weight(p);
}
3977

3978 3979
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
3980
			   const struct sched_attr *attr, bool keep_boost)
3981 3982
{
	__setscheduler_params(p, attr);
3983

3984
	/*
3985 3986
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
3987
	 */
3988 3989 3990 3991
	if (keep_boost)
		p->prio = rt_mutex_get_effective_prio(p, normal_prio(p));
	else
		p->prio = normal_prio(p);
3992

3993 3994 3995
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
3996 3997 3998
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
3999
}
4000 4001 4002 4003 4004 4005 4006 4007 4008

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;
4009
	attr->sched_period = dl_se->dl_period;
4010 4011 4012 4013 4014 4015
	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
4016
 * than the runtime, as well as the period of being zero or
4017
 * greater than deadline. Furthermore, we have to be sure that
4018 4019 4020 4021
 * 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).
4022 4023 4024 4025
 */
static bool
__checkparam_dl(const struct sched_attr *attr)
{
4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051
	/* 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;
4052 4053
}

4054 4055 4056 4057 4058 4059 4060 4061 4062 4063
/*
 * 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);
4064 4065
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
4066 4067 4068 4069
	rcu_read_unlock();
	return match;
}

4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083
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;
}

4084 4085
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
4086
				bool user, bool pi)
L
Linus Torvalds 已提交
4087
{
4088 4089
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
4090
	int retval, oldprio, oldpolicy = -1, queued, running;
4091
	int new_effective_prio, policy = attr->sched_policy;
4092
	const struct sched_class *prev_class;
4093
	struct rq_flags rf;
4094
	int reset_on_fork;
4095
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE;
4096
	struct rq *rq;
L
Linus Torvalds 已提交
4097

4098 4099
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
4100 4101
recheck:
	/* double check policy once rq lock held */
4102 4103
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
4104
		policy = oldpolicy = p->policy;
4105
	} else {
4106
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
4107

4108
		if (!valid_policy(policy))
4109 4110 4111
			return -EINVAL;
	}

4112 4113 4114
	if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK))
		return -EINVAL;

L
Linus Torvalds 已提交
4115 4116
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
4117 4118
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4119
	 */
4120
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4121
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4122
		return -EINVAL;
4123 4124
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4125 4126
		return -EINVAL;

4127 4128 4129
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4130
	if (user && !capable(CAP_SYS_NICE)) {
4131
		if (fair_policy(policy)) {
4132
			if (attr->sched_nice < task_nice(p) &&
4133
			    !can_nice(p, attr->sched_nice))
4134 4135 4136
				return -EPERM;
		}

4137
		if (rt_policy(policy)) {
4138 4139
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4140 4141 4142 4143 4144 4145

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

			/* can't increase priority */
4146 4147
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4148 4149
				return -EPERM;
		}
4150

4151 4152 4153 4154 4155 4156 4157 4158 4159
		 /*
		  * 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
4160
		/*
4161 4162
		 * 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
4163
		 */
4164
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4165
			if (!can_nice(p, task_nice(p)))
4166 4167
				return -EPERM;
		}
4168

4169
		/* can't change other user's priorities */
4170
		if (!check_same_owner(p))
4171
			return -EPERM;
4172 4173 4174 4175

		/* Normal users shall not reset the sched_reset_on_fork flag */
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4176
	}
L
Linus Torvalds 已提交
4177

4178
	if (user) {
4179
		retval = security_task_setscheduler(p);
4180 4181 4182 4183
		if (retval)
			return retval;
	}

4184 4185 4186
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
4187
	 *
L
Lucas De Marchi 已提交
4188
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4189 4190
	 * runqueue lock must be held.
	 */
4191
	rq = task_rq_lock(p, &rf);
4192
	update_rq_clock(rq);
4193

4194 4195 4196 4197
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
4198
		task_rq_unlock(rq, p, &rf);
4199 4200 4201
		return -EINVAL;
	}

4202
	/*
4203 4204
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4205
	 */
4206
	if (unlikely(policy == p->policy)) {
4207
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4208 4209 4210
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4211
		if (dl_policy(policy) && dl_param_changed(p, attr))
4212
			goto change;
4213

4214
		p->sched_reset_on_fork = reset_on_fork;
4215
		task_rq_unlock(rq, p, &rf);
4216 4217
		return 0;
	}
4218
change:
4219

4220
	if (user) {
4221
#ifdef CONFIG_RT_GROUP_SCHED
4222 4223 4224 4225 4226
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4227 4228
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4229
			task_rq_unlock(rq, p, &rf);
4230 4231 4232
			return -EPERM;
		}
#endif
4233 4234 4235 4236 4237 4238 4239 4240 4241
#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.
			 */
4242 4243
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4244
				task_rq_unlock(rq, p, &rf);
4245 4246 4247 4248 4249
				return -EPERM;
			}
		}
#endif
	}
4250

L
Linus Torvalds 已提交
4251 4252 4253
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4254
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4255 4256
		goto recheck;
	}
4257 4258 4259 4260 4261 4262

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4263
	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
4264
		task_rq_unlock(rq, p, &rf);
4265 4266 4267
		return -EBUSY;
	}

4268 4269 4270
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4271 4272 4273 4274 4275 4276 4277 4278 4279
	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);
4280 4281
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4282 4283
	}

4284
	queued = task_on_rq_queued(p);
4285
	running = task_current(rq, p);
4286
	if (queued)
4287
		dequeue_task(rq, p, queue_flags);
4288
	if (running)
4289
		put_prev_task(rq, p);
4290

4291
	prev_class = p->sched_class;
4292
	__setscheduler(rq, p, attr, pi);
4293

4294
	if (queued) {
4295 4296 4297 4298
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4299 4300
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4301

4302
		enqueue_task(rq, p, queue_flags);
4303
	}
4304
	if (running)
4305
		set_curr_task(rq, p);
4306

P
Peter Zijlstra 已提交
4307
	check_class_changed(rq, p, prev_class, oldprio);
4308
	preempt_disable(); /* avoid rq from going away on us */
4309
	task_rq_unlock(rq, p, &rf);
4310

4311 4312
	if (pi)
		rt_mutex_adjust_pi(p);
4313

4314 4315 4316 4317 4318
	/*
	 * Run balance callbacks after we've adjusted the PI chain.
	 */
	balance_callback(rq);
	preempt_enable();
4319

L
Linus Torvalds 已提交
4320 4321
	return 0;
}
4322

4323 4324 4325 4326 4327 4328 4329 4330 4331
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),
	};

4332 4333
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4334 4335 4336 4337 4338
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4339
	return __sched_setscheduler(p, &attr, check, true);
4340
}
4341 4342 4343 4344 4345 4346
/**
 * 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.
 *
4347 4348
 * Return: 0 on success. An error code otherwise.
 *
4349 4350 4351
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4352
		       const struct sched_param *param)
4353
{
4354
	return _sched_setscheduler(p, policy, param, true);
4355
}
L
Linus Torvalds 已提交
4356 4357
EXPORT_SYMBOL_GPL(sched_setscheduler);

4358 4359
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4360
	return __sched_setscheduler(p, attr, true, true);
4361 4362 4363
}
EXPORT_SYMBOL_GPL(sched_setattr);

4364 4365 4366 4367 4368 4369 4370 4371 4372 4373
/**
 * 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.
4374 4375
 *
 * Return: 0 on success. An error code otherwise.
4376 4377
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4378
			       const struct sched_param *param)
4379
{
4380
	return _sched_setscheduler(p, policy, param, false);
4381
}
4382
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4383

I
Ingo Molnar 已提交
4384 4385
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4386 4387 4388
{
	struct sched_param lparam;
	struct task_struct *p;
4389
	int retval;
L
Linus Torvalds 已提交
4390 4391 4392 4393 4394

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4395 4396 4397

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4398
	p = find_process_by_pid(pid);
4399 4400 4401
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4402

L
Linus Torvalds 已提交
4403 4404 4405
	return retval;
}

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 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467
/*
 * 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?
	 */
4468
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4469

4470
	return 0;
4471 4472 4473

err_size:
	put_user(sizeof(*attr), &uattr->size);
4474
	return -E2BIG;
4475 4476
}

L
Linus Torvalds 已提交
4477 4478 4479 4480 4481
/**
 * 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.
4482 4483
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4484
 */
4485 4486
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
4487
{
4488 4489 4490 4491
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4492 4493 4494 4495 4496 4497 4498
	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.
4499 4500
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4501
 */
4502
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4503
{
4504
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4505 4506
}

4507 4508 4509
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4510
 * @uattr: structure containing the extended parameters.
4511
 * @flags: for future extension.
4512
 */
4513 4514
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4515 4516 4517 4518 4519
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4520
	if (!uattr || pid < 0 || flags)
4521 4522
		return -EINVAL;

4523 4524 4525
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4526

4527
	if ((int)attr.sched_policy < 0)
4528
		return -EINVAL;
4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539

	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
4540 4541 4542
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4543 4544 4545
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4546
 */
4547
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4548
{
4549
	struct task_struct *p;
4550
	int retval;
L
Linus Torvalds 已提交
4551 4552

	if (pid < 0)
4553
		return -EINVAL;
L
Linus Torvalds 已提交
4554 4555

	retval = -ESRCH;
4556
	rcu_read_lock();
L
Linus Torvalds 已提交
4557 4558 4559 4560
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4561 4562
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4563
	}
4564
	rcu_read_unlock();
L
Linus Torvalds 已提交
4565 4566 4567 4568
	return retval;
}

/**
4569
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4570 4571
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4572 4573 4574
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4575
 */
4576
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4577
{
4578
	struct sched_param lp = { .sched_priority = 0 };
4579
	struct task_struct *p;
4580
	int retval;
L
Linus Torvalds 已提交
4581 4582

	if (!param || pid < 0)
4583
		return -EINVAL;
L
Linus Torvalds 已提交
4584

4585
	rcu_read_lock();
L
Linus Torvalds 已提交
4586 4587 4588 4589 4590 4591 4592 4593 4594
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4595 4596
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4597
	rcu_read_unlock();
L
Linus Torvalds 已提交
4598 4599 4600 4601 4602 4603 4604 4605 4606

	/*
	 * 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:
4607
	rcu_read_unlock();
L
Linus Torvalds 已提交
4608 4609 4610
	return retval;
}

4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633
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)
4634
				return -EFBIG;
4635 4636 4637 4638 4639
		}

		attr->size = usize;
	}

4640
	ret = copy_to_user(uattr, attr, attr->size);
4641 4642 4643
	if (ret)
		return -EFAULT;

4644
	return 0;
4645 4646 4647
}

/**
4648
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4649
 * @pid: the pid in question.
J
Juri Lelli 已提交
4650
 * @uattr: structure containing the extended parameters.
4651
 * @size: sizeof(attr) for fwd/bwd comp.
4652
 * @flags: for future extension.
4653
 */
4654 4655
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4656 4657 4658 4659 4660 4661 4662 4663
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4664
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677
		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;
4678 4679
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4680 4681 4682
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4683 4684
		attr.sched_priority = p->rt_priority;
	else
4685
		attr.sched_nice = task_nice(p);
4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4697
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4698
{
4699
	cpumask_var_t cpus_allowed, new_mask;
4700 4701
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4702

