core.c 209.0 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 30 31 32
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/nmi.h>
#include <linux/init.h>
33
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
34 35 36
#include <linux/highmem.h>
#include <asm/mmu_context.h>
#include <linux/interrupt.h>
37
#include <linux/capability.h>
L
Linus Torvalds 已提交
38 39
#include <linux/completion.h>
#include <linux/kernel_stat.h>
40
#include <linux/debug_locks.h>
41
#include <linux/perf_event.h>
L
Linus Torvalds 已提交
42 43 44
#include <linux/security.h>
#include <linux/notifier.h>
#include <linux/profile.h>
45
#include <linux/freezer.h>
46
#include <linux/vmalloc.h>
L
Linus Torvalds 已提交
47 48
#include <linux/blkdev.h>
#include <linux/delay.h>
49
#include <linux/pid_namespace.h>
L
Linus Torvalds 已提交
50 51 52 53 54 55 56
#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>
57
#include <linux/proc_fs.h>
L
Linus Torvalds 已提交
58
#include <linux/seq_file.h>
59
#include <linux/sysctl.h>
L
Linus Torvalds 已提交
60 61
#include <linux/syscalls.h>
#include <linux/times.h>
62
#include <linux/tsacct_kern.h>
63
#include <linux/kprobes.h>
64
#include <linux/delayacct.h>
65
#include <linux/unistd.h>
J
Jens Axboe 已提交
66
#include <linux/pagemap.h>
P
Peter Zijlstra 已提交
67
#include <linux/hrtimer.h>
R
Reynes Philippe 已提交
68
#include <linux/tick.h>
P
Peter Zijlstra 已提交
69 70
#include <linux/debugfs.h>
#include <linux/ctype.h>
71
#include <linux/ftrace.h>
72
#include <linux/slab.h>
73
#include <linux/init_task.h>
A
Al Viro 已提交
74
#include <linux/binfmts.h>
75
#include <linux/context_tracking.h>
76
#include <linux/compiler.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
123 124 125 126 127 128 129 130
	0;

#undef SCHED_FEAT

#ifdef CONFIG_SCHED_DEBUG
#define SCHED_FEAT(name, enabled)	\
	#name ,

131
static const char * const sched_feat_names[] = {
132
#include "features.h"
P
Peter Zijlstra 已提交
133 134 135 136
};

#undef SCHED_FEAT

L
Li Zefan 已提交
137
static int sched_feat_show(struct seq_file *m, void *v)
P
Peter Zijlstra 已提交
138 139 140
{
	int i;

141
	for (i = 0; i < __SCHED_FEAT_NR; i++) {
L
Li Zefan 已提交
142 143 144
		if (!(sysctl_sched_features & (1UL << i)))
			seq_puts(m, "NO_");
		seq_printf(m, "%s ", sched_feat_names[i]);
P
Peter Zijlstra 已提交
145
	}
L
Li Zefan 已提交
146
	seq_puts(m, "\n");
P
Peter Zijlstra 已提交
147

L
Li Zefan 已提交
148
	return 0;
P
Peter Zijlstra 已提交
149 150
}

151 152
#ifdef HAVE_JUMP_LABEL

153 154
#define jump_label_key__true  STATIC_KEY_INIT_TRUE
#define jump_label_key__false STATIC_KEY_INIT_FALSE
155 156 157 158

#define SCHED_FEAT(name, enabled)	\
	jump_label_key__##enabled ,

159
struct static_key sched_feat_keys[__SCHED_FEAT_NR] = {
160 161 162 163 164 165 166
#include "features.h"
};

#undef SCHED_FEAT

static void sched_feat_disable(int i)
{
167
	static_key_disable(&sched_feat_keys[i]);
168 169 170 171
}

static void sched_feat_enable(int i)
{
172
	static_key_enable(&sched_feat_keys[i]);
173 174 175 176 177 178
}
#else
static void sched_feat_disable(int i) { };
static void sched_feat_enable(int i) { };
#endif /* HAVE_JUMP_LABEL */

179
static int sched_feat_set(char *cmp)
P
Peter Zijlstra 已提交
180 181
{
	int i;
182
	int neg = 0;
P
Peter Zijlstra 已提交
183

H
Hillf Danton 已提交
184
	if (strncmp(cmp, "NO_", 3) == 0) {
P
Peter Zijlstra 已提交
185 186 187 188
		neg = 1;
		cmp += 3;
	}

189
	for (i = 0; i < __SCHED_FEAT_NR; i++) {
190
		if (strcmp(cmp, sched_feat_names[i]) == 0) {
191
			if (neg) {
P
Peter Zijlstra 已提交
192
				sysctl_sched_features &= ~(1UL << i);
193 194
				sched_feat_disable(i);
			} else {
P
Peter Zijlstra 已提交
195
				sysctl_sched_features |= (1UL << i);
196 197
				sched_feat_enable(i);
			}
P
Peter Zijlstra 已提交
198 199 200 201
			break;
		}
	}

202 203 204 205 206 207 208 209 210 211
	return i;
}

static ssize_t
sched_feat_write(struct file *filp, const char __user *ubuf,
		size_t cnt, loff_t *ppos)
{
	char buf[64];
	char *cmp;
	int i;
212
	struct inode *inode;
213 214 215 216 217 218 219 220 221 222

	if (cnt > 63)
		cnt = 63;

	if (copy_from_user(&buf, ubuf, cnt))
		return -EFAULT;

	buf[cnt] = 0;
	cmp = strstrip(buf);

223 224 225
	/* Ensure the static_key remains in a consistent state */
	inode = file_inode(filp);
	mutex_lock(&inode->i_mutex);
226
	i = sched_feat_set(cmp);
227
	mutex_unlock(&inode->i_mutex);
228
	if (i == __SCHED_FEAT_NR)
P
Peter Zijlstra 已提交
229 230
		return -EINVAL;

231
	*ppos += cnt;
P
Peter Zijlstra 已提交
232 233 234 235

	return cnt;
}

L
Li Zefan 已提交
236 237 238 239 240
static int sched_feat_open(struct inode *inode, struct file *filp)
{
	return single_open(filp, sched_feat_show, NULL);
}

241
static const struct file_operations sched_feat_fops = {
L
Li Zefan 已提交
242 243 244 245 246
	.open		= sched_feat_open,
	.write		= sched_feat_write,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
P
Peter Zijlstra 已提交
247 248 249 250 251 252 253 254 255 256
};

static __init int sched_init_debug(void)
{
	debugfs_create_file("sched_features", 0644, NULL, NULL,
			&sched_feat_fops);

	return 0;
}
late_initcall(sched_init_debug);
257
#endif /* CONFIG_SCHED_DEBUG */
I
Ingo Molnar 已提交
258

259 260 261 262 263 264
/*
 * 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;

265 266 267 268 269 270 271 272
/*
 * 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
273
/*
P
Peter Zijlstra 已提交
274
 * period over which we measure -rt task cpu usage in us.
P
Peter Zijlstra 已提交
275 276
 * default: 1s
 */
P
Peter Zijlstra 已提交
277
unsigned int sysctl_sched_rt_period = 1000000;
P
Peter Zijlstra 已提交
278

279
__read_mostly int scheduler_running;
280

P
Peter Zijlstra 已提交
281 282 283 284 285
/*
 * 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
286

287 288 289
/* cpus with isolated domains */
cpumask_var_t cpu_isolated_map;

L
Linus Torvalds 已提交
290
/*
291
 * this_rq_lock - lock this runqueue and disable interrupts.
L
Linus Torvalds 已提交
292
 */
A
Alexey Dobriyan 已提交
293
static struct rq *this_rq_lock(void)
L
Linus Torvalds 已提交
294 295
	__acquires(rq->lock)
{
296
	struct rq *rq;
L
Linus Torvalds 已提交
297 298 299

	local_irq_disable();
	rq = this_rq();
300
	raw_spin_lock(&rq->lock);
L
Linus Torvalds 已提交
301 302 303 304

	return rq;
}

P
Peter Zijlstra 已提交
305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325
#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());

