core.c 209.8 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>
L
Linus Torvalds 已提交
77

78
#include <asm/switch_to.h>
79
#include <asm/tlb.h>
80
#include <asm/irq_regs.h>
81
#include <asm/mutex.h>
G
Glauber Costa 已提交
82 83 84
#ifdef CONFIG_PARAVIRT
#include <asm/paravirt.h>
#endif
L
Linus Torvalds 已提交
85

86
#include "sched.h"
87
#include "../workqueue_internal.h"
88
#include "../smpboot.h"
89

90
#define CREATE_TRACE_POINTS
91
#include <trace/events/sched.h>
92

93 94
DEFINE_MUTEX(sched_domains_mutex);
DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues);
95

96
static void update_rq_clock_task(struct rq *rq, s64 delta);
97

98
void update_rq_clock(struct rq *rq)
99
{
100
	s64 delta;
101

102 103 104
	lockdep_assert_held(&rq->lock);

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

107
	delta = sched_clock_cpu(cpu_of(rq)) - rq->clock;
108 109
	if (delta < 0)
		return;
110 111
	rq->clock += delta;
	update_rq_clock_task(rq, delta);
112 113
}

I
Ingo Molnar 已提交
114 115 116
/*
 * Debugging: various feature bits
 */
P
Peter Zijlstra 已提交
117 118 119 120

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

I
Ingo Molnar 已提交
121
const_debug unsigned int sysctl_sched_features =
122
#include "features.h"
P
Peter Zijlstra 已提交
123 124 125 126
	0;

#undef SCHED_FEAT

127 128 129 130 131 132
/*
 * 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;

133 134 135 136 137 138 139 140
/*
 * 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
141
/*
P
Peter Zijlstra 已提交
142
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
143 144
 * default: 1s
 */
P
Peter Zijlstra 已提交
145
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
146

147
__read_mostly int scheduler_running;
148

P
Peter Zijlstra 已提交
149 150 151 152 153
/*
 * 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
154

155 156 157
/* cpus with isolated domains */
cpumask_var_t cpu_isolated_map;

L
Linus Torvalds 已提交
158
/*
159
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
160
 */
A
Alexey Dobriyan 已提交
161
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
162 163
	__acquires(rq->lock)
{
164
	struct rq *rq;
L
Linus Torvalds 已提交
165 166 167

	local_irq_disable();
	rq = this_rq();
168
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
169 170 171 172

	return rq;
}

173 174 175
/*
 * __task_rq_lock - lock the rq @p resides on.
 */
176
struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf)
177 178 179 180 181 182 183 184 185 186
	__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))) {
187
			rf->cookie = lockdep_pin_lock(&rq->lock);
188 189 190 191 192 193 194 195 196 197 198 199
			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.
 */
200
struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf)
201 202 203 204 205 206
	__acquires(p->pi_lock)
	__acquires(rq->lock)
{
	struct rq *rq;

	for (;;) {
207
		raw_spin_lock_irqsave(&p->pi_lock, rf->flags);
208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
		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))) {
227
			rf->cookie = lockdep_pin_lock(&rq->lock);
228 229 230
			return rq;
		}
		raw_spin_unlock(&rq->lock);
231
		raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags);
232 233 234 235 236 237

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

P
Peter Zijlstra 已提交
238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258
#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());

259
	raw_spin_lock(&rq->lock);
260
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
261
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
262
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
263 264 265 266

	return HRTIMER_NORESTART;
}

267
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
268

269
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
270 271 272
{
	struct hrtimer *timer = &rq->hrtick_timer;

273
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
274 275
}

276 277 278 279
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
280
{
281
	struct rq *rq = arg;
282

283
	raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
284
	__hrtick_restart(rq);
285
	rq->hrtick_csd_pending = 0;
286
	raw_spin_unlock(&rq->lock);
287 288
}

289 290 291 292 293
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
294
void hrtick_start(struct rq *rq, u64 delay)
295
{
296
	struct hrtimer *timer = &rq->hrtick_timer;
297 298 299 300 301 302 303 304 305
	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);
306

307
	hrtimer_set_expires(timer, time);
308 309

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
310
		__hrtick_restart(rq);
311
	} else if (!rq->hrtick_csd_pending) {
312
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
313 314
		rq->hrtick_csd_pending = 1;
	}
315 316
}

317 318 319 320 321 322
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
323
void hrtick_start(struct rq *rq, u64 delay)
324
{
W
Wanpeng Li 已提交
325 326 327 328 329
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
330 331
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
332 333
}
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
334

335
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
336
{
337 338
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
339

340 341 342 343
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
344

345 346
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
347
}
A
Andrew Morton 已提交
348
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
349 350 351 352 353 354 355
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}
A
Andrew Morton 已提交
356
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
357

358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
/*
 * 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;								\
})

376
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
377 378 379 380 381 382 383 384 385 386
/*
 * 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);
}
387 388 389 390 391 392 393 394 395 396

/*
 * 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);
397
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
398 399 400 401 402 403 404 405 406 407 408 409 410 411

	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;
}

412 413 414 415 416 417
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
418 419 420 421 422 423 424

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
425 426
#endif

427 428 429 430 431 432 433 434 435 436
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
437
	 * barrier implied by the wakeup in wake_up_q().
438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472
	 */
	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
473
/*
474
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
475 476 477 478 479
 *
 * 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.
 */
480
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
481
{
482
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
483 484
	int cpu;

485
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
486

487
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
488 489
		return;

490
	cpu = cpu_of(rq);
491

492
	if (cpu == smp_processor_id()) {
493
		set_tsk_need_resched(curr);
494
		set_preempt_need_resched();
I
Ingo Molnar 已提交
495
		return;
496
	}
I
Ingo Molnar 已提交
497

498
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
499
		smp_send_reschedule(cpu);
500 501
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
502 503
}

504
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
505 506 507 508
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

509
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
510
		return;
511
	resched_curr(rq);
512
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
513
}
514

515
#ifdef CONFIG_SMP
516
#ifdef CONFIG_NO_HZ_COMMON
517 518 519 520 521 522 523 524
/*
 * 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).
 */
525
int get_nohz_timer_target(void)
526
{
527
	int i, cpu = smp_processor_id();
528 529
	struct sched_domain *sd;

530
	if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
531 532
		return cpu;

533
	rcu_read_lock();
534
	for_each_domain(cpu, sd) {
535
		for_each_cpu(i, sched_domain_span(sd)) {
536 537 538 539
			if (cpu == i)
				continue;

			if (!idle_cpu(i) && is_housekeeping_cpu(i)) {
540 541 542 543
				cpu = i;
				goto unlock;
			}
		}
544
	}
545 546 547

	if (!is_housekeeping_cpu(cpu))
		cpu = housekeeping_any_cpu();
548 549
unlock:
	rcu_read_unlock();
550 551
	return cpu;
}
552 553 554 555 556 557 558 559 560 561
/*
 * 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.
 */
562
static void wake_up_idle_cpu(int cpu)
563 564 565 566 567 568
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

569
	if (set_nr_and_not_polling(rq->idle))
570
		smp_send_reschedule(cpu);
571 572
	else
		trace_sched_wake_idle_without_ipi(cpu);
573 574
}

575
static bool wake_up_full_nohz_cpu(int cpu)
576
{
577 578 579 580 581 582
	/*
	 * 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.
	 */
583
	if (tick_nohz_full_cpu(cpu)) {
584 585
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
586
			tick_nohz_full_kick_cpu(cpu);
587 588 589 590 591 592 593 594
		return true;
	}

	return false;
}

void wake_up_nohz_cpu(int cpu)
{
595
	if (!wake_up_full_nohz_cpu(cpu))
596 597 598
		wake_up_idle_cpu(cpu);
}

599
static inline bool got_nohz_idle_kick(void)
600
{
601
	int cpu = smp_processor_id();
602 603 604 605 606 607 608 609 610 611 612 613 614

	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;
615 616
}

617
#else /* CONFIG_NO_HZ_COMMON */
618

619
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
620
{
621
	return false;
P
Peter Zijlstra 已提交
622 623
}

624
#endif /* CONFIG_NO_HZ_COMMON */
625

626
#ifdef CONFIG_NO_HZ_FULL
627
bool sched_can_stop_tick(struct rq *rq)
628
{
629 630 631 632 633 634
	int fifo_nr_running;

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

635
	/*
636 637
	 * If there are more than one RR tasks, we need the tick to effect the
	 * actual RR behaviour.
638
	 */
639 640 641 642 643
	if (rq->rt.rr_nr_running) {
		if (rq->rt.rr_nr_running == 1)
			return true;
		else
			return false;
644 645
	}

646 647 648 649 650 651 652 653 654 655 656 657 658 659
	/*
	 * 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)
660
		return false;
661

662
	return true;
663 664
}
#endif /* CONFIG_NO_HZ_FULL */
665

666
void sched_avg_update(struct rq *rq)
667
{
668 669
	s64 period = sched_avg_period();

670
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
671 672 673 674 675 676
		/*
		 * 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));
677 678 679
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
680 681
}

682
#endif /* CONFIG_SMP */
683

684 685
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
686
/*
687 688 689 690
 * 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.
691
 */
692
int walk_tg_tree_from(struct task_group *from,
693
			     tg_visitor down, tg_visitor up, void *data)
694 695
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
696
	int ret;
697

698 699
	parent = from;

700
down:
P
Peter Zijlstra 已提交
701 702
	ret = (*down)(parent, data);
	if (ret)
703
		goto out;
704 705 706 707 708 709 710
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
711
	ret = (*up)(parent, data);
712 713
	if (ret || parent == from)
		goto out;
714 715 716 717 718

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
719
out:
P
Peter Zijlstra 已提交
720
	return ret;
721 722
}

723
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
724
{
725
	return 0;
P
Peter Zijlstra 已提交
726
}
727 728
#endif

729 730
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
731 732 733
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
734 735 736
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
737
	if (idle_policy(p->policy)) {
738
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
739
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
740 741
		return;
	}
742

743 744
	load->weight = scale_load(sched_prio_to_weight[prio]);
	load->inv_weight = sched_prio_to_wmult[prio];
745 746
}

747
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
748
{
749
	update_rq_clock(rq);
750 751
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
752
	p->sched_class->enqueue_task(rq, p, flags);
753 754
}

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

763
void activate_task(struct rq *rq, struct task_struct *p, int flags)
764 765 766 767
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

768
	enqueue_task(rq, p, flags);
769 770
}

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

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

779
static void update_rq_clock_task(struct rq *rq, s64 delta)
780
{
781 782 783 784 785 786 787 788
/*
 * 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
789
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810

	/*
	 * 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;
811 812
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
813
	if (static_key_false((&paravirt_steal_rq_enabled))) {
814 815 816 817 818 819 820 821 822 823 824
		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

825 826
	rq->clock_task += delta;

827
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
828
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
829 830
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
831 832
}

833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862
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;
	}
}

863
/*
I
Ingo Molnar 已提交
864
 * __normal_prio - return the priority that is based on the static prio
865 866 867
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
868
	return p->static_prio;
869 870
}

871 872 873 874 875 876 877
/*
 * 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.
 */
878
static inline int normal_prio(struct task_struct *p)
879 880 881
{
	int prio;

882 883 884
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
885 886 887 888 889 890 891 892 893 894 895 896 897
		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.
 */
898
static int effective_prio(struct task_struct *p)
899 900 901 902 903 904 905 906 907 908 909 910
{
	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
911 912 913
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
914 915
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
916
 */
917
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
918 919 920 921
{
	return cpu_curr(task_cpu(p)) == p;
}

922
/*
923 924 925 926 927
 * 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().
928
 */
929 930
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
931
				       int oldprio)
932 933 934
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
935
			prev_class->switched_from(rq, p);
936

P
Peter Zijlstra 已提交
937
		p->sched_class->switched_to(rq, p);
938
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
939
		p->sched_class->prio_changed(rq, p, oldprio);
940 941
}

942
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
943 944 945 946 947 948 949 950 951 952
{
	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) {
953
				resched_curr(rq);
954 955 956 957 958 959 960 961 962
				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.
	 */
963
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
964
		rq_clock_skip_update(rq, true);
965 966
}

L
Linus Torvalds 已提交
967
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986
/*
 * 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.
 */
987
static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
988 989 990 991
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
992
	dequeue_task(rq, p, 0);
P
Peter Zijlstra 已提交
993 994 995 996 997 998 999 1000
	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);
1001
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020
	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.
 */
1021
static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
1022 1023
{
	if (unlikely(!cpu_active(dest_cpu)))
1024
		return rq;
P
Peter Zijlstra 已提交
1025 1026 1027

	/* Affinity changed (again). */
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
1028
		return rq;
P
Peter Zijlstra 已提交
1029

1030 1031 1032
	rq = move_queued_task(rq, p, dest_cpu);

	return rq;
P
Peter Zijlstra 已提交
1033 1034 1035 1036 1037 1038 1039 1040 1041 1042
}

/*
 * 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;
1043 1044
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056

	/*
	 * 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();
1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069

	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.
	 */
	if (task_rq(p) == rq && task_on_rq_queued(p))
		rq = __migrate_task(rq, p, arg->dest_cpu);
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock(&p->pi_lock);

P
Peter Zijlstra 已提交
1070 1071 1072 1073
	local_irq_enable();
	return 0;
}

1074 1075 1076 1077 1078
/*
 * 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
1079 1080 1081 1082 1083
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1084 1085
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1086 1087 1088
	struct rq *rq = task_rq(p);
	bool queued, running;

1089
	lockdep_assert_held(&p->pi_lock);
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099

	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);
1100
		dequeue_task(rq, p, DEQUEUE_SAVE);
1101 1102 1103 1104
	}
	if (running)
		put_prev_task(rq, p);

1105
	p->sched_class->set_cpus_allowed(p, new_mask);
1106 1107 1108 1109

	if (running)
		p->sched_class->set_curr_task(rq);
	if (queued)
1110
		enqueue_task(rq, p, ENQUEUE_RESTORE);
1111 1112
}

P
Peter Zijlstra 已提交
1113 1114 1115 1116 1117 1118 1119 1120 1121
/*
 * 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.
 */
1122 1123
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1124
{
1125
	const struct cpumask *cpu_valid_mask = cpu_active_mask;
P
Peter Zijlstra 已提交
1126
	unsigned int dest_cpu;
1127 1128
	struct rq_flags rf;
	struct rq *rq;
P
Peter Zijlstra 已提交
1129 1130
	int ret = 0;

1131
	rq = task_rq_lock(p, &rf);
P
Peter Zijlstra 已提交
1132

1133 1134 1135 1136 1137 1138 1139
	if (p->flags & PF_KTHREAD) {
		/*
		 * Kernel threads are allowed on online && !active CPUs
		 */
		cpu_valid_mask = cpu_online_mask;
	}

1140 1141 1142 1143 1144 1145 1146 1147 1148
	/*
	 * 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
1149 1150 1151
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

1152
	if (!cpumask_intersects(new_mask, cpu_valid_mask)) {
P
Peter Zijlstra 已提交
1153 1154 1155 1156 1157 1158
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

1159 1160 1161 1162 1163 1164 1165 1166 1167 1168
	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
1169 1170 1171 1172
	/* 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;

1173
	dest_cpu = cpumask_any_and(cpu_valid_mask, new_mask);
P
Peter Zijlstra 已提交
1174 1175 1176
	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. */
1177
		task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1178 1179 1180
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1181 1182 1183 1184 1185
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
1186
		lockdep_unpin_lock(&rq->lock, rf.cookie);
1187
		rq = move_queued_task(rq, p, dest_cpu);
1188
		lockdep_repin_lock(&rq->lock, rf.cookie);
1189
	}
P
Peter Zijlstra 已提交
1190
out:
1191
	task_rq_unlock(rq, p, &rf);
P
Peter Zijlstra 已提交
1192 1193 1194

	return ret;
}
1195 1196 1197 1198 1199

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
1200 1201
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1202
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1203
{
1204 1205 1206 1207 1208
#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
1209
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1210
			!p->on_rq);
1211

1212 1213 1214 1215 1216 1217 1218 1219 1220
	/*
	 * 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)));

1221
#ifdef CONFIG_LOCKDEP
1222 1223 1224 1225 1226
	/*
	 * 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
1227
	 * see task_group().
1228 1229 1230 1231
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1232 1233 1234
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1235 1236
#endif

1237
	trace_sched_migrate_task(p, new_cpu);
1238

1239
	if (task_cpu(p) != new_cpu) {
1240
		if (p->sched_class->migrate_task_rq)
1241
			p->sched_class->migrate_task_rq(p);
1242
		p->se.nr_migrations++;
1243
		perf_event_task_migrate(p);
1244
	}
I
Ingo Molnar 已提交
1245 1246

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1247 1248
}

1249 1250
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1251
	if (task_on_rq_queued(p)) {
1252 1253 1254 1255 1256
		struct rq *src_rq, *dst_rq;

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

1257
		p->on_rq = TASK_ON_RQ_MIGRATING;
1258 1259 1260
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1261
		p->on_rq = TASK_ON_RQ_QUEUED;
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283
		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
		 * previous cpu our targer instead of where it really is.
		 */
		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;

1284 1285 1286
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1287 1288 1289
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1290 1291
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1292
	double_rq_lock(src_rq, dst_rq);
1293

1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312
	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);
1313 1314
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336

	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;

1337 1338 1339 1340
	/*
	 * 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.
	 */
1341 1342 1343 1344 1345 1346 1347 1348 1349
	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;

1350
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1351 1352 1353 1354 1355 1356
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1357 1358 1359
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1360 1361 1362 1363 1364 1365 1366
 * 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
1367 1368 1369 1370 1371 1372
 * 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
1373
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1374
{
1375
	int running, queued;
1376
	struct rq_flags rf;
R
Roland McGrath 已提交
1377
	unsigned long ncsw;
1378
	struct rq *rq;
L
Linus Torvalds 已提交
1379

1380 1381 1382 1383 1384 1385 1386 1387
	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);
1388

1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399
		/*
		 * 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
1400 1401 1402
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1403
			cpu_relax();
R
Roland McGrath 已提交
1404
		}
1405

1406 1407 1408 1409 1410
		/*
		 * 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.
		 */
1411
		rq = task_rq_lock(p, &rf);
1412
		trace_sched_wait_task(p);
1413
		running = task_running(rq, p);
1414
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1415
		ncsw = 0;
1416
		if (!match_state || p->state == match_state)
1417
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1418
		task_rq_unlock(rq, p, &rf);
1419

R
Roland McGrath 已提交
1420 1421 1422 1423 1424 1425
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1426 1427 1428 1429 1430 1431 1432 1433 1434 1435
		/*
		 * 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;
		}
1436