4703
	rcu_read_lock();
L
Linus Torvalds 已提交
4704 4705 4706

	p = find_process_by_pid(pid);
	if (!p) {
4707
		rcu_read_unlock();
L
Linus Torvalds 已提交
4708 4709 4710
		return -ESRCH;
	}

4711
	/* Prevent p going away */
L
Linus Torvalds 已提交
4712
	get_task_struct(p);
4713
	rcu_read_unlock();
L
Linus Torvalds 已提交
4714

4715 4716 4717 4718
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4719 4720 4721 4722 4723 4724 4725 4726
	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
4727
	retval = -EPERM;
E
Eric W. Biederman 已提交
4728 4729 4730 4731
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4732
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4733 4734 4735
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4736

4737
	retval = security_task_setscheduler(p);
4738
	if (retval)
4739
		goto out_free_new_mask;
4740

4741 4742 4743 4744

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

4745 4746 4747 4748 4749 4750 4751
	/*
	 * 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
4752 4753 4754
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4755
			retval = -EBUSY;
4756
			rcu_read_unlock();
4757
			goto out_free_new_mask;
4758
		}
4759
		rcu_read_unlock();
4760 4761
	}
#endif
P
Peter Zijlstra 已提交
4762
again:
4763
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4764

P
Paul Menage 已提交
4765
	if (!retval) {
4766 4767
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4768 4769 4770 4771 4772
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4773
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4774 4775 4776
			goto again;
		}
	}
4777
out_free_new_mask:
4778 4779 4780 4781
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4782 4783 4784 4785 4786
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4787
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4788
{
4789 4790 4791 4792 4793
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4794 4795 4796 4797 4798 4799 4800 4801
	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
4802 4803
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4804
 */
4805 4806
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4807
{
4808
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4809 4810
	int retval;

4811 4812
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4813

4814 4815 4816 4817 4818
	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
4819 4820
}

4821
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4822
{
4823
	struct task_struct *p;
4824
	unsigned long flags;
L
Linus Torvalds 已提交
4825 4826
	int retval;

4827
	rcu_read_lock();
L
Linus Torvalds 已提交
4828 4829 4830 4831 4832 4833

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

4834 4835 4836 4837
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4838
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4839
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4840
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4841 4842

out_unlock:
4843
	rcu_read_unlock();
L
Linus Torvalds 已提交
4844

4845
	return retval;
L
Linus Torvalds 已提交
4846 4847 4848 4849 4850 4851 4852
}

/**
 * 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
4853
 *
4854 4855
 * Return: size of CPU mask copied to user_mask_ptr on success. An
 * error code otherwise.
L
Linus Torvalds 已提交
4856
 */
4857 4858
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4859 4860
{
	int ret;
4861
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4862

A
Anton Blanchard 已提交
4863
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4864 4865
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4866 4867
		return -EINVAL;

4868 4869
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4870

4871 4872
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4873
		size_t retlen = min_t(size_t, len, cpumask_size());
4874 4875

		if (copy_to_user(user_mask_ptr, mask, retlen))
4876 4877
			ret = -EFAULT;
		else
4878
			ret = retlen;
4879 4880
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4881

4882
	return ret;
L
Linus Torvalds 已提交
4883 4884 4885 4886 4887
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4888 4889
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4890 4891
 *
 * Return: 0.
L
Linus Torvalds 已提交
4892
 */
4893
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4894
{
4895
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
4896

4897
	schedstat_inc(rq->yld_count);
4898
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4899 4900 4901 4902 4903 4904

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
4905
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4906
	do_raw_spin_unlock(&rq->lock);
4907
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4908 4909 4910 4911 4912 4913

	schedule();

	return 0;
}

4914
#ifndef CONFIG_PREEMPT
4915
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4916
{
4917
	if (should_resched(0)) {
4918
		preempt_schedule_common();
L
Linus Torvalds 已提交
4919 4920 4921 4922
		return 1;
	}
	return 0;
}
4923
EXPORT_SYMBOL(_cond_resched);
4924
#endif
L
Linus Torvalds 已提交
4925 4926

/*
4927
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4928 4929
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4930
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4931 4932 4933
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4934
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4935
{
4936
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4937 4938
	int ret = 0;

4939 4940
	lockdep_assert_held(lock);

4941
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4942
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4943
		if (resched)
4944
			preempt_schedule_common();
N
Nick Piggin 已提交
4945 4946
		else
			cpu_relax();
J
Jan Kara 已提交
4947
		ret = 1;
L
Linus Torvalds 已提交
4948 4949
		spin_lock(lock);
	}
J
Jan Kara 已提交
4950
	return ret;
L
Linus Torvalds 已提交
4951
}
4952
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4953

4954
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4955 4956 4957
{
	BUG_ON(!in_softirq());

4958
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4959
		local_bh_enable();
4960
		preempt_schedule_common();
L
Linus Torvalds 已提交
4961 4962 4963 4964 4965
		local_bh_disable();
		return 1;
	}
	return 0;
}
4966
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4967 4968 4969 4970

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988
 * 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
4989 4990 4991 4992 4993 4994 4995 4996
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4997 4998 4999 5000
/**
 * 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
5001 5002
 * @p: target task
 * @preempt: whether task preemption is allowed or not
5003 5004 5005 5006
 *
 * 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.
 *
5007
 * Return:
5008 5009 5010
 *	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.
5011
 */
5012
int __sched yield_to(struct task_struct *p, bool preempt)
5013 5014 5015 5016
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
5017
	int yielded = 0;
5018 5019 5020 5021 5022 5023

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
5024 5025 5026 5027 5028 5029 5030 5031 5032
	/*
	 * 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;
	}

5033
	double_rq_lock(rq, p_rq);
5034
	if (task_rq(p) != p_rq) {
5035 5036 5037 5038 5039
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
5040
		goto out_unlock;
5041 5042

	if (curr->sched_class != p->sched_class)
5043
		goto out_unlock;
5044 5045

	if (task_running(p_rq, p) || p->state)
5046
		goto out_unlock;
5047 5048

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
5049
	if (yielded) {
5050
		schedstat_inc(rq->yld_count);
5051 5052 5053 5054 5055
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
5056
			resched_curr(p_rq);
5057
	}
5058

5059
out_unlock:
5060
	double_rq_unlock(rq, p_rq);
5061
out_irq:
5062 5063
	local_irq_restore(flags);

5064
	if (yielded > 0)
5065 5066 5067 5068 5069 5070
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
5071
/*
I
Ingo Molnar 已提交
5072
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
5073 5074 5075 5076
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
5077 5078
	int old_iowait = current->in_iowait;
	struct rq *rq;
L
Linus Torvalds 已提交
5079 5080
	long ret;

5081
	current->in_iowait = 1;
5082
	blk_schedule_flush_plug(current);
5083

5084
	delayacct_blkio_start();
5085
	rq = raw_rq();
L
Linus Torvalds 已提交
5086 5087
	atomic_inc(&rq->nr_iowait);
	ret = schedule_timeout(timeout);
5088
	current->in_iowait = old_iowait;
L
Linus Torvalds 已提交
5089
	atomic_dec(&rq->nr_iowait);
5090
	delayacct_blkio_end();
5091

L
Linus Torvalds 已提交
5092 5093
	return ret;
}
5094
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
5095 5096 5097 5098 5099

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
5100 5101 5102
 * 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
5103
 */
5104
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
5105 5106 5107 5108 5109 5110 5111 5112
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
5113
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5114
	case SCHED_NORMAL:
5115
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5116
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5117 5118 5119 5120 5121 5122 5123 5124 5125 5126
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5127 5128 5129
 * 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
5130
 */
5131
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5132 5133 5134 5135 5136 5137 5138 5139
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5140
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5141
	case SCHED_NORMAL:
5142
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5143
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155
		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.
5156 5157 5158
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
L
Linus Torvalds 已提交
5159
 */
5160
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5161
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
5162
{
5163
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5164
	unsigned int time_slice;
5165 5166
	struct rq_flags rf;
	struct timespec t;
5167
	struct rq *rq;
5168
	int retval;
L
Linus Torvalds 已提交
5169 5170

	if (pid < 0)
5171
		return -EINVAL;
L
Linus Torvalds 已提交
5172 5173

	retval = -ESRCH;
5174
	rcu_read_lock();
L
Linus Torvalds 已提交
5175 5176 5177 5178 5179 5180 5181 5182
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5183
	rq = task_rq_lock(p, &rf);
5184 5185 5186
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5187
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5188

5189
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
5190
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
5191 5192
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
5193

L
Linus Torvalds 已提交
5194
out_unlock:
5195
	rcu_read_unlock();
L
Linus Torvalds 已提交
5196 5197 5198
	return retval;
}

5199
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5200

5201
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5202 5203
{
	unsigned long free = 0;
5204
	int ppid;
5205
	unsigned long state = p->state;
L
Linus Torvalds 已提交
5206

5207 5208
	if (!try_get_task_stack(p))
		return;
5209 5210
	if (state)
		state = __ffs(state) + 1;
5211
	printk(KERN_INFO "%-15.15s %c", p->comm,
5212
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
L
Linus Torvalds 已提交
5213
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5214
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5215
#ifdef CONFIG_DEBUG_STACK_USAGE
5216
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5217
#endif
5218
	ppid = 0;
5219
	rcu_read_lock();
5220 5221
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5222
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5223
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5224
		task_pid_nr(p), ppid,
5225
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5226

5227
	print_worker_info(KERN_INFO, p);
5228
	show_stack(p, NULL);
5229
	put_task_stack(p);
L
Linus Torvalds 已提交
5230 5231
}

I
Ingo Molnar 已提交
5232
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5233
{
5234
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5235

5236
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5237 5238
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5239
#else
P
Peter Zijlstra 已提交
5240 5241
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5242
#endif
5243
	rcu_read_lock();
5244
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5245 5246
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5247
		 * console might take a lot of time:
5248 5249 5250
		 * 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
5251 5252
		 */
		touch_nmi_watchdog();
5253
		touch_all_softlockup_watchdogs();
I
Ingo Molnar 已提交
5254
		if (!state_filter || (p->state & state_filter))
5255
			sched_show_task(p);
5256
	}
L
Linus Torvalds 已提交
5257

I
Ingo Molnar 已提交
5258
#ifdef CONFIG_SCHED_DEBUG
5259 5260
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5261
#endif
5262
	rcu_read_unlock();
I
Ingo Molnar 已提交
5263 5264 5265
	/*
	 * Only show locks if all tasks are dumped:
	 */
5266
	if (!state_filter)
I
Ingo Molnar 已提交
5267
		debug_show_all_locks();
L
Linus Torvalds 已提交
5268 5269
}

5270
void init_idle_bootup_task(struct task_struct *idle)
I
Ingo Molnar 已提交
5271
{
I
Ingo Molnar 已提交
5272
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
5273 5274
}

5275 5276 5277 5278 5279 5280 5281 5282
/**
 * 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.
 */
5283
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5284
{
5285
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5286 5287
	unsigned long flags;

5288 5289
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5290

5291
	__sched_fork(0, idle);
5292
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5293
	idle->se.exec_start = sched_clock();
5294
	idle->flags |= PF_IDLE;
I
Ingo Molnar 已提交
5295