326
	raw_spin_lock(&rq->lock);
327
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
328
	rq->curr->sched_class->task_tick(rq, rq->curr, 1);
329
	raw_spin_unlock(&rq->lock);
P
Peter Zijlstra 已提交
330 331 332 333

	return HRTIMER_NORESTART;
}

334
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
335

336
static void __hrtick_restart(struct rq *rq)
P
Peter Zijlstra 已提交
337 338 339
{
	struct hrtimer *timer = &rq->hrtick_timer;

340
	hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
P
Peter Zijlstra 已提交
341 342
}

343 344 345 346
/*
 * called from hardirq (IPI) context
 */
static void __hrtick_start(void *arg)
347
{
348
	struct rq *rq = arg;
349

350
	raw_spin_lock(&rq->lock);
P
Peter Zijlstra 已提交
351
	__hrtick_restart(rq);
352
	rq->hrtick_csd_pending = 0;
353
	raw_spin_unlock(&rq->lock);
354 355
}

356 357 358 359 360
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
361
void hrtick_start(struct rq *rq, u64 delay)
362
{
363
	struct hrtimer *timer = &rq->hrtick_timer;
364 365 366 367 368 369 370 371 372
	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);
373

374
	hrtimer_set_expires(timer, time);
375 376

	if (rq == this_rq()) {
P
Peter Zijlstra 已提交
377
		__hrtick_restart(rq);
378
	} else if (!rq->hrtick_csd_pending) {
379
		smp_call_function_single_async(cpu_of(rq), &rq->hrtick_csd);
380 381
		rq->hrtick_csd_pending = 1;
	}
382 383 384 385 386 387 388 389 390 391 392 393 394 395
}

static int
hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu)
{
	int cpu = (int)(long)hcpu;

	switch (action) {
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
	case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
396
		hrtick_clear(cpu_rq(cpu));
397 398 399 400 401 402
		return NOTIFY_OK;
	}

	return NOTIFY_DONE;
}

403
static __init void init_hrtick(void)
404 405 406
{
	hotcpu_notifier(hotplug_hrtick, 0);
}
407 408 409 410 411 412
#else
/*
 * Called to set the hrtick timer state.
 *
 * called with rq->lock held and irqs disabled
 */
413
void hrtick_start(struct rq *rq, u64 delay)
414
{
W
Wanpeng Li 已提交
415 416 417 418 419
	/*
	 * Don't schedule slices shorter than 10000ns, that just
	 * doesn't make sense. Rely on vruntime for fairness.
	 */
	delay = max_t(u64, delay, 10000LL);
420 421
	hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay),
		      HRTIMER_MODE_REL_PINNED);
422
}
423

A
Andrew Morton 已提交
424
static inline void init_hrtick(void)
P
Peter Zijlstra 已提交
425 426
{
}
427
#endif /* CONFIG_SMP */
P
Peter Zijlstra 已提交
428

429
static void init_rq_hrtick(struct rq *rq)
P
Peter Zijlstra 已提交
430
{
431 432
#ifdef CONFIG_SMP
	rq->hrtick_csd_pending = 0;
P
Peter Zijlstra 已提交
433

434 435 436 437
	rq->hrtick_csd.flags = 0;
	rq->hrtick_csd.func = __hrtick_start;
	rq->hrtick_csd.info = rq;
#endif
P
Peter Zijlstra 已提交
438

439 440
	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
	rq->hrtick_timer.function = hrtick;
P
Peter Zijlstra 已提交
441
}
A
Andrew Morton 已提交
442
#else	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
443 444 445 446 447 448 449 450
static inline void hrtick_clear(struct rq *rq)
{
}

static inline void init_rq_hrtick(struct rq *rq)
{
}

451 452 453
static inline void init_hrtick(void)
{
}
A
Andrew Morton 已提交
454
#endif	/* CONFIG_SCHED_HRTICK */
P
Peter Zijlstra 已提交
455

456 457 458 459 460 461 462 463 464 465 466 467 468 469
/*
 * cmpxchg based fetch_or, macro so it works for different integer types
 */
#define fetch_or(ptr, val)						\
({	typeof(*(ptr)) __old, __val = *(ptr);				\
 	for (;;) {							\
 		__old = cmpxchg((ptr), __val, __val | (val));		\
 		if (__old == __val)					\
 			break;						\
 		__val = __old;						\
 	}								\
 	__old;								\
})

470
#if defined(CONFIG_SMP) && defined(TIF_POLLING_NRFLAG)
471 472 473 474 475 476 477 478 479 480
/*
 * 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);
}
481 482 483 484 485 486 487 488 489 490

/*
 * 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);
491
	typeof(ti->flags) old, val = READ_ONCE(ti->flags);
492 493 494 495 496 497 498 499 500 501 502 503 504 505

	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;
}

506 507 508 509 510 511
#else
static bool set_nr_and_not_polling(struct task_struct *p)
{
	set_tsk_need_resched(p);
	return true;
}
512 513 514 515 516 517 518

#ifdef CONFIG_SMP
static bool set_nr_if_polling(struct task_struct *p)
{
	return false;
}
#endif
519 520
#endif

521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566
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
	 * barrier implied by the wakeup in wake_up_list().
	 */
	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
567
/*
568
 * resched_curr - mark rq's current task 'to be rescheduled now'.
I
Ingo Molnar 已提交
569 570 571 572 573
 *
 * 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.
 */
574
void resched_curr(struct rq *rq)
I
Ingo Molnar 已提交
575
{
576
	struct task_struct *curr = rq->curr;
I
Ingo Molnar 已提交
577 578
	int cpu;

579
	lockdep_assert_held(&rq->lock);
I
Ingo Molnar 已提交
580

581
	if (test_tsk_need_resched(curr))
I
Ingo Molnar 已提交
582 583
		return;

584
	cpu = cpu_of(rq);
585

586
	if (cpu == smp_processor_id()) {
587
		set_tsk_need_resched(curr);
588
		set_preempt_need_resched();
I
Ingo Molnar 已提交
589
		return;
590
	}
I
Ingo Molnar 已提交
591

592
	if (set_nr_and_not_polling(curr))
I
Ingo Molnar 已提交
593
		smp_send_reschedule(cpu);
594 595
	else
		trace_sched_wake_idle_without_ipi(cpu);
I
Ingo Molnar 已提交
596 597
}

598
void resched_cpu(int cpu)
I
Ingo Molnar 已提交
599 600 601 602
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

603
	if (!raw_spin_trylock_irqsave(&rq->lock, flags))
I
Ingo Molnar 已提交
604
		return;
605
	resched_curr(rq);
606
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
607
}
608

609
#ifdef CONFIG_SMP
610
#ifdef CONFIG_NO_HZ_COMMON
611 612 613 614 615 616 617 618
/*
 * 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).
 */
619
int get_nohz_timer_target(void)
620
{
621
	int i, cpu = smp_processor_id();
622 623
	struct sched_domain *sd;

624
	if (!idle_cpu(cpu) && is_housekeeping_cpu(cpu))
625 626
		return cpu;

627
	rcu_read_lock();
628
	for_each_domain(cpu, sd) {
629
		for_each_cpu(i, sched_domain_span(sd)) {
630
			if (!idle_cpu(i) && is_housekeeping_cpu(cpu)) {
631 632 633 634
				cpu = i;
				goto unlock;
			}
		}
635
	}
636 637 638

	if (!is_housekeeping_cpu(cpu))
		cpu = housekeeping_any_cpu();
639 640
unlock:
	rcu_read_unlock();
641 642
	return cpu;
}
643 644 645 646 647 648 649 650 651 652
/*
 * 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.
 */
653
static void wake_up_idle_cpu(int cpu)
654 655 656 657 658 659
{
	struct rq *rq = cpu_rq(cpu);

	if (cpu == smp_processor_id())
		return;

660
	if (set_nr_and_not_polling(rq->idle))
661
		smp_send_reschedule(cpu);
662 663
	else
		trace_sched_wake_idle_without_ipi(cpu);
664 665
}