1437 1438 1439 1440 1441
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1442
		 * So if it was still runnable (but just not actively
1443 1444 1445
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1446
		if (unlikely(queued)) {
1447 1448 1449 1450
			ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1451 1452
			continue;
		}
1453

1454 1455 1456 1457 1458 1459 1460
		/*
		 * 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
1461 1462

	return ncsw;
L
Linus Torvalds 已提交
1463 1464 1465 1466 1467 1468 1469 1470 1471
}

/***
 * 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
1472
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1473 1474 1475 1476 1477
 * 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.
 */
1478
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1479 1480 1481 1482 1483 1484 1485 1486 1487
{
	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
1488
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1489

1490
/*
1491
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510
 *
 * 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.
1511
 */
1512 1513
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1514 1515
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1516 1517
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1518

1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
	/*
	 * 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;
		}
1534
	}
1535

1536 1537
	for (;;) {
		/* Any allowed, online CPU? */
1538
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1539 1540 1541 1542
			if (!cpu_active(dest_cpu))
				continue;
			goto out;
		}
1543

1544
		/* No more Mr. Nice Guy. */
1545 1546
		switch (state) {
		case cpuset:
1547 1548 1549 1550 1551 1552
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
			/* fall-through */
1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571
		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()) {
1572
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1573 1574
					task_pid_nr(p), p->comm, cpu);
		}
1575 1576 1577 1578 1579
	}

	return dest_cpu;
}

1580
/*
1581
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1582
 */
1583
static inline
1584
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1585
{
1586 1587
	lockdep_assert_held(&p->pi_lock);

1588
	if (tsk_nr_cpus_allowed(p) > 1)
1589
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1590 1591
	else
		cpu = cpumask_any(tsk_cpus_allowed(p));
1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602

	/*
	 * 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 ]
	 */
1603
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1604
		     !cpu_online(cpu)))
1605
		cpu = select_fallback_rq(task_cpu(p), p);
1606 1607

	return cpu;
1608
}
1609 1610 1611 1612 1613 1614

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1615 1616 1617 1618 1619 1620 1621 1622 1623

#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
1624
#endif /* CONFIG_SMP */
1625

P
Peter Zijlstra 已提交
1626
static void
1627
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1628
{
P
Peter Zijlstra 已提交
1629
#ifdef CONFIG_SCHEDSTATS
1630 1631
	struct rq *rq = this_rq();

P
Peter Zijlstra 已提交
1632 1633 1634 1635 1636 1637 1638 1639 1640 1641
#ifdef CONFIG_SMP
	int this_cpu = smp_processor_id();

	if (cpu == this_cpu) {
		schedstat_inc(rq, ttwu_local);
		schedstat_inc(p, se.statistics.nr_wakeups_local);
	} else {
		struct sched_domain *sd;

		schedstat_inc(p, se.statistics.nr_wakeups_remote);
1642
		rcu_read_lock();
P
Peter Zijlstra 已提交
1643 1644 1645 1646 1647 1648
		for_each_domain(this_cpu, sd) {
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
				schedstat_inc(sd, ttwu_wake_remote);
				break;
			}
		}
1649
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1650
	}
1651 1652 1653 1654

	if (wake_flags & WF_MIGRATED)
		schedstat_inc(p, se.statistics.nr_wakeups_migrate);

P
Peter Zijlstra 已提交
1655 1656 1657
#endif /* CONFIG_SMP */

	schedstat_inc(rq, ttwu_count);
T
Tejun Heo 已提交
1658
	schedstat_inc(p, se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1659 1660

	if (wake_flags & WF_SYNC)
T
Tejun Heo 已提交
1661
		schedstat_inc(p, se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1662 1663 1664 1665

#endif /* CONFIG_SCHEDSTATS */
}

1666
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1667
{
T
Tejun Heo 已提交
1668
	activate_task(rq, p, en_flags);
1669
	p->on_rq = TASK_ON_RQ_QUEUED;
1670 1671 1672 1673

	/* 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
1674 1675
}

1676 1677 1678
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1679 1680
static void ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags,
			   struct pin_cookie cookie)
T
Tejun Heo 已提交
1681 1682 1683
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1684 1685
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1686
#ifdef CONFIG_SMP
1687 1688
	if (p->sched_class->task_woken) {
		/*
1689 1690
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1691
		 */
1692
		lockdep_unpin_lock(&rq->lock, cookie);
T
Tejun Heo 已提交
1693
		p->sched_class->task_woken(rq, p);
1694
		lockdep_repin_lock(&rq->lock, cookie);
1695
	}
T
Tejun Heo 已提交
1696

1697
	if (rq->idle_stamp) {
1698
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1699
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1700

1701 1702 1703
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1704
			rq->avg_idle = max;
1705

T
Tejun Heo 已提交
1706 1707 1708 1709 1710
		rq->idle_stamp = 0;
	}
#endif
}

1711
static void
1712 1713
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags,
		 struct pin_cookie cookie)
1714
{
1715 1716
	int en_flags = ENQUEUE_WAKEUP;

1717 1718
	lockdep_assert_held(&rq->lock);

1719 1720 1721
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
1722 1723

	if (wake_flags & WF_MIGRATED)
1724
		en_flags |= ENQUEUE_MIGRATED;
1725 1726
#endif

1727
	ttwu_activate(rq, p, en_flags);
1728
	ttwu_do_wakeup(rq, p, wake_flags, cookie);
1729 1730 1731 1732 1733 1734 1735 1736 1737 1738
}

/*
 * 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)
{
1739
	struct rq_flags rf;
1740 1741 1742
	struct rq *rq;
	int ret = 0;

1743
	rq = __task_rq_lock(p, &rf);
1744
	if (task_on_rq_queued(p)) {
1745 1746
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1747
		ttwu_do_wakeup(rq, p, wake_flags, rf.cookie);
1748 1749
		ret = 1;
	}
1750
	__task_rq_unlock(rq, &rf);
1751 1752 1753 1754

	return ret;
}

1755
#ifdef CONFIG_SMP
1756
void sched_ttwu_pending(void)
1757 1758
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1759
	struct llist_node *llist = llist_del_all(&rq->wake_list);
1760
	struct pin_cookie cookie;
P
Peter Zijlstra 已提交
1761
	struct task_struct *p;
1762
	unsigned long flags;
1763

1764 1765 1766 1767
	if (!llist)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
1768
	cookie = lockdep_pin_lock(&rq->lock);
1769

P
Peter Zijlstra 已提交
1770 1771 1772
	while (llist) {
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
1773 1774 1775 1776 1777
		/*
		 * See ttwu_queue(); we only call ttwu_queue_remote() when
		 * its a x-cpu wakeup.
		 */
		ttwu_do_activate(rq, p, WF_MIGRATED, cookie);
1778 1779
	}

1780
	lockdep_unpin_lock(&rq->lock, cookie);
1781
	raw_spin_unlock_irqrestore(&rq->lock, flags);
1782 1783 1784 1785
}

void scheduler_ipi(void)
{
1786 1787 1788 1789 1790
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1791
	preempt_fold_need_resched();
1792

1793
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809
		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
1810
	sched_ttwu_pending();
1811 1812 1813 1814

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1815
	if (unlikely(got_nohz_idle_kick())) {
1816
		this_rq()->idle_balance = 1;
1817
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1818
	}
1819
	irq_exit();
1820 1821 1822 1823
}

static void ttwu_queue_remote(struct task_struct *p, int cpu)
{
1824 1825 1826 1827 1828 1829 1830 1831
	struct rq *rq = cpu_rq(cpu);

	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);
	}
1832
}
1833

1834 1835 1836 1837 1838
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

1839 1840 1841 1842
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1843 1844 1845 1846 1847 1848 1849 1850 1851 1852

	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);
	}
1853 1854 1855

out:
	rcu_read_unlock();
1856 1857
}

1858
bool cpus_share_cache(int this_cpu, int that_cpu)
1859 1860 1861
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1862
#endif /* CONFIG_SMP */
1863

1864
static void ttwu_queue(struct task_struct *p, int cpu, int wake_flags)
1865 1866
{
	struct rq *rq = cpu_rq(cpu);
1867
	struct pin_cookie cookie;
1868

1869
#if defined(CONFIG_SMP)
1870
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1871
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
1872 1873 1874 1875 1876
		ttwu_queue_remote(p, cpu);
		return;
	}
#endif

1877
	raw_spin_lock(&rq->lock);
1878
	cookie = lockdep_pin_lock(&rq->lock);
1879
	ttwu_do_activate(rq, p, wake_flags, cookie);
1880
	lockdep_unpin_lock(&rq->lock, cookie);
1881
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
1882 1883
}

1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 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 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974
/*
 * 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)
 *   2) smp_cond_acquire(!X->on_cpu)
 *
 * 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);
 *
 *                    smp_cond_acquire(!X->on_cpu);
 *                    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,
 * since the waking CPU is the one issueing the ACQUIRE (smp_cond_acquire).
 *
 */

T
Tejun Heo 已提交
1975
/**
L
Linus Torvalds 已提交
1976
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1977
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1978
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1979
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1980 1981 1982 1983 1984 1985 1986
 *
 * Put it on the run-queue if it's not already there. The "current"
 * thread is always on the run-queue (except when the actual
 * re-schedule is in progress), and as such you're allowed to do
 * the simpler "current->state = TASK_RUNNING" to mark yourself
 * runnable without the overhead of this.
 *
1987
 * Return: %true if @p was woken up, %false if it was already running.
T
Tejun Heo 已提交
1988
 * or @state didn't match @p's state.
L
Linus Torvalds 已提交
1989
 */
1990 1991
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1992 1993
{
	unsigned long flags;
1994
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1995

1996 1997 1998 1999 2000 2001 2002
	/*
	 * 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();
2003
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
2004
	if (!(p->state & state))
L
Linus Torvalds 已提交
2005 2006
		goto out;

2007 2008
	trace_sched_waking(p);

2009
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
2010 2011
	cpu = task_cpu(p);

2012 2013
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
2014 2015

#ifdef CONFIG_SMP
2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
	/*
	 * 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
2035
	/*
2036 2037
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
2038 2039 2040 2041 2042
	 *
	 * 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.
2043
	 */
2044
	smp_cond_acquire(!p->on_cpu);
L
Linus Torvalds 已提交
2045

2046
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
2047
	p->state = TASK_WAKING;
2048

2049
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2050 2051
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2052
		set_task_cpu(p, cpu);
2053
	}
L
Linus Torvalds 已提交
2054 2055
#endif /* CONFIG_SMP */

2056
	ttwu_queue(p, cpu, wake_flags);
2057
stat:
2058 2059
	if (schedstat_enabled())
		ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2060
out:
2061
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2062 2063 2064 2065

	return success;
}

T
Tejun Heo 已提交
2066 2067 2068 2069
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
 *
2070
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2071
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2072
 * the current task.
T
Tejun Heo 已提交
2073
 */
2074
static void try_to_wake_up_local(struct task_struct *p, struct pin_cookie cookie)
T
Tejun Heo 已提交
2075 2076 2077
{
	struct rq *rq = task_rq(p);

2078 2079 2080 2081
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2082 2083
	lockdep_assert_held(&rq->lock);

2084
	if (!raw_spin_trylock(&p->pi_lock)) {
2085 2086 2087 2088 2089 2090
		/*
		 * 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.
		 */
2091
		lockdep_unpin_lock(&rq->lock, cookie);
2092 2093 2094
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
2095
		lockdep_repin_lock(&rq->lock, cookie);
2096 2097
	}

T
Tejun Heo 已提交
2098
	if (!(p->state & TASK_NORMAL))
2099
		goto out;
T
Tejun Heo 已提交
2100

2101 2102
	trace_sched_waking(p);

2103
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
2104 2105
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

2106
	ttwu_do_wakeup(rq, p, 0, cookie);
2107 2108
	if (schedstat_enabled())
		ttwu_stat(p, smp_processor_id(), 0);
2109 2110
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2111 2112
}

2113 2114 2115 2116 2117
/**
 * 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
2118 2119 2120
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2121 2122 2123 2124
 *
 * 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.
 */
2125
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2126
{
2127
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2128 2129 2130
}
EXPORT_SYMBOL(wake_up_process);

2131
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2132 2133 2134 2135
{
	return try_to_wake_up(p, state, 0);
}

2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147
/*
 * 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;
2148 2149 2150

	dl_se->dl_throttled = 0;
	dl_se->dl_yielded = 0;
2151 2152
}

L
Linus Torvalds 已提交
2153 2154 2155
/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2156 2157 2158
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2159
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2160
{
P
Peter Zijlstra 已提交
2161 2162 2163
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2164 2165
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2166
	p->se.prev_sum_exec_runtime	= 0;
2167
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2168
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2169
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2170

2171 2172 2173 2174
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2175
#ifdef CONFIG_SCHEDSTATS
2176
	/* Even if schedstat is disabled, there should not be garbage */
2177
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2178
#endif
N
Nick Piggin 已提交
2179

2180
	RB_CLEAR_NODE(&p->dl.rb_node);
2181
	init_dl_task_timer(&p->dl);
2182
	__dl_clear_params(p);
2183

P
Peter Zijlstra 已提交
2184
	INIT_LIST_HEAD(&p->rt.run_list);
2185 2186 2187 2188
	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
2189

2190 2191 2192
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2193 2194 2195

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2196
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2197 2198 2199
		p->mm->numa_scan_seq = 0;
	}

2200 2201 2202 2203 2204
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2205 2206
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2207
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2208
	p->numa_work.next = &p->numa_work;
2209
	p->numa_faults = NULL;
2210 2211
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2212 2213

	p->numa_group = NULL;
2214
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2215 2216
}

2217 2218
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2219
#ifdef CONFIG_NUMA_BALANCING
2220

2221 2222 2223
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2224
		static_branch_enable(&sched_numa_balancing);
2225
	else
2226
		static_branch_disable(&sched_numa_balancing);
2227
}
2228 2229 2230 2231 2232 2233 2234

#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;
2235
	int state = static_branch_likely(&sched_numa_balancing);
2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250

	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
2251

2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313
DEFINE_STATIC_KEY_FALSE(sched_schedstats);

#ifdef CONFIG_SCHEDSTATS
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;

	if (!strcmp(str, "enable")) {
		set_schedstats(true);
		ret = 1;
	} else if (!strcmp(str, "disable")) {
		set_schedstats(false);
		ret = 1;
	}
out:
	if (!ret)
		pr_warn("Unable to parse schedstats=\n");

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

#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;
}
#endif
#endif
I
Ingo Molnar 已提交
2314 2315 2316 2317

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

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

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

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

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

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

P
Peter Zijlstra 已提交
2366 2367 2368
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);

2369 2370 2371 2372 2373 2374 2375
	/*
	 * 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.
	 */
2376
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2377
	set_task_cpu(p, cpu);
2378
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2379

2380
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2381
	if (likely(sched_info_on()))
2382
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2383
#endif
P
Peter Zijlstra 已提交
2384 2385
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2386
#endif
2387
	init_task_preempt_count(p);
2388
#ifdef CONFIG_SMP
2389
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2390
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2391
#endif
2392

N
Nick Piggin 已提交
2393
	put_cpu();
2394
	return 0;
L
Linus Torvalds 已提交
2395 2396
}

2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415
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)
{
2416 2417
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2418 2419 2420
	return &cpu_rq(i)->rd->dl_bw;
}

2421
static inline int dl_bw_cpus(int i)
2422
{
2423 2424 2425
	struct root_domain *rd = cpu_rq(i)->rd;
	int cpus = 0;

2426 2427
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2428 2429 2430 2431
	for_each_cpu_and(i, rd->span, cpu_active_mask)
		cpus++;

	return cpus;
2432 2433 2434 2435 2436 2437 2438
}
#else
inline struct dl_bw *dl_bw_of(int i)
{
	return &cpu_rq(i)->dl.dl_bw;
}

2439
static inline int dl_bw_cpus(int i)
2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451
{
	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.
2452 2453 2454
 *
 * XXX we should delay bw change until the task's 0-lag point, see
 * __setparam_dl().
2455 2456 2457 2458 2459 2460
 */
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));
2461
	u64 period = attr->sched_period ?: attr->sched_deadline;
2462 2463
	u64 runtime = attr->sched_runtime;
	u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2464
	int cpus, err = -1;
2465

2466 2467
	/* !deadline task may carry old deadline bandwidth */
	if (new_bw == p->dl.dl_bw && task_has_dl_policy(p))
2468 2469 2470 2471 2472 2473 2474 2475
		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);
2476
	cpus = dl_bw_cpus(task_cpu(p));
2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496
	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
2497 2498 2499 2500 2501 2502 2503
/*
 * 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.
 */
2504
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2505
{
2506
	struct rq_flags rf;
I
Ingo Molnar 已提交
2507
	struct rq *rq;
2508

2509 2510
	/* Initialize new task's runnable average */
	init_entity_runnable_average(&p->se);
2511
	raw_spin_lock_irqsave(&p->pi_lock, rf.flags);
2512 2513 2514 2515 2516 2517
#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
	 */
2518
	set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2519
#endif
2520 2521
	/* Post initialize new task's util average when its cfs_rq is set */
	post_init_entity_util_avg(&p->se);
2522

2523
	rq = __task_rq_lock(p, &rf);
P
Peter Zijlstra 已提交
2524
	activate_task(rq, p, 0);
2525
	p->on_rq = TASK_ON_RQ_QUEUED;
2526
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2527
	check_preempt_curr(rq, p, WF_FORK);
2528
#ifdef CONFIG_SMP
2529 2530 2531 2532 2533
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
2534
		lockdep_unpin_lock(&rq->lock, rf.cookie);
2535
		p->sched_class->task_woken(rq, p);
2536
		lockdep_repin_lock(&rq->lock, rf.cookie);
2537
	}
2538
#endif
2539
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
2540 2541
}

2542 2543
#ifdef CONFIG_PREEMPT_NOTIFIERS

2544 2545
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557
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);

2558
/**
2559
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2560
 * @notifier: notifier struct to register
2561 2562 2563
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2564 2565 2566
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2567 2568 2569 2570 2571 2572
	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
2573
 * @notifier: notifier struct to unregister
2574
 *
2575
 * This is *not* safe to call from within a preemption notifier.
2576 2577 2578 2579 2580 2581 2582
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2583
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2584 2585 2586
{
	struct preempt_notifier *notifier;

2587
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2588 2589 2590
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2591 2592 2593 2594 2595 2596
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);
}

2597
static void
2598 2599
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2600 2601 2602
{
	struct preempt_notifier *notifier;

2603
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2604 2605 2606
		notifier->ops->sched_out(notifier, next);
}