5296 5297
	kasan_unpoison_task_stack(idle);

5298 5299 5300 5301 5302 5303 5304 5305 5306
#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
5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317
	/*
	 * 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
5318
	__set_task_cpu(idle, cpu);
5319
	rcu_read_unlock();
L
Linus Torvalds 已提交
5320 5321

	rq->curr = rq->idle = idle;
5322
	idle->on_rq = TASK_ON_RQ_QUEUED;
5323
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5324
	idle->on_cpu = 1;
5325
#endif
5326 5327
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5328 5329

	/* Set the preempt count _outside_ the spinlocks! */
5330
	init_idle_preempt_count(idle, cpu);
5331

I
Ingo Molnar 已提交
5332 5333 5334 5335
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5336
	ftrace_graph_init_idle_task(idle, cpu);
5337
	vtime_init_idle(idle, cpu);
5338
#ifdef CONFIG_SMP
5339 5340
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5341 5342
}

5343 5344 5345 5346 5347 5348 5349
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;

5350 5351 5352
	if (!cpumask_weight(cur))
		return ret;

5353
	rcu_read_lock_sched();
5354 5355 5356 5357 5358 5359 5360 5361
	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);
5362
	rcu_read_unlock_sched();
5363 5364 5365 5366

	return ret;
}

5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390
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);
5391
		struct dl_bw *dl_b;
5392 5393 5394 5395
		bool overflow;
		int cpus;
		unsigned long flags;

5396 5397
		rcu_read_lock_sched();
		dl_b = dl_bw_of(dest_cpu);
5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412
		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);
5413
		rcu_read_unlock_sched();
5414 5415 5416 5417 5418 5419 5420

	}
#endif
out:
	return ret;
}

L
Linus Torvalds 已提交
5421 5422
#ifdef CONFIG_SMP

5423 5424
static bool sched_smp_initialized __read_mostly;

5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439
#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 */

5440
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5441 5442
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5443 5444 5445 5446 5447 5448 5449

/*
 * 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)
{
5450
	bool queued, running;
5451 5452
	struct rq_flags rf;
	struct rq *rq;
5453

5454
	rq = task_rq_lock(p, &rf);
5455
	queued = task_on_rq_queued(p);
5456 5457
	running = task_current(rq, p);

5458
	if (queued)
5459
		dequeue_task(rq, p, DEQUEUE_SAVE);
5460
	if (running)
5461
		put_prev_task(rq, p);
5462 5463 5464

	p->numa_preferred_nid = nid;

5465
	if (queued)
5466
		enqueue_task(rq, p, ENQUEUE_RESTORE);
5467
	if (running)
5468
		set_curr_task(rq, p);
5469
	task_rq_unlock(rq, p, &rf);
5470
}
P
Peter Zijlstra 已提交
5471
#endif /* CONFIG_NUMA_BALANCING */
5472

L
Linus Torvalds 已提交
5473
#ifdef CONFIG_HOTPLUG_CPU
5474
/*
5475 5476
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
5477
 */
5478
void idle_task_exit(void)
L
Linus Torvalds 已提交
5479
{
5480
	struct mm_struct *mm = current->active_mm;
5481

5482
	BUG_ON(cpu_online(smp_processor_id()));
5483

5484
	if (mm != &init_mm) {
5485
		switch_mm_irqs_off(mm, &init_mm, current);
5486 5487
		finish_arch_post_lock_switch();
	}
5488
	mmdrop(mm);
L
Linus Torvalds 已提交
5489 5490 5491
}

/*
5492 5493
 * 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
5494 5495 5496
 * 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.
5497 5498
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5499
 */
5500
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5501
{
5502
	long delta = calc_load_fold_active(rq, 1);
5503 5504
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5505 5506
}

5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522
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,
};

5523
/*
5524 5525 5526 5527 5528 5529
 * 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
5530
 */
5531
static void migrate_tasks(struct rq *dead_rq)
L
Linus Torvalds 已提交
5532
{
5533
	struct rq *rq = dead_rq;
5534
	struct task_struct *next, *stop = rq->stop;
5535
	struct rq_flags rf;
5536
	int dest_cpu;
L
Linus Torvalds 已提交
5537 5538

	/*
5539 5540 5541 5542 5543 5544 5545
	 * 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
5546
	 */
5547
	rq->stop = NULL;
5548

5549 5550 5551 5552 5553 5554 5555
	/*
	 * 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);

5556
	for (;;) {
5557 5558 5559 5560 5561
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5562
			break;
5563

5564
		/*
W
Wanpeng Li 已提交
5565
		 * pick_next_task assumes pinned rq->lock.
5566
		 */
5567 5568
		rq_pin_lock(rq, &rf);
		next = pick_next_task(rq, &fake_task, &rf);
5569
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
5570
		next->sched_class->put_prev_task(rq, next);
5571

W
Wanpeng Li 已提交
5572 5573 5574 5575 5576 5577 5578 5579 5580
		/*
		 * 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.
		 */
5581
		rq_unpin_lock(rq, &rf);
W
Wanpeng Li 已提交
5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595
		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;
		}

5596
		/* Find suitable destination for @next, with force if needed. */
5597
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5598

5599 5600 5601 5602 5603 5604
		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
5605
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5606
	}
5607

5608
	rq->stop = stop;
5609
}
L
Linus Torvalds 已提交
5610 5611
#endif /* CONFIG_HOTPLUG_CPU */

5612 5613 5614 5615 5616
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

5617
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636
		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);
		}

5637
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5638 5639 5640 5641
		rq->online = 0;
	}
}

5642
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5643
{
5644
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5645

5646 5647 5648
	rq->age_stamp = sched_clock_cpu(cpu);
}

5649 5650
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

5651
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
5652

5653
static __read_mostly int sched_debug_enabled;
5654

5655
static int __init sched_debug_setup(char *str)
5656
{
5657
	sched_debug_enabled = 1;
5658 5659 5660

	return 0;
}
5661 5662 5663 5664 5665 5666
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
5667

5668
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
5669
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
5670
{
I
Ingo Molnar 已提交
5671
	struct sched_group *group = sd->groups;
L
Linus Torvalds 已提交
5672

5673
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
5674 5675 5676 5677

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

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
5678
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
5679
		if (sd->parent)
P
Peter Zijlstra 已提交
5680 5681
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
5682
		return -1;
N
Nick Piggin 已提交
5683 5684
	}

5685 5686
	printk(KERN_CONT "span %*pbl level %s\n",
	       cpumask_pr_args(sched_domain_span(sd)), sd->name);
I
Ingo Molnar 已提交
5687

5688
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
5689 5690
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
5691
	}
5692
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5693 5694
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
5695
	}
L
Linus Torvalds 已提交
5696

I
Ingo Molnar 已提交
5697
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
5698
	do {
I
Ingo Molnar 已提交
5699
		if (!group) {
P
Peter Zijlstra 已提交
5700 5701
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
5702 5703 5704
			break;
		}

5705
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5706 5707
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
5708 5709
			break;
		}
L
Linus Torvalds 已提交
5710

5711 5712
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5713 5714
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
5715 5716
			break;
		}
L
Linus Torvalds 已提交
5717

5718
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
5719

5720 5721
		printk(KERN_CONT " %*pbl",
		       cpumask_pr_args(sched_group_cpus(group)));
5722
		if (group->sgc->capacity != SCHED_CAPACITY_SCALE) {
5723
			printk(KERN_CONT " (cpu_capacity = %lu)",
5724
				group->sgc->capacity);
5725
		}
L
Linus Torvalds 已提交
5726

I
Ingo Molnar 已提交
5727 5728
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
5729
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
5730

5731
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
5732
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
5733

5734 5735
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
5736 5737
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
5738 5739
	return 0;
}
L
Linus Torvalds 已提交
5740

I
Ingo Molnar 已提交
5741 5742 5743
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
5744

5745
	if (!sched_debug_enabled)
5746 5747
		return;

I
Ingo Molnar 已提交
5748 5749 5750 5751
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
5752

I
Ingo Molnar 已提交
5753 5754 5755
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
5756
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
5757
			break;
L
Linus Torvalds 已提交
5758 5759
		level++;
		sd = sd->parent;
5760
		if (!sd)
I
Ingo Molnar 已提交
5761 5762
			break;
	}
L
Linus Torvalds 已提交
5763
}
5764
#else /* !CONFIG_SCHED_DEBUG */
5765 5766

# define sched_debug_enabled 0
5767
# define sched_domain_debug(sd, cpu) do { } while (0)
5768 5769 5770 5771
static inline bool sched_debug(void)
{
	return false;
}
5772
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
5773

5774
static int sd_degenerate(struct sched_domain *sd)
5775
{
5776
	if (cpumask_weight(sched_domain_span(sd)) == 1)
5777 5778 5779 5780 5781 5782
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
5783
			 SD_BALANCE_EXEC |
5784
			 SD_SHARE_CPUCAPACITY |
5785
			 SD_ASYM_CPUCAPACITY |
5786 5787
			 SD_SHARE_PKG_RESOURCES |
			 SD_SHARE_POWERDOMAIN)) {
5788 5789 5790 5791 5792
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
5793
	if (sd->flags & (SD_WAKE_AFFINE))
5794 5795 5796 5797 5798
		return 0;

	return 1;
}

5799 5800
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
5801 5802 5803 5804 5805 5806
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

5807
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
5808 5809 5810 5811 5812 5813 5814
		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 |
5815
				SD_BALANCE_EXEC |
5816
				SD_ASYM_CPUCAPACITY |
5817
				SD_SHARE_CPUCAPACITY |
5818
				SD_SHARE_PKG_RESOURCES |
5819 5820
				SD_PREFER_SIBLING |
				SD_SHARE_POWERDOMAIN);
5821 5822
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
5823 5824 5825 5826 5827 5828 5829
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

5830
static void free_rootdomain(struct rcu_head *rcu)
5831
{
5832
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
5833

5834
	cpupri_cleanup(&rd->cpupri);
5835
	cpudl_cleanup(&rd->cpudl);
5836
	free_cpumask_var(rd->dlo_mask);
5837 5838 5839 5840 5841 5842
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
5843 5844
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
5845
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
5846 5847
	unsigned long flags;

5848
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5849 5850

	if (rq->rd) {
I
Ingo Molnar 已提交
5851
		old_rd = rq->rd;
G
Gregory Haskins 已提交
5852

5853
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
5854
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5855

5856
		cpumask_clear_cpu(rq->cpu, old_rd->span);
5857

I
Ingo Molnar 已提交
5858
		/*
5859
		 * If we dont want to free the old_rd yet then
I
Ingo Molnar 已提交
5860 5861 5862 5863 5864
		 * 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
5865 5866 5867 5868 5869
	}

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

5870
	cpumask_set_cpu(rq->cpu, rd->span);
5871
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
5872
		set_rq_online(rq);
G
Gregory Haskins 已提交
5873

5874
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
5875 5876

	if (old_rd)
5877
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
5878 5879
}

5880
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
5881 5882 5883
{
	memset(rd, 0, sizeof(*rd));

5884
	if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
5885
		goto out;
5886
	if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
5887
		goto free_span;
5888
	if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
5889
		goto free_online;
5890
	if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
5891
		goto free_dlo_mask;
5892