666
static bool wake_up_full_nohz_cpu(int cpu)
667
{
668 669 670 671 672 673
	/*
	 * 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.
	 */
674
	if (tick_nohz_full_cpu(cpu)) {
675 676
		if (cpu != smp_processor_id() ||
		    tick_nohz_tick_stopped())
677
			tick_nohz_full_kick_cpu(cpu);
678 679 680 681 682 683 684 685
		return true;
	}

	return false;
}

void wake_up_nohz_cpu(int cpu)
{
686
	if (!wake_up_full_nohz_cpu(cpu))
687 688 689
		wake_up_idle_cpu(cpu);
}

690
static inline bool got_nohz_idle_kick(void)
691
{
692
	int cpu = smp_processor_id();
693 694 695 696 697 698 699 700 701 702 703 704 705

	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;
706 707
}

708
#else /* CONFIG_NO_HZ_COMMON */
709

710
static inline bool got_nohz_idle_kick(void)
P
Peter Zijlstra 已提交
711
{
712
	return false;
P
Peter Zijlstra 已提交
713 714
}

715
#endif /* CONFIG_NO_HZ_COMMON */
716

717 718 719
#ifdef CONFIG_NO_HZ_FULL
bool sched_can_stop_tick(void)
{
720 721 722 723 724 725 726 727 728 729 730 731 732 733
	/*
	 * FIFO realtime policy runs the highest priority task. Other runnable
	 * tasks are of a lower priority. The scheduler tick does nothing.
	 */
	if (current->policy == SCHED_FIFO)
		return true;

	/*
	 * Round-robin realtime tasks time slice with other tasks at the same
	 * realtime priority. Is this task the only one at this priority?
	 */
	if (current->policy == SCHED_RR) {
		struct sched_rt_entity *rt_se = &current->rt;

734
		return list_is_singular(&rt_se->run_list);
735 736
	}

737 738 739 740 741
	/*
	 * More than one running task need preemption.
	 * nr_running update is assumed to be visible
	 * after IPI is sent from wakers.
	 */
742 743
	if (this_rq()->nr_running > 1)
		return false;
744

745
	return true;
746 747
}
#endif /* CONFIG_NO_HZ_FULL */
748

749
void sched_avg_update(struct rq *rq)
750
{
751 752
	s64 period = sched_avg_period();

753
	while ((s64)(rq_clock(rq) - rq->age_stamp) > period) {
754 755 756 757 758 759
		/*
		 * 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));
760 761 762
		rq->age_stamp += period;
		rq->rt_avg /= 2;
	}
763 764
}

765
#endif /* CONFIG_SMP */
766

767 768
#if defined(CONFIG_RT_GROUP_SCHED) || (defined(CONFIG_FAIR_GROUP_SCHED) && \
			(defined(CONFIG_SMP) || defined(CONFIG_CFS_BANDWIDTH)))
769
/*
770 771 772 773
 * 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.
774
 */
775
int walk_tg_tree_from(struct task_group *from,
776
			     tg_visitor down, tg_visitor up, void *data)
777 778
{
	struct task_group *parent, *child;
P
Peter Zijlstra 已提交
779
	int ret;
780

781 782
	parent = from;

783
down:
P
Peter Zijlstra 已提交
784 785
	ret = (*down)(parent, data);
	if (ret)
786
		goto out;
787 788 789 790 791 792 793
	list_for_each_entry_rcu(child, &parent->children, siblings) {
		parent = child;
		goto down;

up:
		continue;
	}
P
Peter Zijlstra 已提交
794
	ret = (*up)(parent, data);
795 796
	if (ret || parent == from)
		goto out;
797 798 799 800 801

	child = parent;
	parent = parent->parent;
	if (parent)
		goto up;
802
out:
P
Peter Zijlstra 已提交
803
	return ret;
804 805
}

806
int tg_nop(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
807
{
808
	return 0;
P
Peter Zijlstra 已提交
809
}
810 811
#endif

812 813
static void set_load_weight(struct task_struct *p)
{
N
Nikhil Rao 已提交
814 815 816
	int prio = p->static_prio - MAX_RT_PRIO;
	struct load_weight *load = &p->se.load;

I
Ingo Molnar 已提交
817 818 819
	/*
	 * SCHED_IDLE tasks get minimal weight:
	 */
820
	if (idle_policy(p->policy)) {
821
		load->weight = scale_load(WEIGHT_IDLEPRIO);
N
Nikhil Rao 已提交
822
		load->inv_weight = WMULT_IDLEPRIO;
I
Ingo Molnar 已提交
823 824
		return;
	}
825

826 827
	load->weight = scale_load(sched_prio_to_weight[prio]);
	load->inv_weight = sched_prio_to_wmult[prio];
828 829
}

830
static inline void enqueue_task(struct rq *rq, struct task_struct *p, int flags)
831
{
832
	update_rq_clock(rq);
833 834
	if (!(flags & ENQUEUE_RESTORE))
		sched_info_queued(rq, p);
835
	p->sched_class->enqueue_task(rq, p, flags);
836 837
}

838
static inline void dequeue_task(struct rq *rq, struct task_struct *p, int flags)
839
{
840
	update_rq_clock(rq);
841 842
	if (!(flags & DEQUEUE_SAVE))
		sched_info_dequeued(rq, p);
843
	p->sched_class->dequeue_task(rq, p, flags);
844 845
}

846
void activate_task(struct rq *rq, struct task_struct *p, int flags)
847 848 849 850
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible--;

851
	enqueue_task(rq, p, flags);
852 853
}

854
void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
855 856 857 858
{
	if (task_contributes_to_load(p))
		rq->nr_uninterruptible++;

859
	dequeue_task(rq, p, flags);
860 861
}

862
static void update_rq_clock_task(struct rq *rq, s64 delta)
863
{
864 865 866 867 868 869 870 871
/*
 * 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
872
	irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time;
873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893

	/*
	 * 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;
894 895
#endif
#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
896
	if (static_key_false((&paravirt_steal_rq_enabled))) {
897 898 899 900 901 902 903 904 905 906 907
		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

908 909
	rq->clock_task += delta;

910
#if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING)
911
	if ((irq_delta + steal) && sched_feat(NONTASK_CAPACITY))
912 913
		sched_rt_avg_update(rq, irq_delta + steal);
#endif
914 915
}

916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945
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;
	}
}

946
/*
I
Ingo Molnar 已提交
947
 * __normal_prio - return the priority that is based on the static prio
948 949 950
 */
static inline int __normal_prio(struct task_struct *p)
{
I
Ingo Molnar 已提交
951
	return p->static_prio;
952 953
}

954 955 956 957 958 959 960
/*
 * 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.
 */
961
static inline int normal_prio(struct task_struct *p)
962 963 964
{
	int prio;

965 966 967
	if (task_has_dl_policy(p))
		prio = MAX_DL_PRIO-1;
	else if (task_has_rt_policy(p))
968 969 970 971 972 973 974 975 976 977 978 979 980
		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.
 */
981
static int effective_prio(struct task_struct *p)
982 983 984 985 986 987 988 989 990 991 992 993
{
	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
994 995 996
/**
 * task_curr - is this task currently executing on a CPU?
 * @p: the task in question.
997 998
 *
 * Return: 1 if the task is currently executing. 0 otherwise.
L
Linus Torvalds 已提交
999
 */
1000
inline int task_curr(const struct task_struct *p)
L
Linus Torvalds 已提交
1001 1002 1003 1004
{
	return cpu_curr(task_cpu(p)) == p;
}

1005
/*
1006 1007 1008 1009 1010
 * 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().
1011
 */
1012 1013
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
1014
				       int oldprio)
1015 1016 1017
{
	if (prev_class != p->sched_class) {
		if (prev_class->switched_from)
P
Peter Zijlstra 已提交
1018
			prev_class->switched_from(rq, p);
1019