2607 2608 2609 2610 2611 2612 2613 2614
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);
}

2615
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2616

2617
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2618 2619 2620
{
}

2621
static inline void
2622 2623 2624 2625 2626
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2627
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2628

2629 2630 2631
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2632
 * @prev: the current task that is being switched out
2633 2634 2635 2636 2637 2638 2639 2640 2641
 * @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.
 */
2642 2643 2644
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2645
{
2646
	sched_info_switch(rq, prev, next);
2647
	perf_event_task_sched_out(prev, next);
2648
	fire_sched_out_preempt_notifiers(prev, next);
2649 2650 2651 2652
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2653 2654 2655 2656
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2657 2658 2659 2660
 * 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
2661 2662
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2663
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2664 2665
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2666 2667 2668 2669 2670
 *
 * 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
2671
 */
2672
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2673 2674
	__releases(rq->lock)
{
2675
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2676
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2677
	long prev_state;
L
Linus Torvalds 已提交
2678

2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689
	/*
	 * 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.
	 */
2690 2691 2692 2693
	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);
2694

L
Linus Torvalds 已提交
2695 2696 2697 2698
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2699
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2700 2701
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2702 2703 2704 2705 2706
	 *
	 * 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
2707
	 */
O
Oleg Nesterov 已提交
2708
	prev_state = prev->state;
2709
	vtime_task_switch(prev);
2710
	perf_event_task_sched_in(prev, current);
2711
	finish_lock_switch(rq, prev);
2712
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2713

2714
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
2715 2716
	if (mm)
		mmdrop(mm);
2717
	if (unlikely(prev_state == TASK_DEAD)) {
2718 2719 2720
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2721 2722 2723
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2724
		 */
2725
		kprobe_flush_task(prev);
L
Linus Torvalds 已提交
2726
		put_task_struct(prev);
2727
	}
2728

2729
	tick_nohz_task_switch();
2730
	return rq;
L
Linus Torvalds 已提交
2731 2732
}

2733 2734 2735
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2736
static void __balance_callback(struct rq *rq)
2737
{
2738 2739 2740
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2741

2742 2743 2744 2745 2746 2747 2748 2749
	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;
2750

2751
		func(rq);
2752
	}
2753 2754 2755 2756 2757 2758 2759
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2760 2761 2762
}

#else
2763

2764
static inline void balance_callback(struct rq *rq)
2765
{
L
Linus Torvalds 已提交
2766 2767
}

2768 2769
#endif

L
Linus Torvalds 已提交
2770 2771 2772 2773
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2774
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2775 2776
	__releases(rq->lock)
{
2777
	struct rq *rq;
2778

2779 2780 2781 2782 2783 2784 2785 2786 2787
	/*
	 * 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).
	 */

2788
	rq = finish_task_switch(prev);
2789
	balance_callback(rq);
2790
	preempt_enable();
2791

L
Linus Torvalds 已提交
2792
	if (current->set_child_tid)
2793
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2794 2795 2796
}

/*
2797
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2798
 */
2799
static __always_inline struct rq *
2800
context_switch(struct rq *rq, struct task_struct *prev,
2801
	       struct task_struct *next, struct pin_cookie cookie)
L
Linus Torvalds 已提交
2802
{
I
Ingo Molnar 已提交
2803
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2804

2805
	prepare_task_switch(rq, prev, next);
2806

I
Ingo Molnar 已提交
2807 2808
	mm = next->mm;
	oldmm = prev->active_mm;
2809 2810 2811 2812 2813
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2814
	arch_start_context_switch(prev);
2815

2816
	if (!mm) {
L
Linus Torvalds 已提交
2817 2818 2819 2820
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
2821
		switch_mm_irqs_off(oldmm, mm, next);
L
Linus Torvalds 已提交
2822

2823
	if (!prev->mm) {
L
Linus Torvalds 已提交
2824 2825 2826
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2827 2828 2829 2830 2831 2832
	/*
	 * 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:
	 */
2833
	lockdep_unpin_lock(&rq->lock, cookie);
2834
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2835 2836 2837

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2838
	barrier();
2839 2840

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2841 2842 2843
}

/*
2844
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2845 2846
 *
 * externally visible scheduler statistics: current number of runnable
2847
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2848 2849 2850 2851 2852 2853 2854 2855 2856
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2857
}
L
Linus Torvalds 已提交
2858

2859 2860
/*
 * Check if only the current task is running on the cpu.
2861 2862 2863 2864 2865 2866 2867 2868 2869 2870
 *
 * 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)
2871 2872 2873
 */
bool single_task_running(void)
{
2874
	return raw_rq()->nr_running == 1;
2875 2876 2877
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2878
unsigned long long nr_context_switches(void)
2879
{
2880 2881
	int i;
	unsigned long long sum = 0;
2882

2883
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2884
		sum += cpu_rq(i)->nr_switches;
2885

L
Linus Torvalds 已提交
2886 2887
	return sum;
}
2888

L
Linus Torvalds 已提交
2889 2890 2891
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2892

2893
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2894
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2895

L
Linus Torvalds 已提交
2896 2897
	return sum;
}
2898

2899
unsigned long nr_iowait_cpu(int cpu)
2900
{
2901
	struct rq *this = cpu_rq(cpu);
2902 2903
	return atomic_read(&this->nr_iowait);
}
2904

2905 2906
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2907 2908 2909
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2910 2911
}

I
Ingo Molnar 已提交
2912
#ifdef CONFIG_SMP
2913

2914
/*
P
Peter Zijlstra 已提交
2915 2916
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2917
 */
P
Peter Zijlstra 已提交
2918
void sched_exec(void)
2919
{
P
Peter Zijlstra 已提交
2920
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2921
	unsigned long flags;
2922
	int dest_cpu;
2923

2924
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2925
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2926 2927
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2928

2929
	if (likely(cpu_active(dest_cpu))) {
2930
		struct migration_arg arg = { p, dest_cpu };
2931

2932 2933
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2934 2935
		return;
	}
2936
unlock:
2937
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2938
}
I
Ingo Molnar 已提交
2939

L
Linus Torvalds 已提交
2940 2941 2942
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2943
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2944 2945

EXPORT_PER_CPU_SYMBOL(kstat);
2946
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2947

2948 2949 2950 2951 2952 2953 2954
/*
 * 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)
{
2955
	struct rq_flags rf;
2956
	struct rq *rq;
2957
	u64 ns;
2958

2959 2960 2961 2962 2963 2964 2965 2966 2967
#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.
2968 2969
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
2970
	 */
2971
	if (!p->on_cpu || !task_on_rq_queued(p))
2972 2973 2974
		return p->se.sum_exec_runtime;
#endif

2975
	rq = task_rq_lock(p, &rf);
2976 2977 2978 2979 2980 2981 2982 2983 2984 2985
	/*
	 * 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)) {
		update_rq_clock(rq);
		p->sched_class->update_curr(rq);
	}
	ns = p->se.sum_exec_runtime;
2986
	task_rq_unlock(rq, p, &rf);
2987 2988 2989

	return ns;
}
2990

2991 2992 2993 2994 2995 2996 2997 2998
/*
 * 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
2999
	struct task_struct *curr = rq->curr;
3000 3001

	sched_clock_tick();
I
Ingo Molnar 已提交
3002

3003
	raw_spin_lock(&rq->lock);
3004
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
3005
	curr->sched_class->task_tick(rq, curr, 0);
3006
	cpu_load_update_active(rq);
3007
	calc_global_load_tick(rq);
3008
	raw_spin_unlock(&rq->lock);
3009

3010
	perf_event_task_tick();
3011

3012
#ifdef CONFIG_SMP
3013
	rq->idle_balance = idle_cpu(cpu);
3014
	trigger_load_balance(rq);
3015
#endif
3016
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
3017 3018
}

3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029
#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.
3030 3031
 *
 * Return: Maximum deferment in nanoseconds.
3032 3033 3034 3035
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
3036
	unsigned long next, now = READ_ONCE(jiffies);
3037 3038 3039 3040 3041 3042

	next = rq->last_sched_tick + HZ;

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

3043
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
3044
}
3045
#endif
L
Linus Torvalds 已提交
3046

3047 3048
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))
3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062
/*
 * 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);
	}
}
3063

3064
void preempt_count_add(int val)
L
Linus Torvalds 已提交
3065
{
3066
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3067 3068 3069
	/*
	 * Underflow?
	 */
3070 3071
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
3072
#endif
3073
	__preempt_count_add(val);
3074
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3075 3076 3077
	/*
	 * Spinlock count overflowing soon?
	 */
3078 3079
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
3080
#endif
3081
	preempt_latency_start(val);
L
Linus Torvalds 已提交
3082
}
3083
EXPORT_SYMBOL(preempt_count_add);
3084
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
3085

3086 3087 3088 3089 3090 3091 3092 3093 3094 3095
/*
 * 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());
}

3096
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
3097
{
3098
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
3099 3100 3101
	/*
	 * Underflow?
	 */
3102
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
3103
		return;
L
Linus Torvalds 已提交
3104 3105 3106
	/*
	 * Is the spinlock portion underflowing?
	 */
3107 3108 3109
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
3110
#endif
3111

3112
	preempt_latency_stop(val);
3113
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
3114
}
3115
EXPORT_SYMBOL(preempt_count_sub);
3116
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
3117

3118 3119 3120
#else
static inline void preempt_latency_start(int val) { }
static inline void preempt_latency_stop(int val) { }
L
Linus Torvalds 已提交
3121 3122 3123
#endif

/*
I
Ingo Molnar 已提交
3124
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
3125
 */
I
Ingo Molnar 已提交
3126
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
3127
{
3128 3129 3130
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3131 3132
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3133

I
Ingo Molnar 已提交
3134
	debug_show_held_locks(prev);
3135
	print_modules();
I
Ingo Molnar 已提交
3136 3137
	if (irqs_disabled())
		print_irqtrace_events(prev);
3138 3139 3140 3141 3142 3143 3144
#ifdef CONFIG_DEBUG_PREEMPT
	if (in_atomic_preempt_off()) {
		pr_err("Preemption disabled at:");
		print_ip_sym(current->preempt_disable_ip);
		pr_cont("\n");
	}
#endif
3145
	dump_stack();
3146
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3147
}
L
Linus Torvalds 已提交
3148

I
Ingo Molnar 已提交
3149 3150 3151 3152 3153
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3154
#ifdef CONFIG_SCHED_STACK_END_CHECK
3155
	BUG_ON(task_stack_end_corrupted(prev));
3156
#endif
3157

3158
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3159
		__schedule_bug(prev);
3160 3161
		preempt_count_set(PREEMPT_DISABLED);
	}
3162
	rcu_sleep_check();
I
Ingo Molnar 已提交
3163

L
Linus Torvalds 已提交
3164 3165
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3166
	schedstat_inc(this_rq(), sched_count);
I
Ingo Molnar 已提交
3167 3168 3169 3170 3171 3172
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3173
pick_next_task(struct rq *rq, struct task_struct *prev, struct pin_cookie cookie)
I
Ingo Molnar 已提交
3174
{
3175
	const struct sched_class *class = &fair_sched_class;
I
Ingo Molnar 已提交
3176
	struct task_struct *p;
L
Linus Torvalds 已提交
3177 3178

	/*
I
Ingo Molnar 已提交
3179 3180
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
3181
	 */
3182
	if (likely(prev->sched_class == class &&
3183
		   rq->nr_running == rq->cfs.h_nr_running)) {
3184
		p = fair_sched_class.pick_next_task(rq, prev, cookie);
3185 3186 3187 3188 3189
		if (unlikely(p == RETRY_TASK))
			goto again;

		/* assumes fair_sched_class->next == idle_sched_class */
		if (unlikely(!p))
3190
			p = idle_sched_class.pick_next_task(rq, prev, cookie);
3191 3192

		return p;
L
Linus Torvalds 已提交
3193 3194
	}

3195
again:
3196
	for_each_class(class) {
3197
		p = class->pick_next_task(rq, prev, cookie);
3198 3199 3200
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3201
			return p;
3202
		}
I
Ingo Molnar 已提交
3203
	}
3204 3205

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
3206
}
L
Linus Torvalds 已提交
3207

I
Ingo Molnar 已提交
3208
/*
3209
 * __schedule() is the main scheduler function.
3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243
 *
 * 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
3244
 *
3245
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3246
 */
3247
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3248 3249
{
	struct task_struct *prev, *next;
3250
	unsigned long *switch_count;
3251
	struct pin_cookie cookie;
I
Ingo Molnar 已提交
3252
	struct rq *rq;
3253
	int cpu;
I
Ingo Molnar 已提交
3254 3255 3256 3257 3258

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

3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269
	/*
	 * do_exit() calls schedule() with preemption disabled as an exception;
	 * however we must fix that up, otherwise the next task will see an
	 * inconsistent (higher) preempt count.
	 *
	 * It also avoids the below schedule_debug() test from complaining
	 * about this.
	 */
	if (unlikely(prev->state == TASK_DEAD))
		preempt_enable_no_resched_notrace();

I
Ingo Molnar 已提交
3270
	schedule_debug(prev);
L
Linus Torvalds 已提交
3271

3272
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3273
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3274

3275 3276 3277
	local_irq_disable();
	rcu_note_context_switch();

3278 3279 3280 3281 3282 3283
	/*
	 * 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();
3284
	raw_spin_lock(&rq->lock);
3285
	cookie = lockdep_pin_lock(&rq->lock);
L
Linus Torvalds 已提交
3286

3287 3288
	rq->clock_skip_update <<= 1; /* promote REQ to ACT */

3289
	switch_count = &prev->nivcsw;
3290
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3291
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3292
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3293
		} else {
3294 3295 3296
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
3297
			/*
3298 3299 3300
			 * 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
3301 3302 3303 3304
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

3305
				to_wakeup = wq_worker_sleeping(prev);
T
Tejun Heo 已提交
3306
				if (to_wakeup)
3307
					try_to_wake_up_local(to_wakeup, cookie);
T
Tejun Heo 已提交
3308 3309
			}
		}
I
Ingo Molnar 已提交
3310
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3311 3312
	}

3313
	if (task_on_rq_queued(prev))
3314 3315
		update_rq_clock(rq);

3316
	next = pick_next_task(rq, prev, cookie);
3317
	clear_tsk_need_resched(prev);
3318
	clear_preempt_need_resched();
3319
	rq->clock_skip_update = 0;
L
Linus Torvalds 已提交
3320 3321 3322 3323 3324 3325

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

3326
		trace_sched_switch(preempt, prev, next);
3327
		rq = context_switch(rq, prev, next, cookie); /* unlocks the rq */
3328
	} else {
3329
		lockdep_unpin_lock(&rq->lock, cookie);
3330
		raw_spin_unlock_irq(&rq->lock);
3331
	}
L
Linus Torvalds 已提交
3332

3333
	balance_callback(rq);
L
Linus Torvalds 已提交
3334
}
3335
STACK_FRAME_NON_STANDARD(__schedule); /* switch_to() */
3336

3337 3338
static inline void sched_submit_work(struct task_struct *tsk)
{
3339
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3340 3341 3342 3343 3344 3345 3346 3347 3348
		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);
}

3349
asmlinkage __visible void __sched schedule(void)
3350
{
3351 3352 3353
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3354
	do {
3355
		preempt_disable();
3356
		__schedule(false);
3357
		sched_preempt_enable_no_resched();
3358
	} while (need_resched());
3359
}
L
Linus Torvalds 已提交
3360 3361
EXPORT_SYMBOL(schedule);

3362
#ifdef CONFIG_CONTEXT_TRACKING
3363
asmlinkage __visible void __sched schedule_user(void)
3364 3365 3366 3367 3368 3369
{
	/*
	 * 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.
3370 3371
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3372
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3373
	 * too frequently to make sense yet.
3374
	 */
3375
	enum ctx_state prev_state = exception_enter();
3376
	schedule();
3377
	exception_exit(prev_state);
3378 3379 3380
}
#endif

3381 3382 3383 3384 3385 3386 3387
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3388
	sched_preempt_enable_no_resched();
3389 3390 3391 3392
	schedule();
	preempt_disable();
}

3393
static void __sched notrace preempt_schedule_common(void)
3394 3395
{
	do {
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408
		/*
		 * 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.
		 */
3409
		preempt_disable_notrace();
3410
		preempt_latency_start(1);
3411
		__schedule(true);
3412
		preempt_latency_stop(1);
3413
		preempt_enable_no_resched_notrace();
3414 3415 3416 3417 3418 3419 3420 3421

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

L
Linus Torvalds 已提交
3422 3423
#ifdef CONFIG_PREEMPT
/*
3424
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3425
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3426 3427
 * occur there and call schedule directly.
 */
3428
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3429 3430 3431
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3432
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3433
	 */
3434
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3435 3436
		return;

3437
	preempt_schedule_common();
L
Linus Torvalds 已提交
3438
}
3439
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3440
EXPORT_SYMBOL(preempt_schedule);
3441 3442

/**
3443
 * preempt_schedule_notrace - preempt_schedule called by tracing
3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455
 *
 * 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.
 */
3456
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3457 3458 3459 3460 3461 3462 3463
{
	enum ctx_state prev_ctx;

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

	do {
3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476
		/*
		 * 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.
		 */
3477
		preempt_disable_notrace();
3478
		preempt_latency_start(1);
3479 3480 3481 3482 3483 3484
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3485
		__schedule(true);
3486 3487
		exception_exit(prev_ctx);

3488
		preempt_latency_stop(1);
3489
		preempt_enable_no_resched_notrace();
3490 3491
	} while (need_resched());
}
3492
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3493

3494
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3495 3496

/*
3497
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3498 3499 3500 3501
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3502
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3503
{
3504
	enum ctx_state prev_state;
3505

3506
	/* Catch callers which need to be fixed */
3507
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3508

3509 3510
	prev_state = exception_enter();

3511
	do {
3512
		preempt_disable();
3513
		local_irq_enable();
3514
		__schedule(true);
3515
		local_irq_disable();
3516
		sched_preempt_enable_no_resched();
3517
	} while (need_resched());
3518 3519

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3520 3521
}

P
Peter Zijlstra 已提交
3522
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3523
			  void *key)
L
Linus Torvalds 已提交
3524
{
P
Peter Zijlstra 已提交
3525
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3526 3527 3528
}
EXPORT_SYMBOL(default_wake_function);

3529 3530 3531 3532 3533 3534 3535 3536 3537 3538
#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().
 *
3539 3540
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3541
 */
3542
void rt_mutex_setprio(struct task_struct *p, int prio)
3543
{
3544
	int oldprio, queued, running, queue_flag = DEQUEUE_SAVE | DEQUEUE_MOVE;
3545
	const struct sched_class *prev_class;
3546 3547
	struct rq_flags rf;
	struct rq *rq;
3548