5893
	init_dl_bw(&rd->dl_bw);
5894 5895
	if (cpudl_init(&rd->cpudl) != 0)
		goto free_dlo_mask;
5896

5897
	if (cpupri_init(&rd->cpupri) != 0)
5898
		goto free_rto_mask;
5899
	return 0;
5900

5901 5902
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
5903 5904
free_dlo_mask:
	free_cpumask_var(rd->dlo_mask);
5905 5906 5907 5908
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
5909
out:
5910
	return -ENOMEM;
G
Gregory Haskins 已提交
5911 5912
}

5913 5914 5915 5916 5917 5918
/*
 * 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
5919 5920
static void init_defrootdomain(void)
{
5921
	init_rootdomain(&def_root_domain);
5922

G
Gregory Haskins 已提交
5923 5924 5925
	atomic_set(&def_root_domain.refcount, 1);
}

5926
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5927 5928 5929 5930 5931 5932 5933
{
	struct root_domain *rd;

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

5934
	if (init_rootdomain(rd) != 0) {
5935 5936 5937
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5938 5939 5940 5941

	return rd;
}

5942
static void free_sched_groups(struct sched_group *sg, int free_sgc)
5943 5944 5945 5946 5947 5948 5949 5950 5951 5952
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

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

5953 5954
		if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
			kfree(sg->sgc);
5955 5956 5957 5958 5959 5960

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

5961
static void destroy_sched_domain(struct sched_domain *sd)
5962
{
5963 5964 5965 5966 5967 5968 5969
	/*
	 * 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)) {
5970
		kfree(sd->groups->sgc);
5971
		kfree(sd->groups);
5972
	}
5973 5974
	if (sd->shared && atomic_dec_and_test(&sd->shared->ref))
		kfree(sd->shared);
5975 5976 5977
	kfree(sd);
}

5978
static void destroy_sched_domains_rcu(struct rcu_head *rcu)
5979
{
5980 5981 5982 5983 5984 5985 5986
	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;
	}
5987 5988
}

5989
static void destroy_sched_domains(struct sched_domain *sd)
5990
{
5991 5992
	if (sd)
		call_rcu(&sd->rcu, destroy_sched_domains_rcu);
5993 5994
}

5995 5996 5997 5998 5999 6000 6001
/*
 * 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
6002
 * two cpus are in the same cache domain, see cpus_share_cache().
6003 6004
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
6005
DEFINE_PER_CPU(int, sd_llc_size);
6006
DEFINE_PER_CPU(int, sd_llc_id);
6007
DEFINE_PER_CPU(struct sched_domain_shared *, sd_llc_shared);
6008
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
6009
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
6010 6011 6012

static void update_top_cache_domain(int cpu)
{
6013
	struct sched_domain_shared *sds = NULL;
6014 6015
	struct sched_domain *sd;
	int id = cpu;
6016
	int size = 1;
6017 6018

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
6019
	if (sd) {
6020
		id = cpumask_first(sched_domain_span(sd));
6021
		size = cpumask_weight(sched_domain_span(sd));
6022
		sds = sd->shared;
6023
	}
6024 6025

	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
6026
	per_cpu(sd_llc_size, cpu) = size;
6027
	per_cpu(sd_llc_id, cpu) = id;
6028
	rcu_assign_pointer(per_cpu(sd_llc_shared, cpu), sds);
6029 6030 6031

	sd = lowest_flag_domain(cpu, SD_NUMA);
	rcu_assign_pointer(per_cpu(sd_numa, cpu), sd);
6032 6033 6034

	sd = highest_flag_domain(cpu, SD_ASYM_PACKING);
	rcu_assign_pointer(per_cpu(sd_asym, cpu), sd);
6035 6036
}

L
Linus Torvalds 已提交
6037
/*
I
Ingo Molnar 已提交
6038
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
6039 6040
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
6041 6042
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
6043
{
6044
	struct rq *rq = cpu_rq(cpu);
6045 6046 6047
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
6048
	for (tmp = sd; tmp; ) {
6049 6050 6051
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
6052

6053
		if (sd_parent_degenerate(tmp, parent)) {
6054
			tmp->parent = parent->parent;
6055 6056
			if (parent->parent)
				parent->parent->child = tmp;
6057 6058 6059 6060 6061 6062 6063
			/*
			 * 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;
6064
			destroy_sched_domain(parent);
6065 6066
		} else
			tmp = tmp->parent;
6067 6068
	}

6069
	if (sd && sd_degenerate(sd)) {
6070
		tmp = sd;
6071
		sd = sd->parent;
6072
		destroy_sched_domain(tmp);
6073 6074 6075
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
6076

6077
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
6078

G
Gregory Haskins 已提交
6079
	rq_attach_root(rq, rd);
6080
	tmp = rq->sd;
N
Nick Piggin 已提交
6081
	rcu_assign_pointer(rq->sd, sd);
6082
	destroy_sched_domains(tmp);
6083 6084

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
6085 6086 6087 6088 6089
}

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
6090 6091
	int ret;

R
Rusty Russell 已提交
6092
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
6093 6094 6095 6096 6097
	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
6098 6099
	return 1;
}
I
Ingo Molnar 已提交
6100
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
6101

6102
struct s_data {
6103
	struct sched_domain ** __percpu sd;
6104 6105 6106
	struct root_domain	*rd;
};

6107 6108
enum s_alloc {
	sa_rootdomain,
6109
	sa_sd,
6110
	sa_sd_storage,
6111 6112 6113
	sa_none,
};

P
Peter Zijlstra 已提交
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 6146 6147 6148 6149 6150 6151
/*
 * 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));
}

6152 6153 6154 6155 6156 6157 6158
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;
6159
	struct sched_domain *sibling;
6160 6161 6162 6163 6164 6165 6166 6167 6168 6169
	int i;

	cpumask_clear(covered);

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

		if (cpumask_test_cpu(i, covered))
			continue;

6170
		sibling = *per_cpu_ptr(sdd->sd, i);
P
Peter Zijlstra 已提交
6171 6172

		/* See the comment near build_group_mask(). */
6173
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
P
Peter Zijlstra 已提交
6174 6175
			continue;

6176
		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
6177
				GFP_KERNEL, cpu_to_node(cpu));
6178 6179 6180 6181 6182

		if (!sg)
			goto fail;

		sg_span = sched_group_cpus(sg);
6183 6184 6185
		if (sibling->child)
			cpumask_copy(sg_span, sched_domain_span(sibling->child));
		else
6186 6187 6188 6189
			cpumask_set_cpu(i, sg_span);

		cpumask_or(covered, covered, sg_span);

6190 6191
		sg->sgc = *per_cpu_ptr(sdd->sgc, i);
		if (atomic_inc_return(&sg->sgc->ref) == 1)
P
Peter Zijlstra 已提交
6192 6193
			build_group_mask(sd, sg);

6194
		/*
6195
		 * Initialize sgc->capacity such that even if we mess up the
6196 6197 6198
		 * domains and no possible iteration will get us here, we won't
		 * die on a /0 trap.
		 */
6199
		sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span);
6200
		sg->sgc->min_capacity = SCHED_CAPACITY_SCALE;
6201

P
Peter Zijlstra 已提交
6202 6203 6204 6205 6206
		/*
		 * 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
6207
		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
P
Peter Zijlstra 已提交
6208
		    group_balance_cpu(sg) == cpu)
6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227
			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;
}

6228
static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
L
Linus Torvalds 已提交
6229
{
6230 6231
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	struct sched_domain *child = sd->child;
L
Linus Torvalds 已提交
6232

6233 6234
	if (child)
		cpu = cpumask_first(sched_domain_span(child));
6235

6236
	if (sg) {
6237
		*sg = *per_cpu_ptr(sdd->sg, cpu);
6238 6239
		(*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu);
		atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */
6240
	}
6241 6242

	return cpu;
6243 6244
}

6245
/*
6246 6247
 * 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,
6248
 * and ->cpu_capacity to 0.
6249 6250
 *
 * Assumes the sched_domain tree is fully constructed
6251
 */
6252 6253
static int
build_sched_groups(struct sched_domain *sd, int cpu)
L
Linus Torvalds 已提交
6254
{
6255 6256 6257
	struct sched_group *first = NULL, *last = NULL;
	struct sd_data *sdd = sd->private;
	const struct cpumask *span = sched_domain_span(sd);
6258
	struct cpumask *covered;
6259
	int i;
6260

6261 6262 6263
	get_group(cpu, sdd, &sd->groups);
	atomic_inc(&sd->groups->ref);

6264
	if (cpu != cpumask_first(span))
6265 6266
		return 0;

6267 6268 6269
	lockdep_assert_held(&sched_domains_mutex);
	covered = sched_domains_tmpmask;

6270
	cpumask_clear(covered);
6271

6272 6273
	for_each_cpu(i, span) {
		struct sched_group *sg;
6274
		int group, j;
6275

6276 6277
		if (cpumask_test_cpu(i, covered))
			continue;
6278

6279
		group = get_group(i, sdd, &sg);
P
Peter Zijlstra 已提交
6280
		cpumask_setall(sched_group_mask(sg));
6281

6282 6283 6284
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
				continue;
6285

6286 6287 6288
			cpumask_set_cpu(j, covered);
			cpumask_set_cpu(j, sched_group_cpus(sg));
		}
6289

6290 6291 6292 6293 6294 6295 6296
		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
	}
	last->next = first;
6297 6298

	return 0;
6299
}
6300

6301
/*
6302
 * Initialize sched groups cpu_capacity.
6303
 *
6304
 * cpu_capacity indicates the capacity of sched group, which is used while
6305
 * distributing the load between different sched groups in a sched domain.
6306 6307 6308 6309
 * 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.
6310
 */
6311
static void init_sched_groups_capacity(int cpu, struct sched_domain *sd)
6312
{
6313
	struct sched_group *sg = sd->groups;
6314

6315
	WARN_ON(!sg);
6316 6317

	do {
T
Tim Chen 已提交
6318 6319
		int cpu, max_cpu = -1;

6320
		sg->group_weight = cpumask_weight(sched_group_cpus(sg));
T
Tim Chen 已提交
6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333

		if (!(sd->flags & SD_ASYM_PACKING))
			goto next;

		for_each_cpu(cpu, sched_group_cpus(sg)) {
			if (max_cpu < 0)
				max_cpu = cpu;
			else if (sched_asym_prefer(cpu, max_cpu))
				max_cpu = cpu;
		}
		sg->asym_prefer_cpu = max_cpu;

next:
6334 6335
		sg = sg->next;
	} while (sg != sd->groups);
6336

P
Peter Zijlstra 已提交
6337
	if (cpu != group_balance_cpu(sg))
6338
		return;
6339

6340
	update_group_capacity(sd, cpu);
6341 6342
}

6343 6344 6345 6346 6347
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

6348
static int default_relax_domain_level = -1;
6349
int sched_domain_level_max;
6350 6351 6352

static int __init setup_relax_domain_level(char *str)
{
6353 6354
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
6355

6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370 6371 6372 6373
	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 */
6374
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6375 6376
	} else {
		/* turn on idle balance on this domain */
6377
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6378 6379 6380
	}
}