P
Peter Zijlstra 已提交
1020
		p->sched_class->switched_to(rq, p);
1021
	} else if (oldprio != p->prio || dl_task(p))
P
Peter Zijlstra 已提交
1022
		p->sched_class->prio_changed(rq, p, oldprio);
1023 1024
}

1025
void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags)
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
{
	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) {
1036
				resched_curr(rq);
1037 1038 1039 1040 1041 1042 1043 1044 1045
				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.
	 */
1046
	if (task_on_rq_queued(rq->curr) && test_tsk_need_resched(rq->curr))
1047
		rq_clock_skip_update(rq, true);
1048 1049
}

L
Linus Torvalds 已提交
1050
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069
/*
 * 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.
 */
1070
static struct rq *move_queued_task(struct rq *rq, struct task_struct *p, int new_cpu)
P
Peter Zijlstra 已提交
1071 1072 1073 1074
{
	lockdep_assert_held(&rq->lock);

	p->on_rq = TASK_ON_RQ_MIGRATING;
1075
	dequeue_task(rq, p, 0);
P
Peter Zijlstra 已提交
1076 1077 1078 1079 1080 1081 1082 1083
	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);
1084
	p->on_rq = TASK_ON_RQ_QUEUED;
P
Peter Zijlstra 已提交
1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
	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.
 */
1104
static struct rq *__migrate_task(struct rq *rq, struct task_struct *p, int dest_cpu)
P
Peter Zijlstra 已提交
1105 1106
{
	if (unlikely(!cpu_active(dest_cpu)))
1107
		return rq;
P
Peter Zijlstra 已提交
1108 1109 1110

	/* Affinity changed (again). */
	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
1111
		return rq;
P
Peter Zijlstra 已提交
1112

1113 1114 1115
	rq = move_queued_task(rq, p, dest_cpu);

	return rq;
P
Peter Zijlstra 已提交
1116 1117 1118 1119 1120 1121 1122 1123 1124 1125
}

/*
 * 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;
1126 1127
	struct task_struct *p = arg->task;
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139

	/*
	 * 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();
1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152

	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
1153 1154 1155 1156
	local_irq_enable();
	return 0;
}

1157 1158 1159 1160 1161
/*
 * 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
1162 1163 1164 1165 1166
{
	cpumask_copy(&p->cpus_allowed, new_mask);
	p->nr_cpus_allowed = cpumask_weight(new_mask);
}

1167 1168
void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
{
1169 1170 1171
	struct rq *rq = task_rq(p);
	bool queued, running;

1172
	lockdep_assert_held(&p->pi_lock);
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182

	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);
1183
		dequeue_task(rq, p, DEQUEUE_SAVE);
1184 1185 1186 1187
	}
	if (running)
		put_prev_task(rq, p);

1188
	p->sched_class->set_cpus_allowed(p, new_mask);
1189 1190 1191 1192

	if (running)
		p->sched_class->set_curr_task(rq);
	if (queued)
1193
		enqueue_task(rq, p, ENQUEUE_RESTORE);
1194 1195
}

P
Peter Zijlstra 已提交
1196 1197 1198 1199 1200 1201 1202 1203 1204
/*
 * 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.
 */
1205 1206
static int __set_cpus_allowed_ptr(struct task_struct *p,
				  const struct cpumask *new_mask, bool check)
P
Peter Zijlstra 已提交
1207 1208 1209 1210 1211 1212 1213 1214
{
	unsigned long flags;
	struct rq *rq;
	unsigned int dest_cpu;
	int ret = 0;

	rq = task_rq_lock(p, &flags);

1215 1216 1217 1218 1219 1220 1221 1222 1223
	/*
	 * 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
1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245
	if (cpumask_equal(&p->cpus_allowed, new_mask))
		goto out;

	if (!cpumask_intersects(new_mask, cpu_active_mask)) {
		ret = -EINVAL;
		goto out;
	}

	do_set_cpus_allowed(p, new_mask);

	/* 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;

	dest_cpu = cpumask_any_and(cpu_active_mask, new_mask);
	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. */
		task_rq_unlock(rq, p, &flags);
		stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
		tlb_migrate_finish(p->mm);
		return 0;
1246 1247 1248 1249 1250 1251
	} else if (task_on_rq_queued(p)) {
		/*
		 * OK, since we're going to drop the lock immediately
		 * afterwards anyway.
		 */
		lockdep_unpin_lock(&rq->lock);
1252
		rq = move_queued_task(rq, p, dest_cpu);
1253 1254
		lockdep_pin_lock(&rq->lock);
	}
P
Peter Zijlstra 已提交
1255 1256 1257 1258 1259
out:
	task_rq_unlock(rq, p, &flags);

	return ret;
}
1260 1261 1262 1263 1264

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
1265 1266
EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);

I
Ingo Molnar 已提交
1267
void set_task_cpu(struct task_struct *p, unsigned int new_cpu)
I
Ingo Molnar 已提交
1268
{
1269 1270 1271 1272 1273
#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
1274
	WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING &&
O
Oleg Nesterov 已提交
1275
			!p->on_rq);
1276

1277 1278 1279 1280 1281 1282 1283 1284 1285
	/*
	 * 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)));

1286
#ifdef CONFIG_LOCKDEP
1287 1288 1289 1290 1291
	/*
	 * 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
1292
	 * see task_group().
1293 1294 1295 1296
	 *
	 * Furthermore, all task_rq users should acquire both locks, see
	 * task_rq_lock().
	 */
1297 1298 1299
	WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) ||
				      lockdep_is_held(&task_rq(p)->lock)));
#endif
1300 1301
#endif

1302
	trace_sched_migrate_task(p, new_cpu);
1303

1304
	if (task_cpu(p) != new_cpu) {
1305
		if (p->sched_class->migrate_task_rq)
1306
			p->sched_class->migrate_task_rq(p);
1307
		p->se.nr_migrations++;
1308
		perf_event_task_migrate(p);
1309
	}
I
Ingo Molnar 已提交
1310 1311

	__set_task_cpu(p, new_cpu);
I
Ingo Molnar 已提交
1312 1313
}

1314 1315
static void __migrate_swap_task(struct task_struct *p, int cpu)
{
1316
	if (task_on_rq_queued(p)) {
1317 1318 1319 1320 1321
		struct rq *src_rq, *dst_rq;

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

1322
		p->on_rq = TASK_ON_RQ_MIGRATING;
1323 1324 1325
		deactivate_task(src_rq, p, 0);
		set_task_cpu(p, cpu);
		activate_task(dst_rq, p, 0);
1326
		p->on_rq = TASK_ON_RQ_QUEUED;
1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348
		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;

1349 1350 1351
	if (!cpu_active(arg->src_cpu) || !cpu_active(arg->dst_cpu))
		return -EAGAIN;

1352 1353 1354
	src_rq = cpu_rq(arg->src_cpu);
	dst_rq = cpu_rq(arg->dst_cpu);

1355 1356
	double_raw_lock(&arg->src_task->pi_lock,
			&arg->dst_task->pi_lock);
1357
	double_rq_lock(src_rq, dst_rq);
1358

1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377
	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);
1378 1379
	raw_spin_unlock(&arg->dst_task->pi_lock);
	raw_spin_unlock(&arg->src_task->pi_lock);
1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401

	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;

1402 1403 1404 1405
	/*
	 * 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.
	 */
1406 1407 1408 1409 1410 1411 1412 1413 1414
	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;

1415
	trace_sched_swap_numa(cur, arg.src_cpu, p, arg.dst_cpu);
1416 1417 1418 1419 1420 1421
	ret = stop_two_cpus(arg.dst_cpu, arg.src_cpu, migrate_swap_stop, &arg);

out:
	return ret;
}

L
Linus Torvalds 已提交
1422 1423 1424
/*
 * wait_task_inactive - wait for a thread to unschedule.
 *
R
Roland McGrath 已提交
1425 1426 1427 1428 1429 1430 1431
 * 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
1432 1433 1434 1435 1436 1437
 * 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
1438
unsigned long wait_task_inactive(struct task_struct *p, long match_state)
L
Linus Torvalds 已提交
1439 1440
{
	unsigned long flags;
1441
	int running, queued;
R
Roland McGrath 已提交
1442
	unsigned long ncsw;
1443
	struct rq *rq;
L
Linus Torvalds 已提交
1444