3549
	BUG_ON(prio > MAX_PRIO);
3550

3551
	rq = __task_rq_lock(p, &rf);
3552

3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570
	/*
	 * 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;
	}

3571
	trace_sched_pi_setprio(p, prio);
3572
	oldprio = p->prio;
3573 3574 3575 3576

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

3577
	prev_class = p->sched_class;
3578
	queued = task_on_rq_queued(p);
3579
	running = task_current(rq, p);
3580
	if (queued)
3581
		dequeue_task(rq, p, queue_flag);
3582
	if (running)
3583
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3584

3585 3586 3587 3588 3589 3590 3591 3592 3593 3594
	/*
	 * 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)) {
3595 3596 3597
		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))) {
3598
			p->dl.dl_boosted = 1;
3599
			queue_flag |= ENQUEUE_REPLENISH;
3600 3601
		} else
			p->dl.dl_boosted = 0;
3602
		p->sched_class = &dl_sched_class;
3603 3604 3605 3606
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3607
			queue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3608
		p->sched_class = &rt_sched_class;
3609 3610 3611
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3612 3613
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3614
		p->sched_class = &fair_sched_class;
3615
	}
I
Ingo Molnar 已提交
3616

3617 3618
	p->prio = prio;

3619 3620
	if (running)
		p->sched_class->set_curr_task(rq);
3621
	if (queued)
3622
		enqueue_task(rq, p, queue_flag);
3623

P
Peter Zijlstra 已提交
3624
	check_class_changed(rq, p, prev_class, oldprio);
3625
out_unlock:
3626
	preempt_disable(); /* avoid rq from going away on us */
3627
	__task_rq_unlock(rq, &rf);
3628 3629 3630

	balance_callback(rq);
	preempt_enable();
3631 3632
}
#endif
3633

3634
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3635
{
3636
	int old_prio, delta, queued;
3637
	struct rq_flags rf;
3638
	struct rq *rq;
L
Linus Torvalds 已提交
3639

3640
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3641 3642 3643 3644 3645
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
3646
	rq = task_rq_lock(p, &rf);
L
Linus Torvalds 已提交
3647 3648 3649 3650
	/*
	 * 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
3651
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3652
	 */
3653
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3654 3655 3656
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3657 3658
	queued = task_on_rq_queued(p);
	if (queued)
3659
		dequeue_task(rq, p, DEQUEUE_SAVE);
L
Linus Torvalds 已提交
3660 3661

	p->static_prio = NICE_TO_PRIO(nice);
3662
	set_load_weight(p);
3663 3664 3665
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3666

3667
	if (queued) {
3668
		enqueue_task(rq, p, ENQUEUE_RESTORE);
L
Linus Torvalds 已提交
3669
		/*
3670 3671
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3672
		 */
3673
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3674
			resched_curr(rq);
L
Linus Torvalds 已提交
3675 3676
	}
out_unlock:
3677
	task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
3678 3679 3680
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3681 3682 3683 3684 3685
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3686
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3687
{
3688
	/* convert nice value [19,-20] to rlimit style value [1,40] */
3689
	int nice_rlim = nice_to_rlimit(nice);
3690

3691
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3692 3693 3694
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3695 3696 3697 3698 3699 3700 3701 3702 3703
#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.
 */
3704
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3705
{
3706
	long nice, retval;
L
Linus Torvalds 已提交
3707 3708 3709 3710 3711 3712

	/*
	 * 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.
	 */
3713
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3714
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3715

3716
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3717 3718 3719
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733
	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.
 *
3734
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3735 3736 3737
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3738
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3739 3740 3741 3742 3743 3744 3745
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
3746 3747
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3748 3749 3750
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764
	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
3765 3766 3767 3768 3769
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
3770 3771
 *
 * Return: The idle task for the cpu @cpu.
L
Linus Torvalds 已提交
3772
 */
3773
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3774 3775 3776 3777 3778 3779 3780
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3781 3782
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3783
 */
A
Alexey Dobriyan 已提交
3784
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3785
{
3786
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3787 3788
}

3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803
/*
 * 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;
3804
	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
3805
	dl_se->flags = attr->sched_flags;
3806
	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826

	/*
	 * 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.
	 */
3827 3828
}

3829 3830 3831 3832 3833 3834
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

3835 3836
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
3837
{
3838 3839
	int policy = attr->sched_policy;

3840
	if (policy == SETPARAM_POLICY)
3841 3842
		policy = p->policy;

L
Linus Torvalds 已提交
3843
	p->policy = policy;
3844

3845 3846
	if (dl_policy(policy))
		__setparam_dl(p, attr);
3847
	else if (fair_policy(policy))
3848 3849
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

3850 3851 3852 3853 3854 3855
	/*
	 * __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;
3856
	p->normal_prio = normal_prio(p);
3857 3858
	set_load_weight(p);
}
3859

3860 3861
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
3862
			   const struct sched_attr *attr, bool keep_boost)
3863 3864
{
	__setscheduler_params(p, attr);
3865

3866
	/*
3867 3868
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
3869
	 */
3870 3871 3872 3873
	if (keep_boost)
		p->prio = rt_mutex_get_effective_prio(p, normal_prio(p));
	else
		p->prio = normal_prio(p);
3874

3875 3876 3877
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
3878 3879 3880
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
3881
}
3882 3883 3884 3885 3886 3887 3888 3889 3890

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;
3891
	attr->sched_period = dl_se->dl_period;
3892 3893 3894 3895 3896 3897
	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
3898
 * than the runtime, as well as the period of being zero or
3899
 * greater than deadline. Furthermore, we have to be sure that
3900 3901 3902 3903
 * 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).
3904 3905 3906 3907
 */
static bool
__checkparam_dl(const struct sched_attr *attr)
{
3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933
	/* 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;
3934 3935
}

3936 3937 3938 3939 3940 3941 3942 3943 3944 3945
/*
 * 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);
3946 3947
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
3948 3949 3950 3951
	rcu_read_unlock();
	return match;
}

3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965
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;
}

3966 3967
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
3968
				bool user, bool pi)
L
Linus Torvalds 已提交
3969
{
3970 3971
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
3972
	int retval, oldprio, oldpolicy = -1, queued, running;
3973
	int new_effective_prio, policy = attr->sched_policy;
3974
	const struct sched_class *prev_class;
3975
	struct rq_flags rf;
3976
	int reset_on_fork;
3977
	int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE;
3978
	struct rq *rq;
L
Linus Torvalds 已提交
3979

3980 3981
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
3982 3983
recheck:
	/* double check policy once rq lock held */
3984 3985
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
3986
		policy = oldpolicy = p->policy;
3987
	} else {
3988
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
3989

3990
		if (!valid_policy(policy))
3991 3992 3993
			return -EINVAL;
	}

3994 3995 3996
	if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK))
		return -EINVAL;

L
Linus Torvalds 已提交
3997 3998
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
3999 4000
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
4001
	 */
4002
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
4003
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
4004
		return -EINVAL;
4005 4006
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
4007 4008
		return -EINVAL;

4009 4010 4011
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
4012
	if (user && !capable(CAP_SYS_NICE)) {
4013
		if (fair_policy(policy)) {
4014
			if (attr->sched_nice < task_nice(p) &&
4015
			    !can_nice(p, attr->sched_nice))
4016 4017 4018
				return -EPERM;
		}

4019
		if (rt_policy(policy)) {
4020 4021
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
4022 4023 4024 4025 4026 4027

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

			/* can't increase priority */
4028 4029
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
4030 4031
				return -EPERM;
		}
4032

4033 4034 4035 4036 4037 4038 4039 4040 4041
		 /*
		  * 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
4042
		/*
4043 4044
		 * 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
4045
		 */
4046
		if (idle_policy(p->policy) && !idle_policy(policy)) {
4047
			if (!can_nice(p, task_nice(p)))
4048 4049
				return -EPERM;
		}
4050

4051
		/* can't change other user's priorities */
4052
		if (!check_same_owner(p))
4053
			return -EPERM;
4054 4055 4056 4057

		/* Normal users shall not reset the sched_reset_on_fork flag */
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
4058
	}
L
Linus Torvalds 已提交
4059

4060
	if (user) {
4061
		retval = security_task_setscheduler(p);
4062 4063 4064 4065
		if (retval)
			return retval;
	}

4066 4067 4068
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
4069
	 *
L
Lucas De Marchi 已提交
4070
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
4071 4072
	 * runqueue lock must be held.
	 */
4073
	rq = task_rq_lock(p, &rf);
4074

4075 4076 4077 4078
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
4079
		task_rq_unlock(rq, p, &rf);
4080 4081 4082
		return -EINVAL;
	}

4083
	/*
4084 4085
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
4086
	 */
4087
	if (unlikely(policy == p->policy)) {
4088
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
4089 4090 4091
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
4092
		if (dl_policy(policy) && dl_param_changed(p, attr))
4093
			goto change;
4094

4095
		p->sched_reset_on_fork = reset_on_fork;
4096
		task_rq_unlock(rq, p, &rf);
4097 4098
		return 0;
	}
4099
change:
4100

4101
	if (user) {
4102
#ifdef CONFIG_RT_GROUP_SCHED
4103 4104 4105 4106 4107
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
4108 4109
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
4110
			task_rq_unlock(rq, p, &rf);
4111 4112 4113
			return -EPERM;
		}
#endif
4114 4115 4116 4117 4118 4119 4120 4121 4122
#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.
			 */
4123 4124
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
4125
				task_rq_unlock(rq, p, &rf);
4126 4127 4128 4129 4130
				return -EPERM;
			}
		}
#endif
	}
4131

L
Linus Torvalds 已提交
4132 4133 4134
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
4135
		task_rq_unlock(rq, p, &rf);
L
Linus Torvalds 已提交
4136 4137
		goto recheck;
	}
4138 4139 4140 4141 4142 4143

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
4144
	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
4145
		task_rq_unlock(rq, p, &rf);
4146 4147 4148
		return -EBUSY;
	}

4149 4150 4151
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

4152 4153 4154 4155 4156 4157 4158 4159 4160
	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);
4161 4162
		if (new_effective_prio == oldprio)
			queue_flags &= ~DEQUEUE_MOVE;
4163 4164
	}

4165
	queued = task_on_rq_queued(p);
4166
	running = task_current(rq, p);
4167
	if (queued)
4168
		dequeue_task(rq, p, queue_flags);
4169
	if (running)
4170
		put_prev_task(rq, p);
4171

4172
	prev_class = p->sched_class;
4173
	__setscheduler(rq, p, attr, pi);
4174

4175 4176
	if (running)
		p->sched_class->set_curr_task(rq);
4177
	if (queued) {
4178 4179 4180 4181
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4182 4183
		if (oldprio < p->prio)
			queue_flags |= ENQUEUE_HEAD;
4184

4185
		enqueue_task(rq, p, queue_flags);
4186
	}
4187

P
Peter Zijlstra 已提交
4188
	check_class_changed(rq, p, prev_class, oldprio);
4189
	preempt_disable(); /* avoid rq from going away on us */
4190
	task_rq_unlock(rq, p, &rf);
4191

4192 4193
	if (pi)
		rt_mutex_adjust_pi(p);
4194

4195 4196 4197 4198 4199
	/*
	 * Run balance callbacks after we've adjusted the PI chain.
	 */
	balance_callback(rq);
	preempt_enable();
4200

L
Linus Torvalds 已提交
4201 4202
	return 0;
}
4203

4204 4205 4206 4207 4208 4209 4210 4211 4212
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),
	};

4213 4214
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4215 4216 4217 4218 4219
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4220
	return __sched_setscheduler(p, &attr, check, true);
4221
}
4222 4223 4224 4225 4226 4227
/**
 * 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.
 *
4228 4229
 * Return: 0 on success. An error code otherwise.
 *
4230 4231 4232
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4233
		       const struct sched_param *param)
4234
{
4235
	return _sched_setscheduler(p, policy, param, true);
4236
}
L
Linus Torvalds 已提交
4237 4238
EXPORT_SYMBOL_GPL(sched_setscheduler);

4239 4240
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4241
	return __sched_setscheduler(p, attr, true, true);
4242 4243 4244
}
EXPORT_SYMBOL_GPL(sched_setattr);

4245 4246 4247 4248 4249 4250 4251 4252 4253 4254
/**
 * 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.
4255 4256
 *
 * Return: 0 on success. An error code otherwise.
4257 4258
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4259
			       const struct sched_param *param)
4260
{
4261
	return _sched_setscheduler(p, policy, param, false);
4262
}
4263
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4264

I
Ingo Molnar 已提交
4265 4266
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4267 4268 4269
{
	struct sched_param lparam;
	struct task_struct *p;
4270
	int retval;
L
Linus Torvalds 已提交
4271 4272 4273 4274 4275

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4276 4277 4278

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4279
	p = find_process_by_pid(pid);
4280 4281 4282
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4283

L
Linus Torvalds 已提交
4284 4285 4286
	return retval;
}

4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348
/*
 * 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?
	 */
4349
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4350

4351
	return 0;
4352 4353 4354

err_size:
	put_user(sizeof(*attr), &uattr->size);
4355
	return -E2BIG;
4356 4357
}

L
Linus Torvalds 已提交
4358 4359 4360 4361 4362
/**
 * 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.
4363 4364
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4365
 */
4366 4367
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
4368
{
4369 4370 4371 4372
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4373 4374 4375 4376 4377 4378 4379
	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.
4380 4381
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4382
 */
4383
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4384
{
4385
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4386 4387
}

4388 4389 4390
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4391
 * @uattr: structure containing the extended parameters.
4392
 * @flags: for future extension.
4393
 */
4394 4395
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4396 4397 4398 4399 4400
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4401
	if (!uattr || pid < 0 || flags)
4402 4403
		return -EINVAL;

4404 4405 4406
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4407

4408
	if ((int)attr.sched_policy < 0)
4409
		return -EINVAL;
4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420

	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
4421 4422 4423
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4424 4425 4426
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4427
 */
4428
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4429
{
4430
	struct task_struct *p;
4431
	int retval;
L
Linus Torvalds 已提交
4432 4433

	if (pid < 0)
4434
		return -EINVAL;
L
Linus Torvalds 已提交
4435 4436

	retval = -ESRCH;
4437
	rcu_read_lock();
L
Linus Torvalds 已提交
4438 4439 4440 4441
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4442 4443
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4444
	}
4445
	rcu_read_unlock();
L
Linus Torvalds 已提交
4446 4447 4448 4449
	return retval;
}

/**
4450
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4451 4452
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4453 4454 4455
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4456
 */
4457
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4458
{
4459
	struct sched_param lp = { .sched_priority = 0 };
4460
	struct task_struct *p;
4461
	int retval;
L
Linus Torvalds 已提交
4462 4463

	if (!param || pid < 0)
4464
		return -EINVAL;
L
Linus Torvalds 已提交
4465

4466
	rcu_read_lock();
L
Linus Torvalds 已提交
4467 4468 4469 4470 4471 4472 4473 4474 4475
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4476 4477
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4478
	rcu_read_unlock();
L
Linus Torvalds 已提交
4479 4480 4481 4482 4483 4484 4485 4486 4487

	/*
	 * 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:
4488
	rcu_read_unlock();
L
Linus Torvalds 已提交
4489 4490 4491
	return retval;
}

4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514
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)
4515
				return -EFBIG;
4516 4517 4518 4519 4520
		}

		attr->size = usize;
	}

4521
	ret = copy_to_user(uattr, attr, attr->size);
4522 4523 4524
	if (ret)
		return -EFAULT;

4525
	return 0;
4526 4527 4528
}

/**
4529
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4530
 * @pid: the pid in question.
J
Juri Lelli 已提交
4531
 * @uattr: structure containing the extended parameters.
4532
 * @size: sizeof(attr) for fwd/bwd comp.
4533
 * @flags: for future extension.
4534
 */
4535 4536
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4537 4538 4539 4540 4541 4542 4543 4544
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4545
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558
		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;
4559 4560
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4561 4562 4563
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4564 4565
		attr.sched_priority = p->rt_priority;
	else
4566
		attr.sched_nice = task_nice(p);
4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4578
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4579
{
4580
	cpumask_var_t cpus_allowed, new_mask;
4581 4582
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4583

4584
	rcu_read_lock();
L
Linus Torvalds 已提交
4585 4586 4587

	p = find_process_by_pid(pid);
	if (!p) {
4588
		rcu_read_unlock();
L
Linus Torvalds 已提交
4589 4590 4591
		return -ESRCH;
	}

4592
	/* Prevent p going away */
L
Linus Torvalds 已提交
4593
	get_task_struct(p);
4594
	rcu_read_unlock();
L
Linus Torvalds 已提交
4595

4596 4597 4598 4599
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4600 4601 4602 4603 4604 4605 4606 4607
	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
4608
	retval = -EPERM;
E
Eric W. Biederman 已提交
4609 4610 4611 4612
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4613
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4614 4615 4616
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4617

4618
	retval = security_task_setscheduler(p);
4619
	if (retval)
4620
		goto out_free_new_mask;
4621

4622 4623 4624 4625

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

4626 4627 4628 4629 4630 4631 4632
	/*
	 * 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
4633 4634 4635
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4636
			retval = -EBUSY;
4637
			rcu_read_unlock();
4638
			goto out_free_new_mask;
4639
		}
4640
		rcu_read_unlock();
4641 4642
	}
#endif
P
Peter Zijlstra 已提交
4643
again:
4644
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4645

P
Paul Menage 已提交
4646
	if (!retval) {
4647 4648
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4649 4650 4651 4652 4653
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4654
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4655 4656 4657
			goto again;
		}
	}
4658
out_free_new_mask:
4659 4660 4661 4662
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4663 4664 4665 4666 4667
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4668
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4669
{
4670 4671 4672 4673 4674
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4675 4676 4677 4678 4679 4680 4681 4682
	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
4683 4684
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4685
 */
4686 4687
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4688
{
4689
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4690 4691
	int retval;

4692 4693
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4694

4695 4696 4697 4698 4699
	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
4700 4701
}

4702
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4703
{
4704
	struct task_struct *p;
4705
	unsigned long flags;
L
Linus Torvalds 已提交
4706 4707
	int retval;

4708
	rcu_read_lock();
L
Linus Torvalds 已提交
4709 4710 4711 4712 4713 4714

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

4715 4716 4717 4718
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4719
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4720
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4721
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4722 4723

out_unlock:
4724
	rcu_read_unlock();
L
Linus Torvalds 已提交
4725

4726
	return retval;
L
Linus Torvalds 已提交
4727 4728 4729 4730 4731 4732 4733
}

/**
 * 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
4734 4735
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4736
 */
4737 4738
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4739 4740
{
	int ret;
4741
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4742