6381 6382 6383
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

6384 6385 6386 6387 6388
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
6389 6390
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
6391 6392
	case sa_sd:
		free_percpu(d->sd); /* fall through */
6393
	case sa_sd_storage:
6394
		__sdt_free(cpu_map); /* fall through */
6395 6396 6397 6398
	case sa_none:
		break;
	}
}
6399

6400 6401 6402
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
6403 6404
	memset(d, 0, sizeof(*d));

6405 6406
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
6407 6408 6409
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
6410
	d->rd = alloc_rootdomain();
6411
	if (!d->rd)
6412
		return sa_sd;
6413 6414
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
6415

6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427
/*
 * 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;

6428 6429 6430
	if (atomic_read(&(*per_cpu_ptr(sdd->sds, cpu))->ref))
		*per_cpu_ptr(sdd->sds, cpu) = NULL;

6431
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
6432
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
6433

6434 6435
	if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
		*per_cpu_ptr(sdd->sgc, cpu) = NULL;
6436 6437
}

6438 6439
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
6440
enum numa_topology_type sched_numa_topology_type;
6441
static int *sched_domains_numa_distance;
6442
int sched_max_numa_distance;
6443 6444
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
6445
#endif
6446

6447 6448 6449
/*
 * SD_flags allowed in topology descriptions.
 *
6450 6451 6452
 * These flags are purely descriptive of the topology and do not prescribe
 * behaviour. Behaviour is artificial and mapped in the below sd_init()
 * function:
6453
 *
6454 6455 6456 6457
 *   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
6458
 *   SD_ASYM_CPUCAPACITY    - describes mixed capacity topologies
6459 6460 6461
 *
 * Odd one out, which beside describing the topology has a quirk also
 * prescribes the desired behaviour that goes along with it:
6462
 *
6463
 *   SD_ASYM_PACKING        - describes SMT quirks
6464 6465
 */
#define TOPOLOGY_SD_FLAGS		\
6466
	(SD_SHARE_CPUCAPACITY |		\
6467 6468
	 SD_SHARE_PKG_RESOURCES |	\
	 SD_NUMA |			\
6469
	 SD_ASYM_PACKING |		\
6470
	 SD_ASYM_CPUCAPACITY |		\
6471
	 SD_SHARE_POWERDOMAIN)
6472 6473

static struct sched_domain *
6474
sd_init(struct sched_domain_topology_level *tl,
6475
	const struct cpumask *cpu_map,
6476
	struct sched_domain *child, int cpu)
6477
{
6478 6479 6480
	struct sd_data *sdd = &tl->data;
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	int sd_id, sd_weight, sd_flags = 0;
6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495

#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;
6496 6497 6498 6499 6500

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
6501
		.imbalance_pct		= 125,
6502 6503 6504 6505

		.cache_nice_tries	= 0,
		.busy_idx		= 0,
		.idle_idx		= 0,
6506 6507 6508 6509 6510 6511
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
6512 6513
					| 1*SD_BALANCE_EXEC
					| 1*SD_BALANCE_FORK
6514
					| 0*SD_BALANCE_WAKE
6515
					| 1*SD_WAKE_AFFINE
6516
					| 0*SD_SHARE_CPUCAPACITY
6517
					| 0*SD_SHARE_PKG_RESOURCES
6518
					| 0*SD_SERIALIZE
6519
					| 0*SD_PREFER_SIBLING
6520 6521
					| 0*SD_NUMA
					| sd_flags
6522
					,
6523

6524 6525
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
6526
		.smt_gain		= 0,
6527 6528
		.max_newidle_lb_cost	= 0,
		.next_decay_max_lb_cost	= jiffies,
6529
		.child			= child,
6530 6531 6532
#ifdef CONFIG_SCHED_DEBUG
		.name			= tl->name,
#endif
6533 6534
	};

6535 6536 6537
	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
	sd_id = cpumask_first(sched_domain_span(sd));

6538
	/*
6539
	 * Convert topological properties into behaviour.
6540
	 */
6541

6542 6543 6544 6545 6546 6547 6548
	if (sd->flags & SD_ASYM_CPUCAPACITY) {
		struct sched_domain *t = sd;

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

6549
	if (sd->flags & SD_SHARE_CPUCAPACITY) {
6550
		sd->flags |= SD_PREFER_SIBLING;
6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579
		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;
	}

6580 6581 6582 6583 6584 6585 6586
	/*
	 * 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);
6587
		atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
6588 6589 6590
	}

	sd->private = sdd;
6591 6592 6593 6594

	return sd;
}

6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608
/*
 * 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, },
};

6609 6610
static struct sched_domain_topology_level *sched_domain_topology =
	default_topology;
6611 6612 6613 6614 6615 6616

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

void set_sched_topology(struct sched_domain_topology_level *tl)
{
6617 6618 6619
	if (WARN_ON_ONCE(sched_smp_initialized))
		return;

6620 6621 6622 6623 6624
	sched_domain_topology = tl;
}

#ifdef CONFIG_NUMA

6625 6626 6627 6628 6629
static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650
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");
}

6651
bool find_numa_distance(int distance)
6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665
{
	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;
}

6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690
/*
 * 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;

6691
	if (sched_domains_numa_levels <= 1) {
6692
		sched_numa_topology_type = NUMA_DIRECT;
6693 6694
		return;
	}
6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717

	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;
		}
	}
}

6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738
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++) {
6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762
			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;
6763
		}
6764 6765 6766 6767 6768 6769

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
6770
	}
6771 6772 6773 6774

	if (!level)
		return;

6775 6776 6777 6778
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
6779
	 * The sched_domains_numa_distance[] array includes the actual distance
6780 6781 6782
	 * numbers.
	 */

6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793
	/*
	 * 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;

6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808
	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++) {
6809
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
6810 6811 6812 6813 6814
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

6815
			for_each_node(k) {
6816
				if (node_distance(j, k) > sched_domains_numa_distance[i])
6817 6818 6819 6820 6821 6822 6823
					continue;

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

6824 6825 6826
	/* Compute default topology size */
	for (i = 0; sched_domain_topology[i].mask; i++);

6827
	tl = kzalloc((i + level + 1) *
6828 6829 6830 6831 6832 6833 6834
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
6835 6836
	for (i = 0; sched_domain_topology[i].mask; i++)
		tl[i] = sched_domain_topology[i];
6837 6838 6839 6840 6841 6842 6843

	/*
	 * .. 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,
6844
			.sd_flags = cpu_numa_flags,
6845 6846
			.flags = SDTL_OVERLAP,
			.numa_level = j,
6847
			SD_INIT_NAME(NUMA)
6848 6849 6850 6851
		};
	}

	sched_domain_topology = tl;
6852 6853

	sched_domains_numa_levels = level;
6854
	sched_max_numa_distance = sched_domains_numa_distance[level - 1];
6855 6856

	init_numa_topology_type();
6857
}
6858

6859
static void sched_domains_numa_masks_set(unsigned int cpu)
6860 6861
{
	int node = cpu_to_node(cpu);
6862
	int i, j;
6863 6864 6865 6866 6867 6868 6869 6870 6871

	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]);
		}
	}
}

6872
static void sched_domains_numa_masks_clear(unsigned int cpu)
6873 6874
{
	int i, j;
6875

6876 6877 6878 6879 6880 6881
	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]);
	}
}

6882
#else
6883 6884 6885
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) { }
6886 6887
#endif /* CONFIG_NUMA */

6888 6889 6890 6891 6892
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6893
	for_each_sd_topology(tl) {
6894 6895 6896 6897 6898 6899
		struct sd_data *sdd = &tl->data;

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

6900 6901 6902 6903
		sdd->sds = alloc_percpu(struct sched_domain_shared *);
		if (!sdd->sds)
			return -ENOMEM;

6904 6905 6906 6907
		sdd->sg = alloc_percpu(struct sched_group *);
		if (!sdd->sg)
			return -ENOMEM;

6908 6909
		sdd->sgc = alloc_percpu(struct sched_group_capacity *);
		if (!sdd->sgc)
6910 6911
			return -ENOMEM;

6912 6913
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
6914
			struct sched_domain_shared *sds;
6915
			struct sched_group *sg;
6916
			struct sched_group_capacity *sgc;
6917

P
Peter Zijlstra 已提交
6918
			sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6919 6920 6921 6922 6923 6924
					GFP_KERNEL, cpu_to_node(j));
			if (!sd)
				return -ENOMEM;

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

6925 6926 6927 6928 6929 6930 6931
			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;

6932 6933 6934 6935 6936
			sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
					GFP_KERNEL, cpu_to_node(j));
			if (!sg)
				return -ENOMEM;

6937 6938
			sg->next = sg;

6939
			*per_cpu_ptr(sdd->sg, j) = sg;
6940

6941
			sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
6942
					GFP_KERNEL, cpu_to_node(j));
6943
			if (!sgc)
6944 6945
				return -ENOMEM;

6946
			*per_cpu_ptr(sdd->sgc, j) = sgc;
6947 6948 6949 6950 6951 6952 6953 6954 6955 6956 6957
		}
	}

	return 0;
}

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

6958
	for_each_sd_topology(tl) {
6959 6960 6961
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
6962 6963 6964 6965 6966 6967 6968 6969 6970
			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));
			}

6971 6972
			if (sdd->sds)
				kfree(*per_cpu_ptr(sdd->sds, j));
6973 6974
			if (sdd->sg)
				kfree(*per_cpu_ptr(sdd->sg, j));
6975 6976
			if (sdd->sgc)
				kfree(*per_cpu_ptr(sdd->sgc, j));
6977 6978
		}
		free_percpu(sdd->sd);
6979
		sdd->sd = NULL;
6980 6981
		free_percpu(sdd->sds);
		sdd->sds = NULL;
6982
		free_percpu(sdd->sg);
6983
		sdd->sg = NULL;
6984 6985
		free_percpu(sdd->sgc);
		sdd->sgc = NULL;
6986 6987 6988
	}
}

6989
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6990 6991
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
6992
{
6993
	struct sched_domain *sd = sd_init(tl, cpu_map, child, cpu);
6994

6995 6996 6997
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
6998
		child->parent = sd;
P
Peter Zijlstra 已提交
6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011 7012

		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));
		}

7013
	}
7014
	set_domain_attribute(sd, attr);
7015 7016 7017 7018

	return sd;
}

7019 7020 7021 7022
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
7023 7024
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
7025
{
7026
	enum s_alloc alloc_state;
7027
	struct sched_domain *sd;
7028
	struct s_data d;
7029
	struct rq *rq = NULL;
7030
	int i, ret = -ENOMEM;
7031

7032 7033 7034
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
7035

7036
	/* Set up domains for cpus specified by the cpu_map. */
7037
	for_each_cpu(i, cpu_map) {
7038 7039
		struct sched_domain_topology_level *tl;

7040
		sd = NULL;
7041
		for_each_sd_topology(tl) {
7042
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
7043 7044
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
7045 7046
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
7047 7048
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
7049
		}
7050 7051 7052 7053 7054 7055
	}

	/* 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));
7056 7057 7058 7059 7060 7061 7062
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
7063
		}
7064
	}
7065

7066
	/* Calculate CPU capacity for physical packages and nodes */
7067 7068 7069
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
7070