1445 1446 1447 1448 1449 1450 1451 1452
	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);
1453

1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464
		/*
		 * 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
1465 1466 1467
		while (task_running(rq, p)) {
			if (match_state && unlikely(p->state != match_state))
				return 0;
1468
			cpu_relax();
R
Roland McGrath 已提交
1469
		}
1470

1471 1472 1473 1474 1475 1476
		/*
		 * 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.
		 */
		rq = task_rq_lock(p, &flags);
1477
		trace_sched_wait_task(p);
1478
		running = task_running(rq, p);
1479
		queued = task_on_rq_queued(p);
R
Roland McGrath 已提交
1480
		ncsw = 0;
1481
		if (!match_state || p->state == match_state)
1482
			ncsw = p->nvcsw | LONG_MIN; /* sets MSB */
1483
		task_rq_unlock(rq, p, &flags);
1484

R
Roland McGrath 已提交
1485 1486 1487 1488 1489 1490
		/*
		 * If it changed from the expected state, bail out now.
		 */
		if (unlikely(!ncsw))
			break;

1491 1492 1493 1494 1495 1496 1497 1498 1499 1500
		/*
		 * 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;
		}
1501

1502 1503 1504 1505 1506
		/*
		 * It's not enough that it's not actively running,
		 * it must be off the runqueue _entirely_, and not
		 * preempted!
		 *
1507
		 * So if it was still runnable (but just not actively
1508 1509 1510
		 * running right now), it's preempted, and we should
		 * yield - it could be a while.
		 */
1511
		if (unlikely(queued)) {
1512 1513 1514 1515
			ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ);

			set_current_state(TASK_UNINTERRUPTIBLE);
			schedule_hrtimeout(&to, HRTIMER_MODE_REL);
1516 1517
			continue;
		}
1518

1519 1520 1521 1522 1523 1524 1525
		/*
		 * 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
1526 1527

	return ncsw;
L
Linus Torvalds 已提交
1528 1529 1530 1531 1532 1533 1534 1535 1536
}

/***
 * 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
1537
 * NOTE: this function doesn't have to take the runqueue lock,
L
Linus Torvalds 已提交
1538 1539 1540 1541 1542
 * 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.
 */
1543
void kick_process(struct task_struct *p)
L
Linus Torvalds 已提交
1544 1545 1546 1547 1548 1549 1550 1551 1552
{
	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
1553
EXPORT_SYMBOL_GPL(kick_process);
L
Linus Torvalds 已提交
1554

1555
/*
1556
 * ->cpus_allowed is protected by both rq->lock and p->pi_lock
1557
 */
1558 1559
static int select_fallback_rq(int cpu, struct task_struct *p)
{
1560 1561
	int nid = cpu_to_node(cpu);
	const struct cpumask *nodemask = NULL;
1562 1563
	enum { cpuset, possible, fail } state = cpuset;
	int dest_cpu;
1564

1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581
	/*
	 * 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_online(dest_cpu))
				continue;
			if (!cpu_active(dest_cpu))
				continue;
			if (cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
				return dest_cpu;
		}
1582
	}
1583

1584 1585
	for (;;) {
		/* Any allowed, online CPU? */
1586
		for_each_cpu(dest_cpu, tsk_cpus_allowed(p)) {
1587 1588 1589 1590 1591 1592
			if (!cpu_online(dest_cpu))
				continue;
			if (!cpu_active(dest_cpu))
				continue;
			goto out;
		}
1593

1594
		/* No more Mr. Nice Guy. */
1595 1596
		switch (state) {
		case cpuset:
1597 1598 1599 1600 1601 1602
			if (IS_ENABLED(CONFIG_CPUSETS)) {
				cpuset_cpus_allowed_fallback(p);
				state = possible;
				break;
			}
			/* fall-through */
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621
		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()) {
1622
			printk_deferred("process %d (%s) no longer affine to cpu%d\n",
1623 1624
					task_pid_nr(p), p->comm, cpu);
		}
1625 1626 1627 1628 1629
	}

	return dest_cpu;
}

1630
/*
1631
 * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable.
1632
 */
1633
static inline
1634
int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
1635
{
1636 1637
	lockdep_assert_held(&p->pi_lock);

1638 1639
	if (p->nr_cpus_allowed > 1)
		cpu = p->sched_class->select_task_rq(p, cpu, sd_flags, wake_flags);
1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650

	/*
	 * 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 ]
	 */
1651
	if (unlikely(!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)) ||
P
Peter Zijlstra 已提交
1652
		     !cpu_online(cpu)))
1653
		cpu = select_fallback_rq(task_cpu(p), p);
1654 1655

	return cpu;
1656
}
1657 1658 1659 1660 1661 1662

static void update_avg(u64 *avg, u64 sample)
{
	s64 diff = sample - *avg;
	*avg += diff >> 3;
}
1663 1664 1665 1666 1667 1668 1669 1670 1671

#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
1672
#endif /* CONFIG_SMP */
1673

P
Peter Zijlstra 已提交
1674
static void
1675
ttwu_stat(struct task_struct *p, int cpu, int wake_flags)
T
Tejun Heo 已提交
1676
{
P
Peter Zijlstra 已提交
1677
#ifdef CONFIG_SCHEDSTATS
1678 1679
	struct rq *rq = this_rq();

P
Peter Zijlstra 已提交
1680 1681 1682 1683 1684 1685 1686 1687 1688 1689
#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);
1690
		rcu_read_lock();
P
Peter Zijlstra 已提交
1691 1692 1693 1694 1695 1696
		for_each_domain(this_cpu, sd) {
			if (cpumask_test_cpu(cpu, sched_domain_span(sd))) {
				schedstat_inc(sd, ttwu_wake_remote);
				break;
			}
		}
1697
		rcu_read_unlock();
P
Peter Zijlstra 已提交
1698
	}
1699 1700 1701 1702

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

P
Peter Zijlstra 已提交
1703 1704 1705
#endif /* CONFIG_SMP */

	schedstat_inc(rq, ttwu_count);
T
Tejun Heo 已提交
1706
	schedstat_inc(p, se.statistics.nr_wakeups);
P
Peter Zijlstra 已提交
1707 1708

	if (wake_flags & WF_SYNC)
T
Tejun Heo 已提交
1709
		schedstat_inc(p, se.statistics.nr_wakeups_sync);
P
Peter Zijlstra 已提交
1710 1711 1712 1713

#endif /* CONFIG_SCHEDSTATS */
}

1714
static inline void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags)
P
Peter Zijlstra 已提交
1715
{
T
Tejun Heo 已提交
1716
	activate_task(rq, p, en_flags);
1717
	p->on_rq = TASK_ON_RQ_QUEUED;
1718 1719 1720 1721

	/* 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
1722 1723
}

1724 1725 1726
/*
 * Mark the task runnable and perform wakeup-preemption.
 */
1727
static void
1728
ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags)
T
Tejun Heo 已提交
1729 1730 1731
{
	check_preempt_curr(rq, p, wake_flags);
	p->state = TASK_RUNNING;
1732 1733
	trace_sched_wakeup(p);

T
Tejun Heo 已提交
1734
#ifdef CONFIG_SMP
1735 1736
	if (p->sched_class->task_woken) {
		/*
1737 1738
		 * Our task @p is fully woken up and running; so its safe to
		 * drop the rq->lock, hereafter rq is only used for statistics.
1739
		 */
1740
		lockdep_unpin_lock(&rq->lock);
T
Tejun Heo 已提交
1741
		p->sched_class->task_woken(rq, p);
1742
		lockdep_pin_lock(&rq->lock);
1743
	}
T
Tejun Heo 已提交
1744