A
Anton Blanchard 已提交
4743
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4744 4745
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4746 4747
		return -EINVAL;

4748 4749
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4750

4751 4752
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4753
		size_t retlen = min_t(size_t, len, cpumask_size());
4754 4755

		if (copy_to_user(user_mask_ptr, mask, retlen))
4756 4757
			ret = -EFAULT;
		else
4758
			ret = retlen;
4759 4760
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4761

4762
	return ret;
L
Linus Torvalds 已提交
4763 4764 4765 4766 4767
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4768 4769
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4770 4771
 *
 * Return: 0.
L
Linus Torvalds 已提交
4772
 */
4773
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4774
{
4775
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
4776

4777
	schedstat_inc(rq, yld_count);
4778
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4779 4780 4781 4782 4783 4784

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
4785
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4786
	do_raw_spin_unlock(&rq->lock);
4787
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4788 4789 4790 4791 4792 4793

	schedule();

	return 0;
}

4794
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4795
{
4796
	if (should_resched(0)) {
4797
		preempt_schedule_common();
L
Linus Torvalds 已提交
4798 4799 4800 4801
		return 1;
	}
	return 0;
}
4802
EXPORT_SYMBOL(_cond_resched);
L
Linus Torvalds 已提交
4803 4804

/*
4805
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4806 4807
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4808
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4809 4810 4811
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4812
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4813
{
4814
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4815 4816
	int ret = 0;

4817 4818
	lockdep_assert_held(lock);

4819
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4820
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4821
		if (resched)
4822
			preempt_schedule_common();
N
Nick Piggin 已提交
4823 4824
		else
			cpu_relax();
J
Jan Kara 已提交
4825
		ret = 1;
L
Linus Torvalds 已提交
4826 4827
		spin_lock(lock);
	}
J
Jan Kara 已提交
4828
	return ret;
L
Linus Torvalds 已提交
4829
}
4830
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4831

4832
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4833 4834 4835
{
	BUG_ON(!in_softirq());

4836
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4837
		local_bh_enable();
4838
		preempt_schedule_common();
L
Linus Torvalds 已提交
4839 4840 4841 4842 4843
		local_bh_disable();
		return 1;
	}
	return 0;
}
4844
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4845 4846 4847 4848

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866
 * 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
4867 4868 4869 4870 4871 4872 4873 4874
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4875 4876 4877 4878
/**
 * 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
4879 4880
 * @p: target task
 * @preempt: whether task preemption is allowed or not
4881 4882 4883 4884
 *
 * 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.
 *
4885
 * Return:
4886 4887 4888
 *	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.
4889
 */
4890
int __sched yield_to(struct task_struct *p, bool preempt)
4891 4892 4893 4894
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
4895
	int yielded = 0;
4896 4897 4898 4899 4900 4901

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
4902 4903 4904 4905 4906 4907 4908 4909 4910
	/*
	 * 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;
	}

4911
	double_rq_lock(rq, p_rq);
4912
	if (task_rq(p) != p_rq) {
4913 4914 4915 4916 4917
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
4918
		goto out_unlock;
4919 4920

	if (curr->sched_class != p->sched_class)
4921
		goto out_unlock;
4922 4923

	if (task_running(p_rq, p) || p->state)
4924
		goto out_unlock;
4925 4926

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
4927
	if (yielded) {
4928
		schedstat_inc(rq, yld_count);
4929 4930 4931 4932 4933
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
4934
			resched_curr(p_rq);
4935
	}
4936

4937
out_unlock:
4938
	double_rq_unlock(rq, p_rq);
4939
out_irq:
4940 4941
	local_irq_restore(flags);

4942
	if (yielded > 0)
4943 4944 4945 4946 4947 4948
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
4949
/*
I
Ingo Molnar 已提交
4950
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
4951 4952 4953 4954
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
4955 4956
	int old_iowait = current->in_iowait;
	struct rq *rq;
L
Linus Torvalds 已提交
4957 4958
	long ret;

4959
	current->in_iowait = 1;
4960
	blk_schedule_flush_plug(current);
4961

4962
	delayacct_blkio_start();
4963
	rq = raw_rq();
L
Linus Torvalds 已提交
4964 4965
	atomic_inc(&rq->nr_iowait);
	ret = schedule_timeout(timeout);
4966
	current->in_iowait = old_iowait;
L
Linus Torvalds 已提交
4967
	atomic_dec(&rq->nr_iowait);
4968
	delayacct_blkio_end();
4969

L
Linus Torvalds 已提交
4970 4971
	return ret;
}
4972
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
4973 4974 4975 4976 4977

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
4978 4979 4980
 * 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
4981
 */
4982
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
4983 4984 4985 4986 4987 4988 4989 4990
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
4991
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
4992
	case SCHED_NORMAL:
4993
	case SCHED_BATCH:
I
Ingo Molnar 已提交
4994
	case SCHED_IDLE:
L
Linus Torvalds 已提交
4995 4996 4997 4998 4999 5000 5001 5002 5003 5004
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
5005 5006 5007
 * 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
5008
 */
5009
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
5010 5011 5012 5013 5014 5015 5016 5017
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
5018
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
5019
	case SCHED_NORMAL:
5020
	case SCHED_BATCH:
I
Ingo Molnar 已提交
5021
	case SCHED_IDLE:
L
Linus Torvalds 已提交
5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033
		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.
5034 5035 5036
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
L
Linus Torvalds 已提交
5037
 */
5038
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
5039
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
5040
{
5041
	struct task_struct *p;
D
Dmitry Adamushko 已提交
5042
	unsigned int time_slice;
5043 5044
	struct rq_flags rf;
	struct timespec t;
5045
	struct rq *rq;
5046
	int retval;
L
Linus Torvalds 已提交
5047 5048

	if (pid < 0)
5049
		return -EINVAL;
L
Linus Torvalds 已提交
5050 5051

	retval = -ESRCH;
5052
	rcu_read_lock();
L
Linus Torvalds 已提交
5053 5054 5055 5056 5057 5058 5059 5060
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

5061
	rq = task_rq_lock(p, &rf);
5062 5063 5064
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
5065
	task_rq_unlock(rq, p, &rf);
D
Dmitry Adamushko 已提交
5066

5067
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
5068
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
5069 5070
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
5071

L
Linus Torvalds 已提交
5072
out_unlock:
5073
	rcu_read_unlock();
L
Linus Torvalds 已提交
5074 5075 5076
	return retval;
}

5077
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
5078

5079
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
5080 5081
{
	unsigned long free = 0;
5082
	int ppid;
5083
	unsigned long state = p->state;
L
Linus Torvalds 已提交
5084

5085 5086
	if (state)
		state = __ffs(state) + 1;
5087
	printk(KERN_INFO "%-15.15s %c", p->comm,
5088
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
5089
#if BITS_PER_LONG == 32
L
Linus Torvalds 已提交
5090
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5091
		printk(KERN_CONT " running  ");
L
Linus Torvalds 已提交
5092
	else
P
Peter Zijlstra 已提交
5093
		printk(KERN_CONT " %08lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5094 5095
#else
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
5096
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
5097
	else
P
Peter Zijlstra 已提交
5098
		printk(KERN_CONT " %016lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
5099 5100
#endif
#ifdef CONFIG_DEBUG_STACK_USAGE
5101
	free = stack_not_used(p);
L
Linus Torvalds 已提交
5102
#endif
5103
	ppid = 0;
5104
	rcu_read_lock();
5105 5106
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
5107
	rcu_read_unlock();
P
Peter Zijlstra 已提交
5108
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
5109
		task_pid_nr(p), ppid,
5110
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
5111

5112
	print_worker_info(KERN_INFO, p);
5113
	show_stack(p, NULL);
L
Linus Torvalds 已提交
5114 5115
}

I
Ingo Molnar 已提交
5116
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
5117
{
5118
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
5119

5120
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
5121 5122
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
5123
#else
P
Peter Zijlstra 已提交
5124 5125
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
5126
#endif
5127
	rcu_read_lock();
5128
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
5129 5130
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
5131
		 * console might take a lot of time:
L
Linus Torvalds 已提交
5132 5133
		 */
		touch_nmi_watchdog();
I
Ingo Molnar 已提交
5134
		if (!state_filter || (p->state & state_filter))
5135
			sched_show_task(p);
5136
	}
L
Linus Torvalds 已提交
5137

5138 5139
	touch_all_softlockup_watchdogs();

I
Ingo Molnar 已提交
5140
#ifdef CONFIG_SCHED_DEBUG
5141 5142
	if (!state_filter)
		sysrq_sched_debug_show();
I
Ingo Molnar 已提交
5143
#endif
5144
	rcu_read_unlock();
I
Ingo Molnar 已提交
5145 5146 5147
	/*
	 * Only show locks if all tasks are dumped:
	 */
5148
	if (!state_filter)
I
Ingo Molnar 已提交
5149
		debug_show_all_locks();
L
Linus Torvalds 已提交
5150 5151
}

5152
void init_idle_bootup_task(struct task_struct *idle)
I
Ingo Molnar 已提交
5153
{
I
Ingo Molnar 已提交
5154
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
5155 5156
}

5157 5158 5159 5160 5161 5162 5163 5164
/**
 * 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.
 */
5165
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5166
{
5167
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5168 5169
	unsigned long flags;

5170 5171
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5172

5173
	__sched_fork(0, idle);
5174
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5175 5176
	idle->se.exec_start = sched_clock();

5177 5178
	kasan_unpoison_task_stack(idle);

5179 5180 5181 5182 5183 5184 5185 5186 5187
#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
5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198
	/*
	 * 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
5199
	__set_task_cpu(idle, cpu);
5200
	rcu_read_unlock();
L
Linus Torvalds 已提交
5201 5202

	rq->curr = rq->idle = idle;
5203
	idle->on_rq = TASK_ON_RQ_QUEUED;
5204
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5205
	idle->on_cpu = 1;
5206
#endif
5207 5208
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5209 5210

	/* Set the preempt count _outside_ the spinlocks! */
5211
	init_idle_preempt_count(idle, cpu);
5212

I
Ingo Molnar 已提交
5213 5214 5215 5216
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5217
	ftrace_graph_init_idle_task(idle, cpu);
5218
	vtime_init_idle(idle, cpu);
5219
#ifdef CONFIG_SMP
5220 5221
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5222 5223
}

5224 5225 5226 5227 5228 5229 5230
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;

5231 5232 5233
	if (!cpumask_weight(cur))
		return ret;

5234
	rcu_read_lock_sched();
5235 5236 5237 5238 5239 5240 5241 5242
	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);
5243
	rcu_read_unlock_sched();
5244 5245 5246 5247

	return ret;
}

5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267 5268 5269 5270 5271
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);
5272
		struct dl_bw *dl_b;
5273 5274 5275 5276
		bool overflow;
		int cpus;
		unsigned long flags;

5277 5278
		rcu_read_lock_sched();
		dl_b = dl_bw_of(dest_cpu);
5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293
		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);
5294
		rcu_read_unlock_sched();
5295 5296 5297 5298 5299 5300 5301

	}
#endif
out:
	return ret;
}

L
Linus Torvalds 已提交
5302 5303
#ifdef CONFIG_SMP

5304 5305
static bool sched_smp_initialized __read_mostly;

5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320
#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 */

5321
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5322 5323
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5324 5325 5326 5327 5328 5329 5330

/*
 * 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)
{
5331
	bool queued, running;
5332 5333
	struct rq_flags rf;
	struct rq *rq;
5334

5335
	rq = task_rq_lock(p, &rf);
5336
	queued = task_on_rq_queued(p);
5337 5338
	running = task_current(rq, p);

5339
	if (queued)
5340
		dequeue_task(rq, p, DEQUEUE_SAVE);
5341
	if (running)
5342
		put_prev_task(rq, p);
5343 5344 5345 5346 5347

	p->numa_preferred_nid = nid;

	if (running)
		p->sched_class->set_curr_task(rq);
5348
	if (queued)
5349
		enqueue_task(rq, p, ENQUEUE_RESTORE);
5350
	task_rq_unlock(rq, p, &rf);
5351
}
P
Peter Zijlstra 已提交
5352
#endif /* CONFIG_NUMA_BALANCING */
5353

L
Linus Torvalds 已提交
5354
#ifdef CONFIG_HOTPLUG_CPU
5355
/*
5356 5357
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
5358
 */
5359
void idle_task_exit(void)
L
Linus Torvalds 已提交
5360
{
5361
	struct mm_struct *mm = current->active_mm;
5362

5363
	BUG_ON(cpu_online(smp_processor_id()));
5364

5365
	if (mm != &init_mm) {
5366
		switch_mm_irqs_off(mm, &init_mm, current);
5367 5368
		finish_arch_post_lock_switch();
	}
5369
	mmdrop(mm);
L
Linus Torvalds 已提交
5370 5371 5372
}

/*
5373 5374 5375 5376 5377
 * 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
 * nr_active count is stable.
 *
 * Also see the comment "Global load-average calculations".
L
Linus Torvalds 已提交
5378
 */
5379
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5380
{
5381 5382 5383
	long delta = calc_load_fold_active(rq);
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5384 5385
}

5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401
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,
};

5402
/*
5403 5404 5405 5406 5407 5408
 * 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
5409
 */
5410
static void migrate_tasks(struct rq *dead_rq)
L
Linus Torvalds 已提交
5411
{
5412
	struct rq *rq = dead_rq;
5413
	struct task_struct *next, *stop = rq->stop;
5414
	struct pin_cookie cookie;
5415
	int dest_cpu;
L
Linus Torvalds 已提交
5416 5417

	/*
5418 5419 5420 5421 5422 5423 5424
	 * 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
5425
	 */
5426
	rq->stop = NULL;
5427

5428 5429 5430 5431 5432 5433 5434
	/*
	 * 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);

5435
	for (;;) {
5436 5437 5438 5439 5440
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5441
			break;
5442

5443
		/*
W
Wanpeng Li 已提交
5444
		 * pick_next_task assumes pinned rq->lock.
5445
		 */
5446 5447
		cookie = lockdep_pin_lock(&rq->lock);
		next = pick_next_task(rq, &fake_task, cookie);
5448
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
5449
		next->sched_class->put_prev_task(rq, next);
5450

W
Wanpeng Li 已提交
5451 5452 5453 5454 5455 5456 5457 5458 5459
		/*
		 * 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.
		 */
5460
		lockdep_unpin_lock(&rq->lock, cookie);
W
Wanpeng Li 已提交
5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474
		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;
		}

5475
		/* Find suitable destination for @next, with force if needed. */
5476
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5477

5478 5479 5480 5481 5482 5483
		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
5484
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5485
	}
5486

5487
	rq->stop = stop;
5488
}
L
Linus Torvalds 已提交
5489 5490
#endif /* CONFIG_HOTPLUG_CPU */

5491 5492 5493 5494 5495
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

5496
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515
		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);
		}

5516
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5517 5518 5519 5520
		rq->online = 0;
	}
}

5521
static void set_cpu_rq_start_time(unsigned int cpu)
L
Linus Torvalds 已提交
5522
{
5523
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5524

5525 5526 5527
	rq->age_stamp = sched_clock_cpu(cpu);
}

5528 5529
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

5530
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
5531

5532
static __read_mostly int sched_debug_enabled;
5533

5534
static int __init sched_debug_setup(char *str)
5535
{
5536
	sched_debug_enabled = 1;
5537 5538 5539

	return 0;
}
5540 5541 5542 5543 5544 5545
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
5546

5547
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
5548
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
5549
{
I
Ingo Molnar 已提交
5550
	struct sched_group *group = sd->groups;
L
Linus Torvalds 已提交
5551

5552
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
5553 5554 5555 5556

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

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
5557
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
5558
		if (sd->parent)
P
Peter Zijlstra 已提交
5559 5560
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
5561
		return -1;
N
Nick Piggin 已提交
5562 5563
	}

5564 5565
	printk(KERN_CONT "span %*pbl level %s\n",
	       cpumask_pr_args(sched_domain_span(sd)), sd->name);
I
Ingo Molnar 已提交
5566

5567
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
5568 5569
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
5570
	}
5571
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5572 5573
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
5574
	}
L
Linus Torvalds 已提交
5575

I
Ingo Molnar 已提交
5576
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
5577
	do {
I
Ingo Molnar 已提交
5578
		if (!group) {
P
Peter Zijlstra 已提交
5579 5580
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
5581 5582 5583
			break;
		}

5584
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5585 5586
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
5587 5588
			break;
		}
L
Linus Torvalds 已提交
5589

5590 5591
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5592 5593
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
5594 5595
			break;
		}
L
Linus Torvalds 已提交
5596

5597
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
5598

5599 5600
		printk(KERN_CONT " %*pbl",
		       cpumask_pr_args(sched_group_cpus(group)));
5601
		if (group->sgc->capacity != SCHED_CAPACITY_SCALE) {
5602 5603
			printk(KERN_CONT " (cpu_capacity = %d)",
				group->sgc->capacity);
5604
		}
L
Linus Torvalds 已提交
5605

I
Ingo Molnar 已提交
5606 5607
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
5608
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
5609

5610
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
5611
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
5612

5613 5614
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
5615 5616
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
5617 5618
	return 0;
}
L
Linus Torvalds 已提交
5619

I
Ingo Molnar 已提交
5620 5621 5622
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
5623

5624
	if (!sched_debug_enabled)
5625 5626
		return;

I
Ingo Molnar 已提交
5627 5628 5629 5630
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
5631

I
Ingo Molnar 已提交
5632 5633 5634
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
5635
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
5636
			break;
L
Linus Torvalds 已提交
5637 5638
		level++;
		sd = sd->parent;
5639
		if (!sd)
I
Ingo Molnar 已提交
5640 5641
			break;
	}
L
Linus Torvalds 已提交
5642
}
5643
#else /* !CONFIG_SCHED_DEBUG */
5644
# define sched_domain_debug(sd, cpu) do { } while (0)
5645 5646 5647 5648
static inline bool sched_debug(void)
{
	return false;
}
5649
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
5650

5651
static int sd_degenerate(struct sched_domain *sd)
5652
{
5653
	if (cpumask_weight(sched_domain_span(sd)) == 1)
5654 5655 5656 5657 5658 5659
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
5660
			 SD_BALANCE_EXEC |
5661
			 SD_SHARE_CPUCAPACITY |
5662 5663
			 SD_SHARE_PKG_RESOURCES |
			 SD_SHARE_POWERDOMAIN)) {
5664 5665 5666 5667 5668
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
5669
	if (sd->flags & (SD_WAKE_AFFINE))
5670 5671 5672 5673 5674
		return 0;

	return 1;
}

5675 5676
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
5677 5678 5679 5680 5681 5682
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