7071 7072
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
7073
			init_sched_groups_capacity(i, sd);
7074
		}
7075
	}
7076

L
Linus Torvalds 已提交
7077
	/* Attach the domains */
7078
	rcu_read_lock();
7079
	for_each_cpu(i, cpu_map) {
7080
		rq = cpu_rq(i);
7081
		sd = *per_cpu_ptr(d.sd, i);
7082 7083 7084 7085 7086

		/* 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);

7087
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
7088
	}
7089
	rcu_read_unlock();
7090

7091
	if (rq && sched_debug_enabled) {
7092 7093 7094 7095
		pr_info("span: %*pbl (max cpu_capacity = %lu)\n",
			cpumask_pr_args(cpu_map), rq->rd->max_cpu_capacity);
	}

7096
	ret = 0;
7097
error:
7098
	__free_domain_allocs(&d, alloc_state, cpu_map);
7099
	return ret;
L
Linus Torvalds 已提交
7100
}
P
Paul Jackson 已提交
7101

7102
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
7103
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
7104 7105
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
7106 7107 7108

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
7109 7110
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
7111
 */
7112
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
7113

7114 7115 7116 7117 7118
/*
 * 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.
 */
7119
int __weak arch_update_cpu_topology(void)
7120
{
7121
	return 0;
7122 7123
}

7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148
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);
}

7149
/*
I
Ingo Molnar 已提交
7150
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
7151 7152
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
7153
 */
7154
static int init_sched_domains(const struct cpumask *cpu_map)
7155
{
7156 7157
	int err;

7158
	arch_update_cpu_topology();
P
Paul Jackson 已提交
7159
	ndoms_cur = 1;
7160
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
7161
	if (!doms_cur)
7162 7163
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
7164
	err = build_sched_domains(doms_cur[0], NULL);
7165
	register_sched_domain_sysctl();
7166 7167

	return err;
7168 7169 7170 7171 7172 7173
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
7174
static void detach_destroy_domains(const struct cpumask *cpu_map)
7175 7176 7177
{
	int i;

7178
	rcu_read_lock();
7179
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
7180
		cpu_attach_domain(NULL, &def_root_domain, i);
7181
	rcu_read_unlock();
7182 7183
}

7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199
/* 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
7200 7201
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
7202
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
7203 7204 7205
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
7206
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
7207 7208 7209
 * 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
7210 7211 7212
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
7213 7214 7215 7216 7217 7218
 * 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
7219
 *
7220
 * If doms_new == NULL it will be replaced with cpu_online_mask.
7221 7222
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
7223
 *
P
Paul Jackson 已提交
7224 7225
 * Call with hotplug lock held
 */
7226
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
7227
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
7228
{
7229
	int i, j, n;
7230
	int new_topology;
P
Paul Jackson 已提交
7231

7232
	mutex_lock(&sched_domains_mutex);
7233

7234 7235 7236
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

7237 7238 7239
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

7240
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
7241 7242 7243

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
7244
		for (j = 0; j < n && !new_topology; j++) {
7245
			if (cpumask_equal(doms_cur[i], doms_new[j])
7246
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
7247 7248 7249
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
7250
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
7251 7252 7253 7254
match1:
		;
	}

7255
	n = ndoms_cur;
7256
	if (doms_new == NULL) {
7257
		n = 0;
7258
		doms_new = &fallback_doms;
7259
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
7260
		WARN_ON_ONCE(dattr_new);
7261 7262
	}

P
Paul Jackson 已提交
7263 7264
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
7265
		for (j = 0; j < n && !new_topology; j++) {
7266
			if (cpumask_equal(doms_new[i], doms_cur[j])
7267
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
7268 7269 7270
				goto match2;
		}
		/* no match - add a new doms_new */
7271
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
7272 7273 7274 7275 7276
match2:
		;
	}

	/* Remember the new sched domains */
7277 7278
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
7279
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
7280
	doms_cur = doms_new;
7281
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
7282
	ndoms_cur = ndoms_new;
7283 7284

	register_sched_domain_sysctl();
7285

7286
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
7287 7288
}

7289 7290
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
7291
/*
7292 7293 7294
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
7295 7296 7297
 *
 * 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
7298
 */
7299
static void cpuset_cpu_active(void)
7300
{
7301
	if (cpuhp_tasks_frozen) {
7302 7303 7304 7305 7306 7307 7308 7309 7310
		/*
		 * 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);
7311
			return;
7312 7313 7314 7315 7316 7317
		}
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
7318
	}
7319
	cpuset_update_active_cpus(true);
7320
}
7321

7322
static int cpuset_cpu_inactive(unsigned int cpu)
7323
{
7324 7325
	unsigned long flags;
	struct dl_bw *dl_b;
7326 7327
	bool overflow;
	int cpus;
7328

7329
	if (!cpuhp_tasks_frozen) {
7330 7331
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
7332

7333 7334 7335 7336
		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);
7337

7338
		rcu_read_unlock_sched();
7339

7340
		if (overflow)
7341
			return -EBUSY;
7342
		cpuset_update_active_cpus(false);
7343
	} else {
7344 7345
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
7346
	}
7347
	return 0;
7348 7349
}

7350
int sched_cpu_activate(unsigned int cpu)
7351
{
7352 7353 7354
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

7355
	set_cpu_active(cpu, true);
7356

7357
	if (sched_smp_initialized) {
7358
		sched_domains_numa_masks_set(cpu);
7359
		cpuset_cpu_active();
7360
	}
7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379

	/*
	 * 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();

7380
	return 0;
7381 7382
}

7383
int sched_cpu_deactivate(unsigned int cpu)
7384 7385 7386
{
	int ret;

7387
	set_cpu_active(cpu, false);
7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399 7400 7401
	/*
	 * 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();
7402 7403 7404 7405 7406 7407 7408 7409

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
7410
	}
7411 7412
	sched_domains_numa_masks_clear(cpu);
	return 0;
7413 7414
}

7415 7416 7417 7418 7419 7420 7421 7422
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();
}

7423 7424 7425
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
7426
	sched_rq_cpu_starting(cpu);
7427
	return 0;
7428 7429
}

7430 7431 7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447
#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();
7448
	nohz_balance_exit_idle(cpu);
7449
	hrtick_clear(rq);
7450 7451 7452 7453
	return 0;
}
#endif

P
Peter Zijlstra 已提交
7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469
#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
7470 7471
void __init sched_init_smp(void)
{
7472 7473 7474
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7475
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
7476

7477 7478
	sched_init_numa();

7479 7480 7481 7482 7483
	/*
	 * 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.
	 */
7484
	mutex_lock(&sched_domains_mutex);
7485
	init_sched_domains(cpu_active_mask);
7486 7487 7488
	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);
7489
	mutex_unlock(&sched_domains_mutex);
7490

7491
	/* Move init over to a non-isolated CPU */
7492
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
7493
		BUG();
I
Ingo Molnar 已提交
7494
	sched_init_granularity();
7495
	free_cpumask_var(non_isolated_cpus);
7496

7497
	init_sched_rt_class();
7498
	init_sched_dl_class();
P
Peter Zijlstra 已提交
7499 7500 7501

	sched_init_smt();

7502
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
7503
}
7504 7505 7506

static int __init migration_init(void)
{
7507
	sched_rq_cpu_starting(smp_processor_id());
7508
	return 0;
L
Linus Torvalds 已提交
7509
}
7510 7511
early_initcall(migration_init);

L
Linus Torvalds 已提交
7512 7513 7514
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
7515
	sched_init_granularity();
L
Linus Torvalds 已提交
7516 7517 7518 7519 7520 7521 7522 7523 7524 7525
}
#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);
}

7526
#ifdef CONFIG_CGROUP_SCHED
7527 7528 7529 7530
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
7531
struct task_group root_task_group;
7532
LIST_HEAD(task_groups);
7533 7534 7535

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
7536
#endif
P
Peter Zijlstra 已提交
7537

7538
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
7539
DECLARE_PER_CPU(cpumask_var_t, select_idle_mask);
P
Peter Zijlstra 已提交
7540

7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553
#define WAIT_TABLE_BITS 8
#define WAIT_TABLE_SIZE (1 << WAIT_TABLE_BITS)
static wait_queue_head_t bit_wait_table[WAIT_TABLE_SIZE] __cacheline_aligned;

wait_queue_head_t *bit_waitqueue(void *word, int bit)
{
	const int shift = BITS_PER_LONG == 32 ? 5 : 6;
	unsigned long val = (unsigned long)word << shift | bit;

	return bit_wait_table + hash_long(val, WAIT_TABLE_BITS);
}
EXPORT_SYMBOL(bit_waitqueue);

L
Linus Torvalds 已提交
7554 7555
void __init sched_init(void)
{
I
Ingo Molnar 已提交
7556
	int i, j;
7557 7558
	unsigned long alloc_size = 0, ptr;

7559 7560 7561
	for (i = 0; i < WAIT_TABLE_SIZE; i++)
		init_waitqueue_head(bit_wait_table + i);

7562 7563 7564 7565 7566 7567 7568
#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) {
7569
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
7570 7571

#ifdef CONFIG_FAIR_GROUP_SCHED
7572
		root_task_group.se = (struct sched_entity **)ptr;
7573 7574
		ptr += nr_cpu_ids * sizeof(void **);

7575
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
7576
		ptr += nr_cpu_ids * sizeof(void **);
7577

7578
#endif /* CONFIG_FAIR_GROUP_SCHED */
7579
#ifdef CONFIG_RT_GROUP_SCHED
7580
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
7581 7582
		ptr += nr_cpu_ids * sizeof(void **);

7583
		root_task_group.rt_rq = (struct rt_rq **)ptr;
7584 7585
		ptr += nr_cpu_ids * sizeof(void **);

7586
#endif /* CONFIG_RT_GROUP_SCHED */
7587
	}
7588
#ifdef CONFIG_CPUMASK_OFFSTACK
7589 7590 7591
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7592 7593
		per_cpu(select_idle_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7594
	}
7595
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
7596

7597 7598 7599
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth,
7600
			global_rt_period(), global_rt_runtime());
7601

G
Gregory Haskins 已提交
7602 7603 7604 7605
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

7606
#ifdef CONFIG_RT_GROUP_SCHED
7607
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
7608
			global_rt_period(), global_rt_runtime());
7609
#endif /* CONFIG_RT_GROUP_SCHED */
7610

D
Dhaval Giani 已提交
7611
#ifdef CONFIG_CGROUP_SCHED
7612 7613
	task_group_cache = KMEM_CACHE(task_group, 0);

7614 7615
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
7616
	INIT_LIST_HEAD(&root_task_group.siblings);
7617
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
7618
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
7619

7620
	for_each_possible_cpu(i) {
7621
		struct rq *rq;
L
Linus Torvalds 已提交
7622 7623

		rq = cpu_rq(i);
7624
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
7625
		rq->nr_running = 0;
7626 7627
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
7628
		init_cfs_rq(&rq->cfs);
7629 7630
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
7631
#ifdef CONFIG_FAIR_GROUP_SCHED
7632
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
7633
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
7634
		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
D
Dhaval Giani 已提交
7635
		/*
7636
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
7637 7638 7639 7640
		 *
		 * 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
7641
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
7642 7643 7644
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
7645
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
7646 7647 7648
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
7649
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
7650
		 *
7651 7652
		 * 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
7653
		 */
7654
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
7655
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
7656 7657 7658
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
7659
#ifdef CONFIG_RT_GROUP_SCHED
7660
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
7661
#endif
L
Linus Torvalds 已提交
7662