1745
	if (rq->idle_stamp) {
1746
		u64 delta = rq_clock(rq) - rq->idle_stamp;
1747
		u64 max = 2*rq->max_idle_balance_cost;
T
Tejun Heo 已提交
1748

1749 1750 1751
		update_avg(&rq->avg_idle, delta);

		if (rq->avg_idle > max)
T
Tejun Heo 已提交
1752
			rq->avg_idle = max;
1753

T
Tejun Heo 已提交
1754 1755 1756 1757 1758
		rq->idle_stamp = 0;
	}
#endif
}

1759 1760 1761
static void
ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
{
1762 1763
	lockdep_assert_held(&rq->lock);

1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784
#ifdef CONFIG_SMP
	if (p->sched_contributes_to_load)
		rq->nr_uninterruptible--;
#endif

	ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING);
	ttwu_do_wakeup(rq, p, wake_flags);
}

/*
 * 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)
{
	struct rq *rq;
	int ret = 0;

	rq = __task_rq_lock(p);
1785
	if (task_on_rq_queued(p)) {
1786 1787
		/* check_preempt_curr() may use rq clock */
		update_rq_clock(rq);
1788 1789 1790 1791 1792 1793 1794 1795
		ttwu_do_wakeup(rq, p, wake_flags);
		ret = 1;
	}
	__task_rq_unlock(rq);

	return ret;
}

1796
#ifdef CONFIG_SMP
1797
void sched_ttwu_pending(void)
1798 1799
{
	struct rq *rq = this_rq();
P
Peter Zijlstra 已提交
1800 1801
	struct llist_node *llist = llist_del_all(&rq->wake_list);
	struct task_struct *p;
1802
	unsigned long flags;
1803

1804 1805 1806 1807
	if (!llist)
		return;

	raw_spin_lock_irqsave(&rq->lock, flags);
1808
	lockdep_pin_lock(&rq->lock);
1809

P
Peter Zijlstra 已提交
1810 1811 1812
	while (llist) {
		p = llist_entry(llist, struct task_struct, wake_entry);
		llist = llist_next(llist);
1813 1814 1815
		ttwu_do_activate(rq, p, 0);
	}

1816
	lockdep_unpin_lock(&rq->lock);
1817
	raw_spin_unlock_irqrestore(&rq->lock, flags);
1818 1819 1820 1821
}

void scheduler_ipi(void)
{
1822 1823 1824 1825 1826
	/*
	 * Fold TIF_NEED_RESCHED into the preempt_count; anybody setting
	 * TIF_NEED_RESCHED remotely (for the first time) will also send
	 * this IPI.
	 */
1827
	preempt_fold_need_resched();
1828

1829
	if (llist_empty(&this_rq()->wake_list) && !got_nohz_idle_kick())
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
		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
1846
	sched_ttwu_pending();
1847 1848 1849 1850

	/*
	 * Check if someone kicked us for doing the nohz idle load balance.
	 */
1851
	if (unlikely(got_nohz_idle_kick())) {
1852
		this_rq()->idle_balance = 1;
1853
		raise_softirq_irqoff(SCHED_SOFTIRQ);
1854
	}
1855
	irq_exit();
1856 1857 1858 1859
}

static void ttwu_queue_remote(struct task_struct *p, int cpu)
{
1860 1861 1862 1863 1864 1865 1866 1867
	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);
	}
1868
}
1869

1870 1871 1872 1873 1874
void wake_up_if_idle(int cpu)
{
	struct rq *rq = cpu_rq(cpu);
	unsigned long flags;

1875 1876 1877 1878
	rcu_read_lock();

	if (!is_idle_task(rcu_dereference(rq->curr)))
		goto out;
1879 1880 1881 1882 1883 1884 1885 1886 1887 1888

	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);
	}
1889 1890 1891

out:
	rcu_read_unlock();
1892 1893
}

1894
bool cpus_share_cache(int this_cpu, int that_cpu)
1895 1896 1897
{
	return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
}
1898
#endif /* CONFIG_SMP */
1899

1900 1901 1902 1903
static void ttwu_queue(struct task_struct *p, int cpu)
{
	struct rq *rq = cpu_rq(cpu);

1904
#if defined(CONFIG_SMP)
1905
	if (sched_feat(TTWU_QUEUE) && !cpus_share_cache(smp_processor_id(), cpu)) {
1906
		sched_clock_cpu(cpu); /* sync clocks x-cpu */
1907 1908 1909 1910 1911
		ttwu_queue_remote(p, cpu);
		return;
	}
#endif

1912
	raw_spin_lock(&rq->lock);
1913
	lockdep_pin_lock(&rq->lock);
1914
	ttwu_do_activate(rq, p, 0);
1915
	lockdep_unpin_lock(&rq->lock);
1916
	raw_spin_unlock(&rq->lock);
T
Tejun Heo 已提交
1917 1918 1919
}

/**
L
Linus Torvalds 已提交
1920
 * try_to_wake_up - wake up a thread
T
Tejun Heo 已提交
1921
 * @p: the thread to be awakened
L
Linus Torvalds 已提交
1922
 * @state: the mask of task states that can be woken
T
Tejun Heo 已提交
1923
 * @wake_flags: wake modifier flags (WF_*)
L
Linus Torvalds 已提交
1924 1925 1926 1927 1928 1929 1930
 *
 * 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.
 *
1931
 * Return: %true if @p was woken up, %false if it was already running.
T
Tejun Heo 已提交
1932
 * or @state didn't match @p's state.
L
Linus Torvalds 已提交
1933
 */
1934 1935
static int
try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
L
Linus Torvalds 已提交
1936 1937
{
	unsigned long flags;
1938
	int cpu, success = 0;
P
Peter Zijlstra 已提交
1939

1940 1941 1942 1943 1944 1945 1946
	/*
	 * 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();
1947
	raw_spin_lock_irqsave(&p->pi_lock, flags);
P
Peter Zijlstra 已提交
1948
	if (!(p->state & state))
L
Linus Torvalds 已提交
1949 1950
		goto out;

1951 1952
	trace_sched_waking(p);

1953
	success = 1; /* we're going to change ->state */
L
Linus Torvalds 已提交
1954 1955
	cpu = task_cpu(p);

1956 1957
	if (p->on_rq && ttwu_remote(p, wake_flags))
		goto stat;
L
Linus Torvalds 已提交
1958 1959

#ifdef CONFIG_SMP
1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978
	/*
	 * 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
1979
	/*
1980 1981
	 * If the owning (remote) cpu is still in the middle of schedule() with
	 * this task as prev, wait until its done referencing the task.
P
Peter Zijlstra 已提交
1982
	 */
1983
	while (p->on_cpu)
1984
		cpu_relax();
1985
	/*
1986 1987 1988 1989 1990 1991 1992
	 * Combined with the control dependency above, we have an effective
	 * smp_load_acquire() without the need for full barriers.
	 *
	 * 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.
1993
	 */
1994
	smp_rmb();
L
Linus Torvalds 已提交
1995

1996
	p->sched_contributes_to_load = !!task_contributes_to_load(p);
P
Peter Zijlstra 已提交
1997
	p->state = TASK_WAKING;
1998

1999
	if (p->sched_class->task_waking)
2000
		p->sched_class->task_waking(p);
2001

2002
	cpu = select_task_rq(p, p->wake_cpu, SD_BALANCE_WAKE, wake_flags);
2003 2004
	if (task_cpu(p) != cpu) {
		wake_flags |= WF_MIGRATED;
2005
		set_task_cpu(p, cpu);
2006
	}
L
Linus Torvalds 已提交
2007 2008
#endif /* CONFIG_SMP */