5683
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
5684 5685 5686 5687 5688 5689 5690
		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 |
5691
				SD_BALANCE_EXEC |
5692
				SD_SHARE_CPUCAPACITY |
5693
				SD_SHARE_PKG_RESOURCES |
5694 5695
				SD_PREFER_SIBLING |
				SD_SHARE_POWERDOMAIN);
5696 5697
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
5698 5699 5700 5701 5702 5703 5704
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

5705
static void free_rootdomain(struct rcu_head *rcu)
5706
{
5707
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
5708

5709
	cpupri_cleanup(&rd->cpupri);
5710
	cpudl_cleanup(&rd->cpudl);
5711
	free_cpumask_var(rd->dlo_mask);
5712 5713 5714 5715 5716 5717
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
5718 5719
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
5720
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
5721 5722
	unsigned long flags;

5723
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5724 5725

	if (rq->rd) {
I
Ingo Molnar 已提交
5726
		old_rd = rq->rd;
G
Gregory Haskins 已提交
5727

5728
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
5729
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5730

5731
		cpumask_clear_cpu(rq->cpu, old_rd->span);
5732

I
Ingo Molnar 已提交
5733
		/*
5734
		 * If we dont want to free the old_rd yet then
I
Ingo Molnar 已提交
5735 5736 5737 5738 5739
		 * 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
5740 5741 5742 5743 5744
	}

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

5745
	cpumask_set_cpu(rq->cpu, rd->span);
5746
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
5747
		set_rq_online(rq);
G
Gregory Haskins 已提交
5748

5749
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
5750 5751

	if (old_rd)
5752
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
5753 5754
}

5755
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
5756 5757 5758
{
	memset(rd, 0, sizeof(*rd));

5759
	if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
5760
		goto out;
5761
	if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
5762
		goto free_span;
5763
	if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
5764
		goto free_online;
5765
	if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
5766
		goto free_dlo_mask;
5767

5768
	init_dl_bw(&rd->dl_bw);
5769 5770
	if (cpudl_init(&rd->cpudl) != 0)
		goto free_dlo_mask;
5771

5772
	if (cpupri_init(&rd->cpupri) != 0)
5773
		goto free_rto_mask;
5774
	return 0;
5775

5776 5777
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
5778 5779
free_dlo_mask:
	free_cpumask_var(rd->dlo_mask);
5780 5781 5782 5783
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
5784
out:
5785
	return -ENOMEM;
G
Gregory Haskins 已提交
5786 5787
}

5788 5789 5790 5791 5792 5793
/*
 * 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
5794 5795
static void init_defrootdomain(void)
{
5796
	init_rootdomain(&def_root_domain);
5797

G
Gregory Haskins 已提交
5798 5799 5800
	atomic_set(&def_root_domain.refcount, 1);
}

5801
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5802 5803 5804 5805 5806 5807 5808
{
	struct root_domain *rd;

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

5809
	if (init_rootdomain(rd) != 0) {
5810 5811 5812
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5813 5814 5815 5816

	return rd;
}

5817
static void free_sched_groups(struct sched_group *sg, int free_sgc)
5818 5819 5820 5821 5822 5823 5824 5825 5826 5827
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

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

5828 5829
		if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
			kfree(sg->sgc);
5830 5831 5832 5833 5834 5835

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

5836 5837 5838
static void free_sched_domain(struct rcu_head *rcu)
{
	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
5839 5840 5841 5842 5843 5844 5845 5846

	/*
	 * 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)) {
5847
		kfree(sd->groups->sgc);
5848
		kfree(sd->groups);
5849
	}
5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863
	kfree(sd);
}

static void destroy_sched_domain(struct sched_domain *sd, int cpu)
{
	call_rcu(&sd->rcu, free_sched_domain);
}

static void destroy_sched_domains(struct sched_domain *sd, int cpu)
{
	for (; sd; sd = sd->parent)
		destroy_sched_domain(sd, cpu);
}

5864 5865 5866 5867 5868 5869 5870
/*
 * 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
5871
 * two cpus are in the same cache domain, see cpus_share_cache().
5872 5873
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
5874
DEFINE_PER_CPU(int, sd_llc_size);
5875
DEFINE_PER_CPU(int, sd_llc_id);
5876
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
5877 5878
DEFINE_PER_CPU(struct sched_domain *, sd_busy);
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
5879 5880 5881 5882

static void update_top_cache_domain(int cpu)
{
	struct sched_domain *sd;
5883
	struct sched_domain *busy_sd = NULL;
5884
	int id = cpu;
5885
	int size = 1;
5886 5887

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
5888
	if (sd) {
5889
		id = cpumask_first(sched_domain_span(sd));
5890
		size = cpumask_weight(sched_domain_span(sd));
5891
		busy_sd = sd->parent; /* sd_busy */
5892
	}
5893
	rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd);
5894 5895

	rcu_assign_pointer(per_cpu(sd_llc, cpu), sd);
5896
	per_cpu(sd_llc_size, cpu) = size;
5897
	per_cpu(sd_llc_id, cpu) = id;
5898 5899 5900

	sd = lowest_flag_domain(cpu, SD_NUMA);
	rcu_assign_pointer(per_cpu(sd_numa, cpu), sd);
5901 5902 5903

	sd = highest_flag_domain(cpu, SD_ASYM_PACKING);
	rcu_assign_pointer(per_cpu(sd_asym, cpu), sd);
5904 5905
}

L
Linus Torvalds 已提交
5906
/*
I
Ingo Molnar 已提交
5907
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
5908 5909
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
5910 5911
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
5912
{
5913
	struct rq *rq = cpu_rq(cpu);
5914 5915 5916
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
5917
	for (tmp = sd; tmp; ) {
5918 5919 5920
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
5921

5922
		if (sd_parent_degenerate(tmp, parent)) {
5923
			tmp->parent = parent->parent;
5924 5925
			if (parent->parent)
				parent->parent->child = tmp;
5926 5927 5928 5929 5930 5931 5932
			/*
			 * 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;
5933
			destroy_sched_domain(parent, cpu);
5934 5935
		} else
			tmp = tmp->parent;
5936 5937
	}

5938
	if (sd && sd_degenerate(sd)) {
5939
		tmp = sd;
5940
		sd = sd->parent;
5941
		destroy_sched_domain(tmp, cpu);
5942 5943 5944
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
5945

5946
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
5947

G
Gregory Haskins 已提交
5948
	rq_attach_root(rq, rd);
5949
	tmp = rq->sd;
N
Nick Piggin 已提交
5950
	rcu_assign_pointer(rq->sd, sd);
5951
	destroy_sched_domains(tmp, cpu);
5952 5953

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
5954 5955 5956 5957 5958
}

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
5959 5960
	int ret;

R
Rusty Russell 已提交
5961
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
5962 5963 5964 5965 5966
	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
5967 5968
	return 1;
}
I
Ingo Molnar 已提交
5969
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
5970

5971
struct s_data {
5972
	struct sched_domain ** __percpu sd;
5973 5974 5975
	struct root_domain	*rd;
};

5976 5977
enum s_alloc {
	sa_rootdomain,
5978
	sa_sd,
5979
	sa_sd_storage,
5980 5981 5982
	sa_none,
};

P
Peter Zijlstra 已提交
5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020
/*
 * 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));
}

6021 6022 6023 6024 6025 6026 6027
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;
6028
	struct sched_domain *sibling;
6029 6030 6031 6032 6033 6034 6035 6036 6037 6038
	int i;

	cpumask_clear(covered);

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

		if (cpumask_test_cpu(i, covered))
			continue;

6039
		sibling = *per_cpu_ptr(sdd->sd, i);
P
Peter Zijlstra 已提交
6040 6041

		/* See the comment near build_group_mask(). */
6042
		if (!cpumask_test_cpu(i, sched_domain_span(sibling)))
P
Peter Zijlstra 已提交
6043 6044
			continue;

6045
		sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
6046
				GFP_KERNEL, cpu_to_node(cpu));
6047 6048 6049 6050 6051

		if (!sg)
			goto fail;

		sg_span = sched_group_cpus(sg);
6052 6053 6054
		if (sibling->child)
			cpumask_copy(sg_span, sched_domain_span(sibling->child));
		else
6055 6056 6057 6058
			cpumask_set_cpu(i, sg_span);

		cpumask_or(covered, covered, sg_span);

6059 6060
		sg->sgc = *per_cpu_ptr(sdd->sgc, i);
		if (atomic_inc_return(&sg->sgc->ref) == 1)
P
Peter Zijlstra 已提交
6061 6062
			build_group_mask(sd, sg);

6063
		/*
6064
		 * Initialize sgc->capacity such that even if we mess up the
6065 6066 6067
		 * domains and no possible iteration will get us here, we won't
		 * die on a /0 trap.
		 */
6068
		sg->sgc->capacity = SCHED_CAPACITY_SCALE * cpumask_weight(sg_span);
6069

P
Peter Zijlstra 已提交
6070 6071 6072 6073 6074
		/*
		 * 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
6075
		if ((!groups && cpumask_test_cpu(cpu, sg_span)) ||
P
Peter Zijlstra 已提交
6076
		    group_balance_cpu(sg) == cpu)
6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093 6094 6095
			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;
}

6096
static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
L
Linus Torvalds 已提交
6097
{
6098 6099
	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
	struct sched_domain *child = sd->child;
L
Linus Torvalds 已提交
6100

6101 6102
	if (child)
		cpu = cpumask_first(sched_domain_span(child));
6103

6104
	if (sg) {
6105
		*sg = *per_cpu_ptr(sdd->sg, cpu);
6106 6107
		(*sg)->sgc = *per_cpu_ptr(sdd->sgc, cpu);
		atomic_set(&(*sg)->sgc->ref, 1); /* for claim_allocations */
6108
	}
6109 6110

	return cpu;
6111 6112
}

6113
/*
6114 6115
 * 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,
6116
 * and ->cpu_capacity to 0.
6117 6118
 *
 * Assumes the sched_domain tree is fully constructed
6119
 */
6120 6121
static int
build_sched_groups(struct sched_domain *sd, int cpu)
L
Linus Torvalds 已提交
6122
{
6123 6124 6125
	struct sched_group *first = NULL, *last = NULL;
	struct sd_data *sdd = sd->private;
	const struct cpumask *span = sched_domain_span(sd);
6126
	struct cpumask *covered;
6127
	int i;
6128

6129 6130 6131
	get_group(cpu, sdd, &sd->groups);
	atomic_inc(&sd->groups->ref);

6132
	if (cpu != cpumask_first(span))
6133 6134
		return 0;

6135 6136 6137
	lockdep_assert_held(&sched_domains_mutex);
	covered = sched_domains_tmpmask;

6138
	cpumask_clear(covered);
6139

6140 6141
	for_each_cpu(i, span) {
		struct sched_group *sg;
6142
		int group, j;
6143

6144 6145
		if (cpumask_test_cpu(i, covered))
			continue;
6146

6147
		group = get_group(i, sdd, &sg);
P
Peter Zijlstra 已提交
6148
		cpumask_setall(sched_group_mask(sg));
6149

6150 6151 6152
		for_each_cpu(j, span) {
			if (get_group(j, sdd, NULL) != group)
				continue;
6153

6154 6155 6156
			cpumask_set_cpu(j, covered);
			cpumask_set_cpu(j, sched_group_cpus(sg));
		}
6157

6158 6159 6160 6161 6162 6163 6164
		if (!first)
			first = sg;
		if (last)
			last->next = sg;
		last = sg;
	}
	last->next = first;
6165 6166

	return 0;
6167
}
6168

6169
/*
6170
 * Initialize sched groups cpu_capacity.
6171
 *
6172
 * cpu_capacity indicates the capacity of sched group, which is used while
6173
 * distributing the load between different sched groups in a sched domain.
6174 6175 6176 6177
 * 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.
6178
 */
6179
static void init_sched_groups_capacity(int cpu, struct sched_domain *sd)
6180
{
6181
	struct sched_group *sg = sd->groups;
6182

6183
	WARN_ON(!sg);
6184 6185 6186 6187 6188

	do {
		sg->group_weight = cpumask_weight(sched_group_cpus(sg));
		sg = sg->next;
	} while (sg != sd->groups);
6189

P
Peter Zijlstra 已提交
6190
	if (cpu != group_balance_cpu(sg))
6191
		return;
6192

6193 6194
	update_group_capacity(sd, cpu);
	atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight);
6195 6196
}

6197 6198 6199 6200 6201
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

6202
static int default_relax_domain_level = -1;
6203
int sched_domain_level_max;
6204 6205 6206

static int __init setup_relax_domain_level(char *str)
{
6207 6208
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
6209

6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227
	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 */
6228
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6229 6230
	} else {
		/* turn on idle balance on this domain */
6231
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6232 6233 6234
	}
}

6235 6236 6237
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

6238 6239 6240 6241 6242
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
6243 6244
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
6245 6246
	case sa_sd:
		free_percpu(d->sd); /* fall through */
6247
	case sa_sd_storage:
6248
		__sdt_free(cpu_map); /* fall through */
6249 6250 6251 6252
	case sa_none:
		break;
	}
}
6253

6254 6255 6256
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
6257 6258
	memset(d, 0, sizeof(*d));

6259 6260
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
6261 6262 6263
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
6264
	d->rd = alloc_rootdomain();
6265
	if (!d->rd)
6266
		return sa_sd;
6267 6268
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
6269

6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281
/*
 * 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;

6282
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
6283
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
6284

6285 6286
	if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
		*per_cpu_ptr(sdd->sgc, cpu) = NULL;
6287 6288
}

6289 6290
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
6291
enum numa_topology_type sched_numa_topology_type;
6292
static int *sched_domains_numa_distance;
6293
int sched_max_numa_distance;
6294 6295
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
6296
#endif
6297

6298 6299 6300
/*
 * SD_flags allowed in topology descriptions.
 *
6301
 * SD_SHARE_CPUCAPACITY      - describes SMT topologies
6302 6303
 * SD_SHARE_PKG_RESOURCES - describes shared caches
 * SD_NUMA                - describes NUMA topologies
6304
 * SD_SHARE_POWERDOMAIN   - describes shared power domain
6305 6306 6307 6308 6309
 *
 * Odd one out:
 * SD_ASYM_PACKING        - describes SMT quirks
 */
#define TOPOLOGY_SD_FLAGS		\
6310
	(SD_SHARE_CPUCAPACITY |		\
6311 6312
	 SD_SHARE_PKG_RESOURCES |	\
	 SD_NUMA |			\
6313 6314
	 SD_ASYM_PACKING |		\
	 SD_SHARE_POWERDOMAIN)
6315 6316

static struct sched_domain *
6317
sd_init(struct sched_domain_topology_level *tl, int cpu)
6318 6319
{
	struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);
6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335
	int sd_weight, sd_flags = 0;

#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;
6336 6337 6338 6339 6340

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
6341
		.imbalance_pct		= 125,
6342 6343 6344 6345

		.cache_nice_tries	= 0,
		.busy_idx		= 0,
		.idle_idx		= 0,
6346 6347 6348 6349 6350 6351
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
6352 6353
					| 1*SD_BALANCE_EXEC
					| 1*SD_BALANCE_FORK
6354
					| 0*SD_BALANCE_WAKE
6355
					| 1*SD_WAKE_AFFINE
6356
					| 0*SD_SHARE_CPUCAPACITY
6357
					| 0*SD_SHARE_PKG_RESOURCES
6358
					| 0*SD_SERIALIZE
6359
					| 0*SD_PREFER_SIBLING
6360 6361
					| 0*SD_NUMA
					| sd_flags
6362
					,
6363

6364 6365
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
6366
		.smt_gain		= 0,
6367 6368
		.max_newidle_lb_cost	= 0,
		.next_decay_max_lb_cost	= jiffies,
6369 6370 6371
#ifdef CONFIG_SCHED_DEBUG
		.name			= tl->name,
#endif
6372 6373 6374
	};

	/*
6375
	 * Convert topological properties into behaviour.
6376
	 */
6377

6378
	if (sd->flags & SD_SHARE_CPUCAPACITY) {
6379
		sd->flags |= SD_PREFER_SIBLING;
6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409
		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;
	}

	sd->private = &tl->data;
6410 6411 6412 6413

	return sd;
}

6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427
/*
 * 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, },
};

6428 6429
static struct sched_domain_topology_level *sched_domain_topology =
	default_topology;
6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440

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

void set_sched_topology(struct sched_domain_topology_level *tl)
{
	sched_domain_topology = tl;
}

#ifdef CONFIG_NUMA

6441 6442 6443 6444 6445
static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466
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");
}

6467
bool find_numa_distance(int distance)
6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481
{
	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;
}

6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506
/*
 * 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;

6507
	if (sched_domains_numa_levels <= 1) {
6508
		sched_numa_topology_type = NUMA_DIRECT;
6509 6510
		return;
	}
6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533

	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;
		}
	}
}

6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554
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++) {
6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578
			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;
6579
		}
6580 6581 6582 6583 6584 6585

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
6586
	}
6587 6588 6589 6590

	if (!level)
		return;

6591 6592 6593 6594
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
6595
	 * The sched_domains_numa_distance[] array includes the actual distance
6596 6597 6598
	 * numbers.
	 */

6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609
	/*
	 * 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;

6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624
	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++) {
6625
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
6626 6627 6628 6629 6630
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

6631
			for_each_node(k) {
6632
				if (node_distance(j, k) > sched_domains_numa_distance[i])
6633 6634 6635 6636 6637 6638 6639
					continue;

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

6640 6641 6642
	/* Compute default topology size */
	for (i = 0; sched_domain_topology[i].mask; i++);

6643
	tl = kzalloc((i + level + 1) *
6644 6645 6646 6647 6648 6649 6650
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
6651 6652
	for (i = 0; sched_domain_topology[i].mask; i++)
		tl[i] = sched_domain_topology[i];
6653 6654 6655 6656 6657 6658 6659

	/*
	 * .. 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,
6660
			.sd_flags = cpu_numa_flags,
6661 6662
			.flags = SDTL_OVERLAP,
			.numa_level = j,
6663
			SD_INIT_NAME(NUMA)
6664 6665 6666 6667
		};
	}

	sched_domain_topology = tl;
6668 6669

	sched_domains_numa_levels = level;
6670
	sched_max_numa_distance = sched_domains_numa_distance[level - 1];
6671 6672

	init_numa_topology_type();
6673
}
6674

6675
static void sched_domains_numa_masks_set(unsigned int cpu)
6676 6677
{
	int node = cpu_to_node(cpu);
6678
	int i, j;
6679 6680 6681 6682 6683 6684 6685 6686 6687