I
Ingo Molnar 已提交
7663 7664
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
7665

L
Linus Torvalds 已提交
7666
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
7667
		rq->sd = NULL;
G
Gregory Haskins 已提交
7668
		rq->rd = NULL;
7669
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
7670
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
7671
		rq->active_balance = 0;
I
Ingo Molnar 已提交
7672
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
7673
		rq->push_cpu = 0;
7674
		rq->cpu = i;
7675
		rq->online = 0;
7676 7677
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
7678
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
7679 7680 7681

		INIT_LIST_HEAD(&rq->cfs_tasks);

7682
		rq_attach_root(rq, &def_root_domain);
7683
#ifdef CONFIG_NO_HZ_COMMON
7684
		rq->last_load_update_tick = jiffies;
7685
		rq->nohz_flags = 0;
7686
#endif
7687 7688 7689
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
7690
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
7691
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
7692 7693 7694
		atomic_set(&rq->nr_iowait, 0);
	}

7695
	set_load_weight(&init_task);
7696

L
Linus Torvalds 已提交
7697 7698 7699 7700 7701 7702 7703 7704 7705 7706 7707 7708 7709
	/*
	 * 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());
7710 7711 7712

	calc_load_update = jiffies + LOAD_FREQ;

7713
#ifdef CONFIG_SMP
7714
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
7715 7716 7717
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
7718
	idle_thread_set_boot_cpu();
7719
	set_cpu_rq_start_time(smp_processor_id());
7720 7721
#endif
	init_sched_fair_class();
7722

7723 7724
	init_schedstats();

7725
	scheduler_running = 1;
L
Linus Torvalds 已提交
7726 7727
}

7728
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
7729 7730
static inline int preempt_count_equals(int preempt_offset)
{
7731
	int nested = preempt_count() + rcu_preempt_depth();
7732

A
Arnd Bergmann 已提交
7733
	return (nested == preempt_offset);
7734 7735
}

7736
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7737
{
P
Peter Zijlstra 已提交
7738 7739 7740 7741 7742
	/*
	 * 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.
	 */
7743
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
7744 7745 7746 7747
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
7748
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
7749

7750 7751 7752 7753 7754
	___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
7755 7756
{
	static unsigned long prev_jiffy;	/* ratelimiting */
7757
	unsigned long preempt_disable_ip;
L
Linus Torvalds 已提交
7758

7759
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
7760 7761
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
7762
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
7763 7764 7765 7766 7767
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

7768 7769 7770
	/* Save this before calling printk(), since that will clobber it */
	preempt_disable_ip = get_preempt_disable_ip(current);

P
Peter Zijlstra 已提交
7771 7772 7773 7774 7775 7776 7777
	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
7778

7779 7780 7781
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
7782 7783 7784
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
7785 7786
	if (IS_ENABLED(CONFIG_DEBUG_PREEMPT)
	    && !preempt_count_equals(preempt_offset)) {
7787
		pr_err("Preemption disabled at:");
7788
		print_ip_sym(preempt_disable_ip);
7789 7790
		pr_cont("\n");
	}
I
Ingo Molnar 已提交
7791
	dump_stack();
7792
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
L
Linus Torvalds 已提交
7793
}
7794
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
7795 7796 7797
#endif

#ifdef CONFIG_MAGIC_SYSRQ
7798
void normalize_rt_tasks(void)
7799
{
7800
	struct task_struct *g, *p;
7801 7802 7803
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
7804

7805
	read_lock(&tasklist_lock);
7806
	for_each_process_thread(g, p) {
7807 7808 7809
		/*
		 * Only normalize user tasks:
		 */
7810
		if (p->flags & PF_KTHREAD)
7811 7812
			continue;

7813 7814 7815 7816
		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
7817

7818
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
7819 7820 7821 7822
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
7823
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
7824
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
7825
			continue;
I
Ingo Molnar 已提交
7826
		}
L
Linus Torvalds 已提交
7827

7828
		__sched_setscheduler(p, &attr, false, false);
7829
	}
7830
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
7831 7832 7833
}

#endif /* CONFIG_MAGIC_SYSRQ */
7834

7835
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
7836
/*
7837
 * These functions are only useful for the IA64 MCA handling, or kdb.
7838 7839 7840 7841 7842 7843 7844 7845 7846 7847 7848 7849 7850
 *
 * 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!
7851 7852
 *
 * Return: The current task for @cpu.
7853
 */
7854
struct task_struct *curr_task(int cpu)
7855 7856 7857 7858
{
	return cpu_curr(cpu);
}

7859 7860 7861
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
7862 7863 7864 7865 7866 7867
/**
 * 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
7868 7869
 * 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
7870 7871 7872 7873 7874 7875 7876
 * 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!
 */
7877
void ia64_set_curr_task(int cpu, struct task_struct *p)
7878 7879 7880 7881 7882
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
7883

D
Dhaval Giani 已提交
7884
#ifdef CONFIG_CGROUP_SCHED
7885 7886 7887
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

7888
static void sched_free_group(struct task_group *tg)
7889 7890 7891
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
7892
	autogroup_free(tg);
7893
	kmem_cache_free(task_group_cache, tg);
7894 7895 7896
}

/* allocate runqueue etc for a new task group */
7897
struct task_group *sched_create_group(struct task_group *parent)
7898 7899 7900
{
	struct task_group *tg;

7901
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
7902 7903 7904
	if (!tg)
		return ERR_PTR(-ENOMEM);

7905
	if (!alloc_fair_sched_group(tg, parent))
7906 7907
		goto err;

7908
	if (!alloc_rt_sched_group(tg, parent))
7909 7910
		goto err;

7911 7912 7913
	return tg;

err:
7914
	sched_free_group(tg);
7915 7916 7917 7918 7919 7920 7921
	return ERR_PTR(-ENOMEM);
}

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

7922
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7923
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
7924 7925 7926 7927 7928

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
7929
	list_add_rcu(&tg->siblings, &parent->children);
7930
	spin_unlock_irqrestore(&task_group_lock, flags);
7931 7932

	online_fair_sched_group(tg);
S
Srivatsa Vaddagiri 已提交
7933 7934
}

7935
/* rcu callback to free various structures associated with a task group */
7936
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
7937 7938
{
	/* now it should be safe to free those cfs_rqs */
7939
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
7940 7941
}

7942
void sched_destroy_group(struct task_group *tg)
7943 7944
{
	/* wait for possible concurrent references to cfs_rqs complete */
7945
	call_rcu(&tg->rcu, sched_free_group_rcu);
7946 7947 7948
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
7949
{
7950
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
7951

7952
	/* end participation in shares distribution */
7953
	unregister_fair_sched_group(tg);
7954 7955

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7956
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
7957
	list_del_rcu(&tg->siblings);
7958
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7959 7960
}

7961
static void sched_change_group(struct task_struct *tsk, int type)
S
Srivatsa Vaddagiri 已提交
7962
{
P
Peter Zijlstra 已提交
7963
	struct task_group *tg;
S
Srivatsa Vaddagiri 已提交
7964

7965 7966 7967 7968 7969 7970
	/*
	 * 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
7971 7972 7973 7974
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
7975
#ifdef CONFIG_FAIR_GROUP_SCHED
7976 7977
	if (tsk->sched_class->task_change_group)
		tsk->sched_class->task_change_group(tsk, type);
7978
	else
P
Peter Zijlstra 已提交
7979
#endif
7980
		set_task_rq(tsk, task_cpu(tsk));
7981 7982 7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006
}

/*
 * 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
8007

8008
	if (queued)
8009
		enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE);
8010
	if (unlikely(running))
8011
		set_curr_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
8012

8013
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
8014
}
D
Dhaval Giani 已提交
8015
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
8016

8017 8018 8019 8020 8021
#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
8022

P
Peter Zijlstra 已提交
8023 8024
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
8025
{
P
Peter Zijlstra 已提交
8026
	struct task_struct *g, *p;
8027

8028 8029 8030 8031 8032 8033
	/*
	 * Autogroups do not have RT tasks; see autogroup_create().
	 */
	if (task_group_is_autogroup(tg))
		return 0;

8034
	for_each_process_thread(g, p) {
8035
		if (rt_task(p) && task_group(p) == tg)
P
Peter Zijlstra 已提交
8036
			return 1;
8037
	}
8038

P
Peter Zijlstra 已提交
8039 8040
	return 0;
}
8041

P
Peter Zijlstra 已提交
8042 8043 8044 8045 8046
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
8047

8048
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
8049 8050 8051 8052 8053
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
8054

P
Peter Zijlstra 已提交
8055 8056
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
8057

P
Peter Zijlstra 已提交
8058 8059 8060
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
8061 8062
	}

8063 8064 8065 8066 8067
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
8068

8069 8070 8071
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
8072 8073
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
8074

P
Peter Zijlstra 已提交
8075
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
8076

8077 8078 8079 8080 8081
	/*
	 * 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
8082

8083 8084 8085
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
8086 8087 8088
	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
8089

P
Peter Zijlstra 已提交
8090 8091 8092 8093
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
8094

P
Peter Zijlstra 已提交
8095
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
8096
	}
P
Peter Zijlstra 已提交
8097

P
Peter Zijlstra 已提交
8098 8099 8100 8101
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
8102 8103
}

P
Peter Zijlstra 已提交
8104
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
8105
{
8106 8107
	int ret;

P
Peter Zijlstra 已提交
8108 8109 8110 8111 8112 8113
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

8114 8115 8116 8117 8118
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8119 8120
}

8121
static int tg_set_rt_bandwidth(struct task_group *tg,
8122
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
8123
{
P
Peter Zijlstra 已提交
8124
	int i, err = 0;
P
Peter Zijlstra 已提交
8125

8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136
	/*
	 * 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
8137
	mutex_lock(&rt_constraints_mutex);
8138
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
8139 8140
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
8141
		goto unlock;
P
Peter Zijlstra 已提交
8142

8143
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8144 8145
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
8146 8147 8148 8149

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

8150
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8151
		rt_rq->rt_runtime = rt_runtime;
8152
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8153
	}
8154
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
8155
unlock:
8156
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
8157 8158 8159
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
8160 8161
}

8162
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
8163 8164 8165 8166 8167 8168 8169 8170
{
	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;

8171
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8172 8173
}

8174
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
8175 8176 8177
{
	u64 rt_runtime_us;

8178
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
8179 8180
		return -1;

8181
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
8182 8183 8184
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
8185

8186
static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
8187 8188 8189
{
	u64 rt_runtime, rt_period;

8190
	rt_period = rt_period_us * NSEC_PER_USEC;
8191 8192
	rt_runtime = tg->rt_bandwidth.rt_runtime;

8193
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8194 8195
}

8196
static long sched_group_rt_period(struct task_group *tg)
8197 8198 8199 8200 8201 8202 8203
{
	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;
}
8204
#endif /* CONFIG_RT_GROUP_SCHED */
8205