2009 2010
	ttwu_queue(p, cpu);
stat:
2011
	ttwu_stat(p, cpu, wake_flags);
L
Linus Torvalds 已提交
2012
out:
2013
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2014 2015 2016 2017

	return success;
}

T
Tejun Heo 已提交
2018 2019 2020 2021
/**
 * try_to_wake_up_local - try to wake up a local task with rq lock held
 * @p: the thread to be awakened
 *
2022
 * Put @p on the run-queue if it's not already there. The caller must
T
Tejun Heo 已提交
2023
 * ensure that this_rq() is locked, @p is bound to this_rq() and not
2024
 * the current task.
T
Tejun Heo 已提交
2025 2026 2027 2028 2029
 */
static void try_to_wake_up_local(struct task_struct *p)
{
	struct rq *rq = task_rq(p);

2030 2031 2032 2033
	if (WARN_ON_ONCE(rq != this_rq()) ||
	    WARN_ON_ONCE(p == current))
		return;

T
Tejun Heo 已提交
2034 2035
	lockdep_assert_held(&rq->lock);

2036
	if (!raw_spin_trylock(&p->pi_lock)) {
2037 2038 2039 2040 2041 2042 2043
		/*
		 * 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.
		 */
		lockdep_unpin_lock(&rq->lock);
2044 2045 2046
		raw_spin_unlock(&rq->lock);
		raw_spin_lock(&p->pi_lock);
		raw_spin_lock(&rq->lock);
2047
		lockdep_pin_lock(&rq->lock);
2048 2049
	}

T
Tejun Heo 已提交
2050
	if (!(p->state & TASK_NORMAL))
2051
		goto out;
T
Tejun Heo 已提交
2052

2053 2054
	trace_sched_waking(p);

2055
	if (!task_on_rq_queued(p))
P
Peter Zijlstra 已提交
2056 2057
		ttwu_activate(rq, p, ENQUEUE_WAKEUP);

2058
	ttwu_do_wakeup(rq, p, 0);
2059
	ttwu_stat(p, smp_processor_id(), 0);
2060 2061
out:
	raw_spin_unlock(&p->pi_lock);
T
Tejun Heo 已提交
2062 2063
}

2064 2065 2066 2067 2068
/**
 * 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
2069 2070 2071
 * processes.
 *
 * Return: 1 if the process was woken up, 0 if it was already running.
2072 2073 2074 2075
 *
 * 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.
 */
2076
int wake_up_process(struct task_struct *p)
L
Linus Torvalds 已提交
2077
{
2078
	return try_to_wake_up(p, TASK_NORMAL, 0);
L
Linus Torvalds 已提交
2079 2080 2081
}
EXPORT_SYMBOL(wake_up_process);

2082
int wake_up_state(struct task_struct *p, unsigned int state)
L
Linus Torvalds 已提交
2083 2084 2085 2086
{
	return try_to_wake_up(p, state, 0);
}

2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098
/*
 * 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;
2099 2100 2101 2102

	dl_se->dl_throttled = 0;
	dl_se->dl_new = 1;
	dl_se->dl_yielded = 0;
2103 2104
}

L
Linus Torvalds 已提交
2105 2106 2107
/*
 * Perform scheduler related setup for a newly forked process p.
 * p is forked by current.
I
Ingo Molnar 已提交
2108 2109 2110
 *
 * __sched_fork() is basic setup used by init_idle() too:
 */
2111
static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2112
{
P
Peter Zijlstra 已提交
2113 2114 2115
	p->on_rq			= 0;

	p->se.on_rq			= 0;
I
Ingo Molnar 已提交
2116 2117
	p->se.exec_start		= 0;
	p->se.sum_exec_runtime		= 0;
2118
	p->se.prev_sum_exec_runtime	= 0;
2119
	p->se.nr_migrations		= 0;
P
Peter Zijlstra 已提交
2120
	p->se.vruntime			= 0;
P
Peter Zijlstra 已提交
2121
	INIT_LIST_HEAD(&p->se.group_node);
I
Ingo Molnar 已提交
2122

2123 2124 2125 2126
#ifdef CONFIG_FAIR_GROUP_SCHED
	p->se.cfs_rq			= NULL;
#endif

I
Ingo Molnar 已提交
2127
#ifdef CONFIG_SCHEDSTATS
2128
	memset(&p->se.statistics, 0, sizeof(p->se.statistics));
I
Ingo Molnar 已提交
2129
#endif
N
Nick Piggin 已提交
2130

2131
	RB_CLEAR_NODE(&p->dl.rb_node);
2132
	init_dl_task_timer(&p->dl);
2133
	__dl_clear_params(p);
2134

P
Peter Zijlstra 已提交
2135
	INIT_LIST_HEAD(&p->rt.run_list);
N
Nick Piggin 已提交
2136

2137 2138 2139
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&p->preempt_notifiers);
#endif
2140 2141 2142

#ifdef CONFIG_NUMA_BALANCING
	if (p->mm && atomic_read(&p->mm->mm_users) == 1) {
2143
		p->mm->numa_next_scan = jiffies + msecs_to_jiffies(sysctl_numa_balancing_scan_delay);
2144 2145 2146
		p->mm->numa_scan_seq = 0;
	}

2147 2148 2149 2150 2151
	if (clone_flags & CLONE_VM)
		p->numa_preferred_nid = current->numa_preferred_nid;
	else
		p->numa_preferred_nid = -1;

2152 2153
	p->node_stamp = 0ULL;
	p->numa_scan_seq = p->mm ? p->mm->numa_scan_seq : 0;
2154
	p->numa_scan_period = sysctl_numa_balancing_scan_delay;
2155
	p->numa_work.next = &p->numa_work;
2156
	p->numa_faults = NULL;
2157 2158
	p->last_task_numa_placement = 0;
	p->last_sum_exec_runtime = 0;
2159 2160

	p->numa_group = NULL;
2161
#endif /* CONFIG_NUMA_BALANCING */
I
Ingo Molnar 已提交
2162 2163
}

2164 2165
DEFINE_STATIC_KEY_FALSE(sched_numa_balancing);

2166
#ifdef CONFIG_NUMA_BALANCING
2167

2168 2169 2170
void set_numabalancing_state(bool enabled)
{
	if (enabled)
2171
		static_branch_enable(&sched_numa_balancing);
2172
	else
2173
		static_branch_disable(&sched_numa_balancing);
2174
}
2175 2176 2177 2178 2179 2180 2181

#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;
2182
	int state = static_branch_likely(&sched_numa_balancing);
2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197

	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
2198 2199 2200 2201

/*
 * fork()/clone()-time setup:
 */
2202
int sched_fork(unsigned long clone_flags, struct task_struct *p)
I
Ingo Molnar 已提交
2203
{
2204
	unsigned long flags;
I
Ingo Molnar 已提交
2205 2206
	int cpu = get_cpu();

2207
	__sched_fork(clone_flags, p);
2208
	/*
2209
	 * We mark the process as running here. This guarantees that
2210 2211 2212
	 * nobody will actually run it, and a signal or other external
	 * event cannot wake it up and insert it on the runqueue either.
	 */
2213
	p->state = TASK_RUNNING;
I
Ingo Molnar 已提交
2214

2215 2216 2217 2218 2219
	/*
	 * Make sure we do not leak PI boosting priority to the child.
	 */
	p->prio = current->normal_prio;

2220 2221 2222 2223
	/*
	 * Revert to default priority/policy on fork if requested.
	 */
	if (unlikely(p->sched_reset_on_fork)) {
2224
		if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
2225
			p->policy = SCHED_NORMAL;
2226
			p->static_prio = NICE_TO_PRIO(0);
2227 2228 2229 2230 2231 2232
			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);
2233

2234 2235 2236 2237 2238 2239
		/*
		 * We don't need the reset flag anymore after the fork. It has
		 * fulfilled its duty:
		 */
		p->sched_reset_on_fork = 0;
	}
2240

2241 2242 2243 2244 2245 2246
	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
2247
		p->sched_class = &fair_sched_class;
2248
	}
2249

P
Peter Zijlstra 已提交
2250 2251 2252
	if (p->sched_class->task_fork)
		p->sched_class->task_fork(p);

2253 2254 2255 2256 2257 2258 2259
	/*
	 * 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.
	 */
2260
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2261
	set_task_cpu(p, cpu);
2262
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
2263