	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]);
		}
	}
}

6688
static void sched_domains_numa_masks_clear(unsigned int cpu)
6689 6690
{
	int i, j;
6691

6692 6693 6694 6695 6696 6697
	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]);
	}
}

6698
#else
6699 6700 6701
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) { }
6702 6703
#endif /* CONFIG_NUMA */

6704 6705 6706 6707 6708
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6709
	for_each_sd_topology(tl) {
6710 6711 6712 6713 6714 6715 6716 6717 6718 6719
		struct sd_data *sdd = &tl->data;

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

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

6720 6721
		sdd->sgc = alloc_percpu(struct sched_group_capacity *);
		if (!sdd->sgc)
6722 6723
			return -ENOMEM;

6724 6725 6726
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
			struct sched_group *sg;
6727
			struct sched_group_capacity *sgc;
6728

P
Peter Zijlstra 已提交
6729
			sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740
					GFP_KERNEL, cpu_to_node(j));
			if (!sd)
				return -ENOMEM;

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

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

6741 6742
			sg->next = sg;

6743
			*per_cpu_ptr(sdd->sg, j) = sg;
6744

6745
			sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
6746
					GFP_KERNEL, cpu_to_node(j));
6747
			if (!sgc)
6748 6749
				return -ENOMEM;

6750
			*per_cpu_ptr(sdd->sgc, j) = sgc;
6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761
		}
	}

	return 0;
}

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

6762
	for_each_sd_topology(tl) {
6763 6764 6765
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776
			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));
			}

			if (sdd->sg)
				kfree(*per_cpu_ptr(sdd->sg, j));
6777 6778
			if (sdd->sgc)
				kfree(*per_cpu_ptr(sdd->sgc, j));
6779 6780
		}
		free_percpu(sdd->sd);
6781
		sdd->sd = NULL;
6782
		free_percpu(sdd->sg);
6783
		sdd->sg = NULL;
6784 6785
		free_percpu(sdd->sgc);
		sdd->sgc = NULL;
6786 6787 6788
	}
}

6789
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6790 6791
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
6792
{
6793
	struct sched_domain *sd = sd_init(tl, cpu);
6794
	if (!sd)
6795
		return child;
6796 6797

	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
6798 6799 6800
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
6801
		child->parent = sd;
6802
		sd->child = child;
P
Peter Zijlstra 已提交
6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816

		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));
		}

6817
	}
6818
	set_domain_attribute(sd, attr);
6819 6820 6821 6822

	return sd;
}

6823 6824 6825 6826
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
6827 6828
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
6829
{
6830
	enum s_alloc alloc_state;
6831
	struct sched_domain *sd;
6832
	struct s_data d;
6833
	int i, ret = -ENOMEM;
6834

6835 6836 6837
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
6838

6839
	/* Set up domains for cpus specified by the cpu_map. */
6840
	for_each_cpu(i, cpu_map) {
6841 6842
		struct sched_domain_topology_level *tl;

6843
		sd = NULL;
6844
		for_each_sd_topology(tl) {
6845
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
6846 6847
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
6848 6849
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
6850 6851
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
6852
		}
6853 6854 6855 6856 6857 6858
	}

	/* 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));
6859 6860 6861 6862 6863 6864 6865
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
6866
		}
6867
	}
6868

6869
	/* Calculate CPU capacity for physical packages and nodes */
6870 6871 6872
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
6873

6874 6875
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
6876
			init_sched_groups_capacity(i, sd);
6877
		}
6878
	}
6879

L
Linus Torvalds 已提交
6880
	/* Attach the domains */
6881
	rcu_read_lock();
6882
	for_each_cpu(i, cpu_map) {
6883
		sd = *per_cpu_ptr(d.sd, i);
6884
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
6885
	}
6886
	rcu_read_unlock();
6887

6888
	ret = 0;
6889
error:
6890
	__free_domain_allocs(&d, alloc_state, cpu_map);
6891
	return ret;
L
Linus Torvalds 已提交
6892
}
P
Paul Jackson 已提交
6893

6894
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
6895
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
6896 6897
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
6898 6899 6900

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
6901 6902
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
6903
 */
6904
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
6905

6906 6907 6908 6909 6910
/*
 * 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.
 */
6911
int __weak arch_update_cpu_topology(void)
6912
{
6913
	return 0;
6914 6915
}

6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932 6933 6934 6935 6936 6937 6938 6939 6940
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);
}

6941
/*
I
Ingo Molnar 已提交
6942
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
6943 6944
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
6945
 */
6946
static int init_sched_domains(const struct cpumask *cpu_map)
6947
{
6948 6949
	int err;

6950
	arch_update_cpu_topology();
P
Paul Jackson 已提交
6951
	ndoms_cur = 1;
6952
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
6953
	if (!doms_cur)
6954 6955
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
6956
	err = build_sched_domains(doms_cur[0], NULL);
6957
	register_sched_domain_sysctl();
6958 6959

	return err;
6960 6961 6962 6963 6964 6965
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
6966
static void detach_destroy_domains(const struct cpumask *cpu_map)
6967 6968 6969
{
	int i;

6970
	rcu_read_lock();
6971
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
6972
		cpu_attach_domain(NULL, &def_root_domain, i);
6973
	rcu_read_unlock();
6974 6975
}

6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990 6991
/* 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
6992 6993
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
6994
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
6995 6996 6997
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
6998
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
6999 7000 7001
 * 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
7002 7003 7004
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
7005 7006 7007 7008 7009 7010
 * 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
7011
 *
7012
 * If doms_new == NULL it will be replaced with cpu_online_mask.
7013 7014
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
7015
 *
P
Paul Jackson 已提交
7016 7017
 * Call with hotplug lock held
 */
7018
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
7019
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
7020
{
7021
	int i, j, n;
7022
	int new_topology;
P
Paul Jackson 已提交
7023

7024
	mutex_lock(&sched_domains_mutex);
7025

7026 7027 7028
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

7029 7030 7031
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

7032
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
7033 7034 7035

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
7036
		for (j = 0; j < n && !new_topology; j++) {
7037
			if (cpumask_equal(doms_cur[i], doms_new[j])
7038
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
7039 7040 7041
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
7042
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
7043 7044 7045 7046
match1:
		;
	}

7047
	n = ndoms_cur;
7048
	if (doms_new == NULL) {
7049
		n = 0;
7050
		doms_new = &fallback_doms;
7051
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
7052
		WARN_ON_ONCE(dattr_new);
7053 7054
	}

P
Paul Jackson 已提交
7055 7056
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
7057
		for (j = 0; j < n && !new_topology; j++) {
7058
			if (cpumask_equal(doms_new[i], doms_cur[j])
7059
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
7060 7061 7062
				goto match2;
		}
		/* no match - add a new doms_new */
7063
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
7064 7065 7066 7067 7068
match2:
		;
	}

	/* Remember the new sched domains */
7069 7070
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
7071
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
7072
	doms_cur = doms_new;
7073
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
7074
	ndoms_cur = ndoms_new;
7075 7076

	register_sched_domain_sysctl();
7077

7078
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
7079 7080
}

7081 7082
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
7083
/*
7084 7085 7086
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
7087 7088 7089
 *
 * 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
7090
 */
7091
static void cpuset_cpu_active(void)
7092
{
7093
	if (cpuhp_tasks_frozen) {
7094 7095 7096 7097 7098 7099 7100 7101 7102
		/*
		 * 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);
7103
			return;
7104 7105 7106 7107 7108 7109
		}
		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */
7110
	}
7111
	cpuset_update_active_cpus(true);
7112
}
7113

7114
static int cpuset_cpu_inactive(unsigned int cpu)
7115
{
7116 7117
	unsigned long flags;
	struct dl_bw *dl_b;
7118 7119
	bool overflow;
	int cpus;
7120

7121
	if (!cpuhp_tasks_frozen) {
7122 7123
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
7124

7125 7126 7127 7128
		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);
7129

7130
		rcu_read_unlock_sched();
7131

7132
		if (overflow)
7133
			return -EBUSY;
7134
		cpuset_update_active_cpus(false);
7135
	} else {
7136 7137
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
7138
	}
7139
	return 0;
7140 7141
}

7142
int sched_cpu_activate(unsigned int cpu)
7143
{
7144 7145 7146
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

7147
	set_cpu_active(cpu, true);
7148

7149
	if (sched_smp_initialized) {
7150
		sched_domains_numa_masks_set(cpu);
7151
		cpuset_cpu_active();
7152
	}
7153 7154 7155 7156 7157 7158 7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171

	/*
	 * 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();

7172
	return 0;
7173 7174
}

7175
int sched_cpu_deactivate(unsigned int cpu)
7176 7177 7178
{
	int ret;

7179
	set_cpu_active(cpu, false);
7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193
	/*
	 * 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();
7194 7195 7196 7197 7198 7199 7200 7201

	if (!sched_smp_initialized)
		return 0;

	ret = cpuset_cpu_inactive(cpu);
	if (ret) {
		set_cpu_active(cpu, true);
		return ret;
7202
	}
7203 7204
	sched_domains_numa_masks_clear(cpu);
	return 0;
7205 7206
}

7207 7208 7209 7210 7211 7212 7213 7214 7215
static void sched_rq_cpu_starting(unsigned int cpu)
{
	struct rq *rq = cpu_rq(cpu);

	rq->calc_load_update = calc_load_update;
	account_reset_rq(rq);
	update_max_interval();
}

7216 7217 7218
int sched_cpu_starting(unsigned int cpu)
{
	set_cpu_rq_start_time(cpu);
7219
	sched_rq_cpu_starting(cpu);
7220
	return 0;
7221 7222
}

7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240
#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();
7241
	nohz_balance_exit_idle(cpu);
7242
	hrtick_clear(rq);
7243 7244 7245 7246
	return 0;
}
#endif

L
Linus Torvalds 已提交
7247 7248
void __init sched_init_smp(void)
{
7249 7250 7251
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7252
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
7253

7254 7255
	sched_init_numa();

7256 7257 7258 7259 7260
	/*
	 * 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.
	 */
7261
	mutex_lock(&sched_domains_mutex);
7262
	init_sched_domains(cpu_active_mask);
7263 7264 7265
	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);
7266
	mutex_unlock(&sched_domains_mutex);
7267

7268
	/* Move init over to a non-isolated CPU */
7269
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
7270
		BUG();
I
Ingo Molnar 已提交
7271
	sched_init_granularity();
7272
	free_cpumask_var(non_isolated_cpus);
7273

7274
	init_sched_rt_class();
7275
	init_sched_dl_class();
7276
	sched_smp_initialized = true;
L
Linus Torvalds 已提交
7277
}
7278 7279 7280

static int __init migration_init(void)
{
7281
	sched_rq_cpu_starting(smp_processor_id());
7282
	return 0;
L
Linus Torvalds 已提交
7283
}
7284 7285
early_initcall(migration_init);

L
Linus Torvalds 已提交
7286 7287 7288
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
7289
	sched_init_granularity();
L
Linus Torvalds 已提交
7290 7291 7292 7293 7294 7295 7296 7297 7298 7299
}
#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);
}

7300
#ifdef CONFIG_CGROUP_SCHED
7301 7302 7303 7304
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
7305
struct task_group root_task_group;
7306
LIST_HEAD(task_groups);
7307 7308 7309

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
7310
#endif
P
Peter Zijlstra 已提交
7311

7312
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
P
Peter Zijlstra 已提交
7313

L
Linus Torvalds 已提交
7314 7315
void __init sched_init(void)
{
I
Ingo Molnar 已提交
7316
	int i, j;
7317 7318 7319 7320 7321 7322 7323 7324 7325
	unsigned long alloc_size = 0, ptr;

#ifdef CONFIG_FAIR_GROUP_SCHED
	alloc_size += 2 * nr_cpu_ids * sizeof(void **);
#endif
#ifdef CONFIG_RT_GROUP_SCHED
	alloc_size += 2 * nr_cpu_ids * sizeof(void **);
#endif
	if (alloc_size) {
7326
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
7327 7328

#ifdef CONFIG_FAIR_GROUP_SCHED
7329
		root_task_group.se = (struct sched_entity **)ptr;
7330 7331
		ptr += nr_cpu_ids * sizeof(void **);

7332
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
7333
		ptr += nr_cpu_ids * sizeof(void **);
7334

7335
#endif /* CONFIG_FAIR_GROUP_SCHED */
7336
#ifdef CONFIG_RT_GROUP_SCHED
7337
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
7338 7339
		ptr += nr_cpu_ids * sizeof(void **);

7340
		root_task_group.rt_rq = (struct rt_rq **)ptr;
7341 7342
		ptr += nr_cpu_ids * sizeof(void **);

7343
#endif /* CONFIG_RT_GROUP_SCHED */
7344
	}
7345
#ifdef CONFIG_CPUMASK_OFFSTACK
7346 7347 7348
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7349
	}
7350
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
7351

7352 7353 7354
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth,
7355
			global_rt_period(), global_rt_runtime());
7356

G
Gregory Haskins 已提交
7357 7358 7359 7360
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

7361
#ifdef CONFIG_RT_GROUP_SCHED
7362
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
7363
			global_rt_period(), global_rt_runtime());
7364
#endif /* CONFIG_RT_GROUP_SCHED */
7365

D
Dhaval Giani 已提交
7366
#ifdef CONFIG_CGROUP_SCHED
7367 7368
	task_group_cache = KMEM_CACHE(task_group, 0);

7369 7370
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
7371
	INIT_LIST_HEAD(&root_task_group.siblings);
7372
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
7373
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
7374

7375
	for_each_possible_cpu(i) {
7376
		struct rq *rq;
L
Linus Torvalds 已提交
7377 7378

		rq = cpu_rq(i);
7379
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
7380
		rq->nr_running = 0;
7381 7382
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
7383
		init_cfs_rq(&rq->cfs);
7384 7385
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
7386
#ifdef CONFIG_FAIR_GROUP_SCHED
7387
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
7388
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
D
Dhaval Giani 已提交
7389
		/*
7390
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
7391 7392 7393 7394
		 *
		 * 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
7395
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
7396 7397 7398
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
7399
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
7400 7401 7402
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
7403
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
7404
		 *
7405 7406
		 * 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
7407
		 */
7408
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
7409
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
7410 7411 7412
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
7413
#ifdef CONFIG_RT_GROUP_SCHED
7414
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
7415
#endif
L
Linus Torvalds 已提交
7416

I
Ingo Molnar 已提交
7417 7418
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
7419

L
Linus Torvalds 已提交
7420
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
7421
		rq->sd = NULL;
G
Gregory Haskins 已提交
7422
		rq->rd = NULL;
7423
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
7424
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
7425
		rq->active_balance = 0;
I
Ingo Molnar 已提交
7426
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
7427
		rq->push_cpu = 0;
7428
		rq->cpu = i;
7429
		rq->online = 0;
7430 7431
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
7432
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
7433 7434 7435

		INIT_LIST_HEAD(&rq->cfs_tasks);

7436
		rq_attach_root(rq, &def_root_domain);
7437
#ifdef CONFIG_NO_HZ_COMMON
7438
		rq->last_load_update_tick = jiffies;
7439
		rq->nohz_flags = 0;
7440
#endif
7441 7442 7443
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
7444
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
7445
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
7446 7447 7448
		atomic_set(&rq->nr_iowait, 0);
	}

7449
	set_load_weight(&init_task);
7450

7451 7452 7453 7454
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&init_task.preempt_notifiers);
#endif

L
Linus Torvalds 已提交
7455 7456 7457 7458 7459 7460
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
	atomic_inc(&init_mm.mm_count);
	enter_lazy_tlb(&init_mm, current);

7461 7462 7463 7464 7465
	/*
	 * During early bootup we pretend to be a normal task:
	 */
	current->sched_class = &fair_sched_class;

L
Linus Torvalds 已提交
7466 7467 7468 7469 7470 7471 7472
	/*
	 * 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());
7473 7474 7475

	calc_load_update = jiffies + LOAD_FREQ;

7476
#ifdef CONFIG_SMP
7477
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
7478 7479 7480
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
7481
	idle_thread_set_boot_cpu();
7482
	set_cpu_rq_start_time(smp_processor_id());
7483 7484
#endif
	init_sched_fair_class();
7485

7486
	scheduler_running = 1;
L
Linus Torvalds 已提交
7487 7488
}

7489
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
7490 7491
static inline int preempt_count_equals(int preempt_offset)
{
7492
	int nested = preempt_count() + rcu_preempt_depth();
7493

A
Arnd Bergmann 已提交
7494
	return (nested == preempt_offset);
7495 7496
}

7497
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7498
{
P
Peter Zijlstra 已提交
7499 7500 7501 7502 7503
	/*
	 * 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.
	 */
7504
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
7505 7506 7507 7508
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
7509
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
7510

7511 7512 7513 7514 7515
	___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
7516 7517 7518
{
	static unsigned long prev_jiffy;	/* ratelimiting */

7519
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
7520 7521
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
7522
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
7523 7524 7525 7526 7527
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

P
Peter Zijlstra 已提交
7528 7529 7530 7531 7532 7533 7534
	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
7535

7536 7537 7538
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
7539 7540 7541
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
7542 7543 7544 7545 7546 7547 7548
#ifdef CONFIG_DEBUG_PREEMPT
	if (!preempt_count_equals(preempt_offset)) {
		pr_err("Preemption disabled at:");
		print_ip_sym(current->preempt_disable_ip);
		pr_cont("\n");
	}
#endif
I
Ingo Molnar 已提交
7549
	dump_stack();
L
Linus Torvalds 已提交
7550
}
7551
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
7552 7553 7554
#endif

#ifdef CONFIG_MAGIC_SYSRQ
7555
void normalize_rt_tasks(void)
7556
{
7557
	struct task_struct *g, *p;
7558 7559 7560
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
7561

7562
	read_lock(&tasklist_lock);
7563
	for_each_process_thread(g, p) {
7564 7565 7566
		/*
		 * Only normalize user tasks:
		 */
7567
		if (p->flags & PF_KTHREAD)
7568 7569
			continue;

I
Ingo Molnar 已提交
7570 7571
		p->se.exec_start		= 0;
#ifdef CONFIG_SCHEDSTATS
7572 7573 7574
		p->se.statistics.wait_start	= 0;
		p->se.statistics.sleep_start	= 0;
		p->se.statistics.block_start	= 0;
I
Ingo Molnar 已提交
7575
#endif
I
Ingo Molnar 已提交
7576