8206
#ifdef CONFIG_RT_GROUP_SCHED
8207 8208 8209 8210 8211
static int sched_rt_global_constraints(void)
{
	int ret = 0;

	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
8212
	read_lock(&tasklist_lock);
8213
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
8214
	read_unlock(&tasklist_lock);
8215 8216 8217 8218
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
8219

8220
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8221 8222 8223 8224 8225 8226 8227 8228
{
	/* 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;
}

8229
#else /* !CONFIG_RT_GROUP_SCHED */
8230 8231
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
8232
	unsigned long flags;
8233
	int i;
8234

8235
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8236 8237 8238
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

8239
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8240
		rt_rq->rt_runtime = global_rt_runtime();
8241
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8242
	}
8243
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8244

8245
	return 0;
8246
}
8247
#endif /* CONFIG_RT_GROUP_SCHED */
8248

8249
static int sched_dl_global_validate(void)
8250
{
8251 8252
	u64 runtime = global_rt_runtime();
	u64 period = global_rt_period();
8253
	u64 new_bw = to_ratio(period, runtime);
8254
	struct dl_bw *dl_b;
8255
	int cpu, ret = 0;
8256
	unsigned long flags;
8257 8258 8259 8260 8261 8262 8263 8264 8265 8266

	/*
	 * 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!
	 */
8267
	for_each_possible_cpu(cpu) {
8268 8269
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8270

8271
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8272 8273
		if (new_bw < dl_b->total_bw)
			ret = -EBUSY;
8274
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8275

8276 8277
		rcu_read_unlock_sched();

8278 8279
		if (ret)
			break;
8280 8281
	}

8282
	return ret;
8283 8284
}

8285
static void sched_dl_do_global(void)
8286
{
8287
	u64 new_bw = -1;
8288
	struct dl_bw *dl_b;
8289
	int cpu;
8290
	unsigned long flags;
8291

8292 8293 8294 8295 8296 8297 8298 8299 8300 8301
	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) {
8302 8303
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8304

8305
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8306
		dl_b->bw = new_bw;
8307
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8308 8309

		rcu_read_unlock_sched();
8310
	}
8311 8312 8313 8314 8315 8316 8317
}

static int sched_rt_global_validate(void)
{
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

8318 8319
	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
8320 8321 8322 8323 8324 8325 8326 8327 8328
		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());
8329 8330
}

8331
int sched_rt_handler(struct ctl_table *table, int write,
8332
		void __user *buffer, size_t *lenp,
8333 8334 8335 8336
		loff_t *ppos)
{
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);
8337
	int ret;
8338 8339 8340 8341 8342

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

8343
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8344 8345

	if (!ret && write) {
8346 8347 8348 8349
		ret = sched_rt_global_validate();
		if (ret)
			goto undo;

8350
		ret = sched_dl_global_validate();
8351 8352 8353
		if (ret)
			goto undo;

8354
		ret = sched_rt_global_constraints();
8355 8356 8357 8358 8359 8360 8361 8362 8363 8364
		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;
8365 8366 8367 8368 8369
	}
	mutex_unlock(&mutex);

	return ret;
}
8370

8371
int sched_rr_handler(struct ctl_table *table, int write,
8372 8373 8374 8375 8376 8377 8378 8379
		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);
8380 8381
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
8382
	if (!ret && write) {
8383 8384
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
8385 8386 8387 8388 8389
	}
	mutex_unlock(&mutex);
	return ret;
}

8390
#ifdef CONFIG_CGROUP_SCHED
8391

8392
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
8393
{
8394
	return css ? container_of(css, struct task_group, css) : NULL;
8395 8396
}

8397 8398
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
8399
{
8400 8401
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
8402

8403
	if (!parent) {
8404
		/* This is early initialization for the top cgroup */
8405
		return &root_task_group.css;
8406 8407
	}

8408
	tg = sched_create_group(parent);
8409 8410 8411
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

8412 8413
	sched_online_group(tg, parent);

8414 8415 8416
	return &tg->css;
}

8417
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
8418
{
8419
	struct task_group *tg = css_tg(css);
8420

8421
	sched_offline_group(tg);
8422 8423
}

8424
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
8425
{
8426
	struct task_group *tg = css_tg(css);
8427

8428 8429 8430 8431
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
8432 8433
}

8434 8435 8436 8437
/*
 * 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.
 */
8438
static void cpu_cgroup_fork(struct task_struct *task)
8439
{
8440 8441 8442 8443 8444
	struct rq_flags rf;
	struct rq *rq;

	rq = task_rq_lock(task, &rf);

8445
	update_rq_clock(rq);
8446 8447 8448
	sched_change_group(task, TASK_SET_GROUP);

	task_rq_unlock(rq, task, &rf);
8449 8450
}

8451
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
8452
{
8453
	struct task_struct *task;
8454
	struct cgroup_subsys_state *css;
8455
	int ret = 0;
8456

8457
	cgroup_taskset_for_each(task, css, tset) {
8458
#ifdef CONFIG_RT_GROUP_SCHED
8459
		if (!sched_rt_can_attach(css_tg(css), task))
8460
			return -EINVAL;
8461
#else
8462 8463 8464
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
8465
#endif
8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481
		/*
		 * 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;
8482
	}
8483
	return ret;
8484
}
8485

8486
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
8487
{
8488
	struct task_struct *task;
8489
	struct cgroup_subsys_state *css;
8490

8491
	cgroup_taskset_for_each(task, css, tset)
8492
		sched_move_task(task);
8493 8494
}

8495
#ifdef CONFIG_FAIR_GROUP_SCHED
8496 8497
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
8498
{
8499
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
8500 8501
}

8502 8503
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
8504
{
8505
	struct task_group *tg = css_tg(css);
8506

8507
	return (u64) scale_load_down(tg->shares);
8508
}
8509 8510

#ifdef CONFIG_CFS_BANDWIDTH
8511 8512
static DEFINE_MUTEX(cfs_constraints_mutex);

8513 8514 8515
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

8516 8517
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

8518 8519
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
8520
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
8521
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8522 8523 8524 8525 8526 8527 8528 8529 8530 8531 8532 8533 8534 8535 8536 8537 8538 8539 8540 8541

	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;

8542 8543 8544 8545 8546
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
8547 8548 8549 8550 8551
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

8552
	runtime_enabled = quota != RUNTIME_INF;
8553
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
8554 8555 8556 8557 8558 8559
	/*
	 * 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();
8560 8561 8562
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
8563

P
Paul Turner 已提交
8564
	__refill_cfs_bandwidth_runtime(cfs_b);
8565
	/* restart the period timer (if active) to handle new period expiry */
P
Peter Zijlstra 已提交
8566 8567
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
8568 8569
	raw_spin_unlock_irq(&cfs_b->lock);

8570
	for_each_online_cpu(i) {
8571
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
8572
		struct rq *rq = cfs_rq->rq;
8573 8574

		raw_spin_lock_irq(&rq->lock);
8575
		cfs_rq->runtime_enabled = runtime_enabled;
8576
		cfs_rq->runtime_remaining = 0;
8577

8578
		if (cfs_rq->throttled)
8579
			unthrottle_cfs_rq(cfs_rq);
8580 8581
		raw_spin_unlock_irq(&rq->lock);
	}
8582 8583
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
8584 8585
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
8586
	put_online_cpus();
8587

8588
	return ret;
8589 8590 8591 8592 8593 8594
}

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

8595
	period = ktime_to_ns(tg->cfs_bandwidth.period);
8596 8597 8598 8599 8600 8601 8602 8603 8604 8605 8606 8607
	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;

8608
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
8609 8610
		return -1;

8611
	quota_us = tg->cfs_bandwidth.quota;
8612 8613 8614 8615 8616 8617 8618 8619 8620 8621
	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;
8622
	quota = tg->cfs_bandwidth.quota;
8623 8624 8625 8626 8627 8628 8629 8630

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

8631
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
8632 8633 8634 8635 8636
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

8637 8638
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
8639
{
8640
	return tg_get_cfs_quota(css_tg(css));
8641 8642
}

8643 8644
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
8645
{
8646
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
8647 8648
}

8649 8650
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8651
{
8652
	return tg_get_cfs_period(css_tg(css));
8653 8654
}

8655 8656
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
8657
{
8658
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
8659 8660
}

8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672 8673 8674 8675 8676 8677 8678 8679 8680 8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691 8692
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;
8693
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8694 8695 8696 8697 8698
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
8699
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
8700 8701

		quota = normalize_cfs_quota(tg, d);
8702
		parent_quota = parent_b->hierarchical_quota;
8703 8704 8705 8706 8707 8708 8709 8710 8711 8712

		/*
		 * 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;
	}
8713
	cfs_b->hierarchical_quota = quota;
8714 8715 8716 8717 8718 8719

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
8720
	int ret;
8721 8722 8723 8724 8725 8726 8727 8728 8729 8730 8731
	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);
	}

8732 8733 8734 8735 8736
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8737
}
8738

8739
static int cpu_stats_show(struct seq_file *sf, void *v)
8740
{
8741
	struct task_group *tg = css_tg(seq_css(sf));
8742
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8743

8744 8745 8746
	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);
8747 8748 8749

	return 0;
}
8750
#endif /* CONFIG_CFS_BANDWIDTH */
8751
#endif /* CONFIG_FAIR_GROUP_SCHED */
8752

8753
#ifdef CONFIG_RT_GROUP_SCHED
8754 8755
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
8756
{
8757
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
8758 8759
}

8760 8761
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
8762
{
8763
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
8764
}
8765

8766 8767
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
8768
{
8769
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
8770 8771
}

8772 8773
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8774
{
8775
	return sched_group_rt_period(css_tg(css));
8776
}
8777
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
8778

8779
static struct cftype cpu_files[] = {
8780
#ifdef CONFIG_FAIR_GROUP_SCHED
8781 8782
	{
		.name = "shares",
8783 8784
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
8785
	},
8786
#endif
8787 8788 8789 8790 8791 8792 8793 8794 8795 8796 8797
#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,
	},
8798 8799
	{
		.name = "stat",
8800
		.seq_show = cpu_stats_show,
8801
	},
8802
#endif
8803
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
8804
	{
P
Peter Zijlstra 已提交
8805
		.name = "rt_runtime_us",
8806 8807
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
8808
	},
8809 8810
	{
		.name = "rt_period_us",
8811 8812
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
8813
	},
8814
#endif
8815
	{ }	/* terminate */
8816 8817
};

8818
struct cgroup_subsys cpu_cgrp_subsys = {
8819
	.css_alloc	= cpu_cgroup_css_alloc,
8820
	.css_released	= cpu_cgroup_css_released,
8821
	.css_free	= cpu_cgroup_css_free,
8822
	.fork		= cpu_cgroup_fork,
8823 8824
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
8825
	.legacy_cftypes	= cpu_files,
8826
	.early_init	= true,
8827 8828
};

8829
#endif	/* CONFIG_CGROUP_SCHED */
8830

8831 8832 8833 8834 8835
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
8836 8837 8838 8839 8840 8841 8842 8843 8844 8845 8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868 8869 8870 8871 8872 8873 8874 8875 8876

/*
 * 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,
};