7577
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
7578 7579 7580 7581
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
7582
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
7583
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
7584
			continue;
I
Ingo Molnar 已提交
7585
		}
L
Linus Torvalds 已提交
7586

7587
		__sched_setscheduler(p, &attr, false, false);
7588
	}
7589
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
7590 7591 7592
}

#endif /* CONFIG_MAGIC_SYSRQ */
7593

7594
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
7595
/*
7596
 * These functions are only useful for the IA64 MCA handling, or kdb.
7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609
 *
 * 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!
7610 7611
 *
 * Return: The current task for @cpu.
7612
 */
7613
struct task_struct *curr_task(int cpu)
7614 7615 7616 7617
{
	return cpu_curr(cpu);
}

7618 7619 7620
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
7621 7622 7623 7624 7625 7626
/**
 * 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
7627 7628
 * 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
7629 7630 7631 7632 7633 7634 7635
 * 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!
 */
7636
void set_curr_task(int cpu, struct task_struct *p)
7637 7638 7639 7640 7641
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
7642

D
Dhaval Giani 已提交
7643
#ifdef CONFIG_CGROUP_SCHED
7644 7645 7646
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

7647
static void sched_free_group(struct task_group *tg)
7648 7649 7650
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
7651
	autogroup_free(tg);
7652
	kmem_cache_free(task_group_cache, tg);
7653 7654 7655
}

/* allocate runqueue etc for a new task group */
7656
struct task_group *sched_create_group(struct task_group *parent)
7657 7658 7659
{
	struct task_group *tg;

7660
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
7661 7662 7663
	if (!tg)
		return ERR_PTR(-ENOMEM);

7664
	if (!alloc_fair_sched_group(tg, parent))
7665 7666
		goto err;

7667
	if (!alloc_rt_sched_group(tg, parent))
7668 7669
		goto err;

7670 7671 7672
	return tg;

err:
7673
	sched_free_group(tg);
7674 7675 7676 7677 7678 7679 7680
	return ERR_PTR(-ENOMEM);
}

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

7681
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7682
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
7683 7684 7685 7686 7687

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
7688
	list_add_rcu(&tg->siblings, &parent->children);
7689
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7690 7691
}

7692
/* rcu callback to free various structures associated with a task group */
7693
static void sched_free_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
7694 7695
{
	/* now it should be safe to free those cfs_rqs */
7696
	sched_free_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
7697 7698
}

7699
void sched_destroy_group(struct task_group *tg)
7700 7701
{
	/* wait for possible concurrent references to cfs_rqs complete */
7702
	call_rcu(&tg->rcu, sched_free_group_rcu);
7703 7704 7705
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
7706
{
7707
	unsigned long flags;
S
Srivatsa Vaddagiri 已提交
7708

7709
	/* end participation in shares distribution */
7710
	unregister_fair_sched_group(tg);
7711 7712

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7713
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
7714
	list_del_rcu(&tg->siblings);
7715
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7716 7717
}

7718
/* change task's runqueue when it moves between groups.
I
Ingo Molnar 已提交
7719 7720 7721
 *	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.
7722 7723
 */
void sched_move_task(struct task_struct *tsk)
S
Srivatsa Vaddagiri 已提交
7724
{
P
Peter Zijlstra 已提交
7725
	struct task_group *tg;
7726
	int queued, running;
7727
	struct rq_flags rf;
S
Srivatsa Vaddagiri 已提交
7728 7729
	struct rq *rq;

7730
	rq = task_rq_lock(tsk, &rf);
S
Srivatsa Vaddagiri 已提交
7731

7732
	running = task_current(rq, tsk);
7733
	queued = task_on_rq_queued(tsk);
S
Srivatsa Vaddagiri 已提交
7734

7735
	if (queued)
7736
		dequeue_task(rq, tsk, DEQUEUE_SAVE | DEQUEUE_MOVE);
7737
	if (unlikely(running))
7738
		put_prev_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
7739

7740 7741 7742 7743 7744 7745
	/*
	 * 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
7746 7747 7748 7749
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
7750
#ifdef CONFIG_FAIR_GROUP_SCHED
7751
	if (tsk->sched_class->task_move_group)
7752
		tsk->sched_class->task_move_group(tsk);
7753
	else
P
Peter Zijlstra 已提交
7754
#endif
7755
		set_task_rq(tsk, task_cpu(tsk));
P
Peter Zijlstra 已提交
7756

7757 7758
	if (unlikely(running))
		tsk->sched_class->set_curr_task(rq);
7759
	if (queued)
7760
		enqueue_task(rq, tsk, ENQUEUE_RESTORE | ENQUEUE_MOVE);
S
Srivatsa Vaddagiri 已提交
7761

7762
	task_rq_unlock(rq, tsk, &rf);
S
Srivatsa Vaddagiri 已提交
7763
}
D
Dhaval Giani 已提交
7764
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
7765

7766 7767 7768 7769 7770
#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
7771

P
Peter Zijlstra 已提交
7772 7773
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
7774
{
P
Peter Zijlstra 已提交
7775
	struct task_struct *g, *p;
7776

7777 7778 7779 7780 7781 7782
	/*
	 * Autogroups do not have RT tasks; see autogroup_create().
	 */
	if (task_group_is_autogroup(tg))
		return 0;

7783
	for_each_process_thread(g, p) {
7784
		if (rt_task(p) && task_group(p) == tg)
P
Peter Zijlstra 已提交
7785
			return 1;
7786
	}
7787

P
Peter Zijlstra 已提交
7788 7789
	return 0;
}
7790

P
Peter Zijlstra 已提交
7791 7792 7793 7794 7795
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
7796

7797
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
7798 7799 7800 7801 7802
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
7803

P
Peter Zijlstra 已提交
7804 7805
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
7806

P
Peter Zijlstra 已提交
7807 7808 7809
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
7810 7811
	}

7812 7813 7814 7815 7816
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
7817

7818 7819 7820
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
7821 7822
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
7823

P
Peter Zijlstra 已提交
7824
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7825

7826 7827 7828 7829 7830
	/*
	 * 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
7831

7832 7833 7834
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
7835 7836 7837
	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
7838

P
Peter Zijlstra 已提交
7839 7840 7841 7842
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
7843

P
Peter Zijlstra 已提交
7844
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7845
	}
P
Peter Zijlstra 已提交
7846

P
Peter Zijlstra 已提交
7847 7848 7849 7850
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
7851 7852
}

P
Peter Zijlstra 已提交
7853
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
7854
{
7855 7856
	int ret;

P
Peter Zijlstra 已提交
7857 7858 7859 7860 7861 7862
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

7863 7864 7865 7866 7867
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
7868 7869
}

7870
static int tg_set_rt_bandwidth(struct task_group *tg,
7871
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
7872
{
P
Peter Zijlstra 已提交
7873
	int i, err = 0;
P
Peter Zijlstra 已提交
7874

7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885
	/*
	 * 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
7886
	mutex_lock(&rt_constraints_mutex);
7887
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
7888 7889
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
7890
		goto unlock;
P
Peter Zijlstra 已提交
7891

7892
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
7893 7894
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
7895 7896 7897 7898

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

7899
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7900
		rt_rq->rt_runtime = rt_runtime;
7901
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7902
	}
7903
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
7904
unlock:
7905
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
7906 7907 7908
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
7909 7910
}

7911
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
7912 7913 7914 7915 7916 7917 7918 7919
{
	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;

7920
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
7921 7922
}

7923
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
7924 7925 7926
{
	u64 rt_runtime_us;

7927
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
7928 7929
		return -1;

7930
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
7931 7932 7933
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
7934

7935
static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
7936 7937 7938
{
	u64 rt_runtime, rt_period;

7939
	rt_period = rt_period_us * NSEC_PER_USEC;
7940 7941
	rt_runtime = tg->rt_bandwidth.rt_runtime;

7942
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
7943 7944
}

7945
static long sched_group_rt_period(struct task_group *tg)
7946 7947 7948 7949 7950 7951 7952
{
	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;
}
7953
#endif /* CONFIG_RT_GROUP_SCHED */
7954

7955
#ifdef CONFIG_RT_GROUP_SCHED
7956 7957 7958 7959 7960
static int sched_rt_global_constraints(void)
{
	int ret = 0;

	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
7961
	read_lock(&tasklist_lock);
7962
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
7963
	read_unlock(&tasklist_lock);
7964 7965 7966 7967
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
7968

7969
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
7970 7971 7972 7973 7974 7975 7976 7977
{
	/* 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;
}

7978
#else /* !CONFIG_RT_GROUP_SCHED */
7979 7980
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
7981
	unsigned long flags;
7982
	int i;
7983

7984
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
7985 7986 7987
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

7988
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7989
		rt_rq->rt_runtime = global_rt_runtime();
7990
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7991
	}
7992
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
7993

7994
	return 0;
7995
}
7996
#endif /* CONFIG_RT_GROUP_SCHED */
7997

7998
static int sched_dl_global_validate(void)
7999
{
8000 8001
	u64 runtime = global_rt_runtime();
	u64 period = global_rt_period();
8002
	u64 new_bw = to_ratio(period, runtime);
8003
	struct dl_bw *dl_b;
8004
	int cpu, ret = 0;
8005
	unsigned long flags;
8006 8007 8008 8009 8010 8011 8012 8013 8014 8015

	/*
	 * 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!
	 */
8016
	for_each_possible_cpu(cpu) {
8017 8018
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8019

8020
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8021 8022
		if (new_bw < dl_b->total_bw)
			ret = -EBUSY;
8023
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8024

8025 8026
		rcu_read_unlock_sched();

8027 8028
		if (ret)
			break;
8029 8030
	}

8031
	return ret;
8032 8033
}

8034
static void sched_dl_do_global(void)
8035
{
8036
	u64 new_bw = -1;
8037
	struct dl_bw *dl_b;
8038
	int cpu;
8039
	unsigned long flags;
8040

8041 8042 8043 8044 8045 8046 8047 8048 8049 8050
	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) {
8051 8052
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8053

8054
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8055
		dl_b->bw = new_bw;
8056
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8057 8058

		rcu_read_unlock_sched();
8059
	}
8060 8061 8062 8063 8064 8065 8066
}

static int sched_rt_global_validate(void)
{
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

8067 8068
	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
8069 8070 8071 8072 8073 8074 8075 8076 8077
		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());
8078 8079
}

8080
int sched_rt_handler(struct ctl_table *table, int write,
8081
		void __user *buffer, size_t *lenp,
8082 8083 8084 8085
		loff_t *ppos)
{
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);
8086
	int ret;
8087 8088 8089 8090 8091

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

8092
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8093 8094

	if (!ret && write) {
8095 8096 8097 8098
		ret = sched_rt_global_validate();
		if (ret)
			goto undo;

8099
		ret = sched_dl_global_validate();
8100 8101 8102
		if (ret)
			goto undo;

8103
		ret = sched_rt_global_constraints();
8104 8105 8106 8107 8108 8109 8110 8111 8112 8113
		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;
8114 8115 8116 8117 8118
	}
	mutex_unlock(&mutex);

	return ret;
}
8119

8120
int sched_rr_handler(struct ctl_table *table, int write,
8121 8122 8123 8124 8125 8126 8127 8128
		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);
8129 8130
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
8131
	if (!ret && write) {
8132 8133
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
8134 8135 8136 8137 8138
	}
	mutex_unlock(&mutex);
	return ret;
}

8139
#ifdef CONFIG_CGROUP_SCHED
8140

8141
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
8142
{
8143
	return css ? container_of(css, struct task_group, css) : NULL;
8144 8145
}

8146 8147
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
8148
{
8149 8150
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
8151

8152
	if (!parent) {
8153
		/* This is early initialization for the top cgroup */
8154
		return &root_task_group.css;
8155 8156
	}

8157
	tg = sched_create_group(parent);
8158 8159 8160
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

8161 8162
	sched_online_group(tg, parent);

8163 8164 8165
	return &tg->css;
}

8166
static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
8167
{
8168
	struct task_group *tg = css_tg(css);
8169

8170
	sched_offline_group(tg);
8171 8172
}

8173
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
8174
{
8175
	struct task_group *tg = css_tg(css);
8176

8177 8178 8179 8180
	/*
	 * Relies on the RCU grace period between css_released() and this.
	 */
	sched_free_group(tg);
8181 8182
}

8183
static void cpu_cgroup_fork(struct task_struct *task)
8184 8185 8186 8187
{
	sched_move_task(task);
}

8188
static int cpu_cgroup_can_attach(struct cgroup_taskset *tset)
8189
{
8190
	struct task_struct *task;
8191
	struct cgroup_subsys_state *css;
8192

8193
	cgroup_taskset_for_each(task, css, tset) {
8194
#ifdef CONFIG_RT_GROUP_SCHED
8195
		if (!sched_rt_can_attach(css_tg(css), task))
8196
			return -EINVAL;
8197
#else
8198 8199 8200
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
8201
#endif
8202
	}
8203 8204
	return 0;
}
8205

8206
static void cpu_cgroup_attach(struct cgroup_taskset *tset)
8207
{
8208
	struct task_struct *task;
8209
	struct cgroup_subsys_state *css;
8210

8211
	cgroup_taskset_for_each(task, css, tset)
8212
		sched_move_task(task);
8213 8214
}

8215
#ifdef CONFIG_FAIR_GROUP_SCHED
8216 8217
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
8218
{
8219
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
8220 8221
}

8222 8223
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
8224
{
8225
	struct task_group *tg = css_tg(css);
8226

8227
	return (u64) scale_load_down(tg->shares);
8228
}
8229 8230

#ifdef CONFIG_CFS_BANDWIDTH
8231 8232
static DEFINE_MUTEX(cfs_constraints_mutex);

8233 8234 8235
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

8236 8237
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

8238 8239
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
8240
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
8241
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8242 8243 8244 8245 8246 8247 8248 8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261

	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;

8262 8263 8264 8265 8266
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
8267 8268 8269 8270 8271
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

8272
	runtime_enabled = quota != RUNTIME_INF;
8273
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
8274 8275 8276 8277 8278 8279
	/*
	 * 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();
8280 8281 8282
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
8283

P
Paul Turner 已提交
8284
	__refill_cfs_bandwidth_runtime(cfs_b);
8285
	/* restart the period timer (if active) to handle new period expiry */
P
Peter Zijlstra 已提交
8286 8287
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
8288 8289
	raw_spin_unlock_irq(&cfs_b->lock);

8290
	for_each_online_cpu(i) {
8291
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
8292
		struct rq *rq = cfs_rq->rq;
8293 8294

		raw_spin_lock_irq(&rq->lock);
8295
		cfs_rq->runtime_enabled = runtime_enabled;
8296
		cfs_rq->runtime_remaining = 0;
8297

8298
		if (cfs_rq->throttled)
8299
			unthrottle_cfs_rq(cfs_rq);
8300 8301
		raw_spin_unlock_irq(&rq->lock);
	}
8302 8303
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
8304 8305
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
8306
	put_online_cpus();
8307

8308
	return ret;
8309 8310 8311 8312 8313 8314
}

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

8315
	period = ktime_to_ns(tg->cfs_bandwidth.period);
8316 8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327
	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;

8328
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
8329 8330
		return -1;

8331
	quota_us = tg->cfs_bandwidth.quota;
8332 8333 8334 8335 8336 8337 8338 8339 8340 8341
	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;
8342
	quota = tg->cfs_bandwidth.quota;
8343 8344 8345 8346 8347 8348 8349 8350

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

8351
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
8352 8353 8354 8355 8356
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

8357 8358
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
8359
{
8360
	return tg_get_cfs_quota(css_tg(css));
8361 8362
}

8363 8364
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
8365
{
8366
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
8367 8368
}

8369 8370
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8371
{
8372
	return tg_get_cfs_period(css_tg(css));
8373 8374
}

8375 8376
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
8377
{
8378
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
8379 8380
}

8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402 8403 8404 8405 8406 8407 8408 8409 8410 8411 8412
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;
8413
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8414 8415 8416 8417 8418
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
8419
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
8420 8421

		quota = normalize_cfs_quota(tg, d);
8422
		parent_quota = parent_b->hierarchical_quota;
8423 8424 8425 8426 8427 8428 8429 8430 8431 8432

		/*
		 * 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;
	}
8433
	cfs_b->hierarchical_quota = quota;
8434 8435 8436 8437 8438 8439

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
8440
	int ret;
8441 8442 8443 8444 8445 8446 8447 8448 8449 8450 8451
	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);
	}

8452 8453 8454 8455 8456
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8457
}
8458

8459
static int cpu_stats_show(struct seq_file *sf, void *v)
8460
{
8461
	struct task_group *tg = css_tg(seq_css(sf));
8462
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8463

8464 8465 8466
	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);
8467 8468 8469

	return 0;
}
8470
#endif /* CONFIG_CFS_BANDWIDTH */
8471
#endif /* CONFIG_FAIR_GROUP_SCHED */
8472

8473
#ifdef CONFIG_RT_GROUP_SCHED
8474 8475
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
8476
{
8477
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
8478 8479
}

8480 8481
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
8482
{
8483
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
8484
}
8485

8486 8487
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
8488
{
8489
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
8490 8491
}

8492 8493
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8494
{
8495
	return sched_group_rt_period(css_tg(css));
8496
}
8497
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
8498

8499
static struct cftype cpu_files[] = {
8500
#ifdef CONFIG_FAIR_GROUP_SCHED
8501 8502
	{
		.name = "shares",
8503 8504
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
8505
	},
8506
#endif
8507 8508 8509 8510 8511 8512 8513 8514 8515 8516 8517
#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,
	},
8518 8519
	{
		.name = "stat",
8520
		.seq_show = cpu_stats_show,
8521
	},
8522
#endif
8523
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
8524
	{
P
Peter Zijlstra 已提交
8525
		.name = "rt_runtime_us",
8526 8527
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
8528
	},
8529 8530
	{
		.name = "rt_period_us",
8531 8532
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
8533
	},
8534
#endif
8535
	{ }	/* terminate */
8536 8537
};

8538
struct cgroup_subsys cpu_cgrp_subsys = {
8539
	.css_alloc	= cpu_cgroup_css_alloc,
8540
	.css_released	= cpu_cgroup_css_released,
8541
	.css_free	= cpu_cgroup_css_free,
8542
	.fork		= cpu_cgroup_fork,
8543 8544
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
8545
	.legacy_cftypes	= cpu_files,
8546
	.early_init	= true,
8547 8548
};

8549
#endif	/* CONFIG_CGROUP_SCHED */
8550

8551 8552 8553 8554 8555
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
8556 8557 8558 8559 8560 8561 8562 8563 8564 8565 8566 8567 8568 8569 8570 8571 8572 8573 8574 8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592 8593 8594 8595 8596

/*
 * 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,
};