2264
#ifdef CONFIG_SCHED_INFO
I
Ingo Molnar 已提交
2265
	if (likely(sched_info_on()))
2266
		memset(&p->sched_info, 0, sizeof(p->sched_info));
L
Linus Torvalds 已提交
2267
#endif
P
Peter Zijlstra 已提交
2268 2269
#if defined(CONFIG_SMP)
	p->on_cpu = 0;
2270
#endif
2271
	init_task_preempt_count(p);
2272
#ifdef CONFIG_SMP
2273
	plist_node_init(&p->pushable_tasks, MAX_PRIO);
2274
	RB_CLEAR_NODE(&p->pushable_dl_tasks);
2275
#endif
2276

N
Nick Piggin 已提交
2277
	put_cpu();
2278
	return 0;
L
Linus Torvalds 已提交
2279 2280
}

2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299
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)
{
2300 2301
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2302 2303 2304
	return &cpu_rq(i)->rd->dl_bw;
}

2305
static inline int dl_bw_cpus(int i)
2306
{
2307 2308 2309
	struct root_domain *rd = cpu_rq(i)->rd;
	int cpus = 0;

2310 2311
	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
			 "sched RCU must be held");
2312 2313 2314 2315
	for_each_cpu_and(i, rd->span, cpu_active_mask)
		cpus++;

	return cpus;
2316 2317 2318 2319 2320 2321 2322
}
#else
inline struct dl_bw *dl_bw_of(int i)
{
	return &cpu_rq(i)->dl.dl_bw;
}

2323
static inline int dl_bw_cpus(int i)
2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335
{
	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.
2336 2337 2338
 *
 * XXX we should delay bw change until the task's 0-lag point, see
 * __setparam_dl().
2339 2340 2341 2342 2343 2344
 */
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));
2345
	u64 period = attr->sched_period ?: attr->sched_deadline;
2346 2347
	u64 runtime = attr->sched_runtime;
	u64 new_bw = dl_policy(policy) ? to_ratio(period, runtime) : 0;
2348
	int cpus, err = -1;
2349 2350 2351 2352 2353 2354 2355 2356 2357 2358

	if (new_bw == p->dl.dl_bw)
		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);
2359
	cpus = dl_bw_cpus(task_cpu(p));
2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379
	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
2380 2381 2382 2383 2384 2385 2386
/*
 * 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.
 */
2387
void wake_up_new_task(struct task_struct *p)
L
Linus Torvalds 已提交
2388 2389
{
	unsigned long flags;
I
Ingo Molnar 已提交
2390
	struct rq *rq;
2391

2392
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2393 2394
	/* Initialize new task's runnable average */
	init_entity_runnable_average(&p->se);
2395 2396 2397 2398 2399 2400
#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
	 */
2401
	set_task_cpu(p, select_task_rq(p, task_cpu(p), SD_BALANCE_FORK, 0));
2402 2403
#endif

2404
	rq = __task_rq_lock(p);
P
Peter Zijlstra 已提交
2405
	activate_task(rq, p, 0);
2406
	p->on_rq = TASK_ON_RQ_QUEUED;
2407
	trace_sched_wakeup_new(p);
P
Peter Zijlstra 已提交
2408
	check_preempt_curr(rq, p, WF_FORK);
2409
#ifdef CONFIG_SMP
2410 2411 2412 2413 2414 2415
	if (p->sched_class->task_woken) {
		/*
		 * Nothing relies on rq->lock after this, so its fine to
		 * drop it.
		 */
		lockdep_unpin_lock(&rq->lock);
2416
		p->sched_class->task_woken(rq, p);
2417 2418
		lockdep_pin_lock(&rq->lock);
	}
2419
#endif
2420
	task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
2421 2422
}

2423 2424
#ifdef CONFIG_PREEMPT_NOTIFIERS

2425 2426
static struct static_key preempt_notifier_key = STATIC_KEY_INIT_FALSE;

2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438
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);

2439
/**
2440
 * preempt_notifier_register - tell me when current is being preempted & rescheduled
R
Randy Dunlap 已提交
2441
 * @notifier: notifier struct to register
2442 2443 2444
 */
void preempt_notifier_register(struct preempt_notifier *notifier)
{
2445 2446 2447
	if (!static_key_false(&preempt_notifier_key))
		WARN(1, "registering preempt_notifier while notifiers disabled\n");

2448 2449 2450 2451 2452 2453
	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
2454
 * @notifier: notifier struct to unregister
2455
 *
2456
 * This is *not* safe to call from within a preemption notifier.
2457 2458 2459 2460 2461 2462 2463
 */
void preempt_notifier_unregister(struct preempt_notifier *notifier)
{
	hlist_del(&notifier->link);
}
EXPORT_SYMBOL_GPL(preempt_notifier_unregister);

2464
static void __fire_sched_in_preempt_notifiers(struct task_struct *curr)
2465 2466 2467
{
	struct preempt_notifier *notifier;

2468
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2469 2470 2471
		notifier->ops->sched_in(notifier, raw_smp_processor_id());
}

2472 2473 2474 2475 2476 2477
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);
}

2478
static void
2479 2480
__fire_sched_out_preempt_notifiers(struct task_struct *curr,
				   struct task_struct *next)
2481 2482 2483
{
	struct preempt_notifier *notifier;

2484
	hlist_for_each_entry(notifier, &curr->preempt_notifiers, link)
2485 2486 2487
		notifier->ops->sched_out(notifier, next);
}

2488 2489 2490 2491 2492 2493 2494 2495
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);
}

2496
#else /* !CONFIG_PREEMPT_NOTIFIERS */
2497

2498
static inline void fire_sched_in_preempt_notifiers(struct task_struct *curr)
2499 2500 2501
{
}

2502
static inline void
2503 2504 2505 2506 2507
fire_sched_out_preempt_notifiers(struct task_struct *curr,
				 struct task_struct *next)
{
}

2508
#endif /* CONFIG_PREEMPT_NOTIFIERS */
2509

2510 2511 2512
/**
 * prepare_task_switch - prepare to switch tasks
 * @rq: the runqueue preparing to switch
R
Randy Dunlap 已提交
2513
 * @prev: the current task that is being switched out
2514 2515 2516 2517 2518 2519 2520 2521 2522
 * @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.
 */
2523 2524 2525
static inline void
prepare_task_switch(struct rq *rq, struct task_struct *prev,
		    struct task_struct *next)
2526
{
2527
	sched_info_switch(rq, prev, next);
2528
	perf_event_task_sched_out(prev, next);
2529
	fire_sched_out_preempt_notifiers(prev, next);
2530 2531 2532 2533
	prepare_lock_switch(rq, next);
	prepare_arch_switch(next);
}

L
Linus Torvalds 已提交
2534 2535 2536 2537
/**
 * finish_task_switch - clean up after a task-switch
 * @prev: the thread we just switched away from.
 *
2538 2539 2540 2541
 * 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
2542 2543
 *
 * Note that we may have delayed dropping an mm in context_switch(). If
I
Ingo Molnar 已提交
2544
 * so, we finish that here outside of the runqueue lock. (Doing it
L
Linus Torvalds 已提交
2545 2546
 * with the lock held can cause deadlocks; see schedule() for
 * details.)
2547 2548 2549 2550 2551
 *
 * 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
2552
 */
2553
static struct rq *finish_task_switch(struct task_struct *prev)
L
Linus Torvalds 已提交
2554 2555
	__releases(rq->lock)
{
2556
	struct rq *rq = this_rq();
L
Linus Torvalds 已提交
2557
	struct mm_struct *mm = rq->prev_mm;
O
Oleg Nesterov 已提交
2558
	long prev_state;
L
Linus Torvalds 已提交
2559

2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570
	/*
	 * 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.
	 */
2571 2572 2573 2574
	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);
2575

L
Linus Torvalds 已提交
2576 2577 2578 2579
	rq->prev_mm = NULL;

	/*
	 * A task struct has one reference for the use as "current".
2580
	 * If a task dies, then it sets TASK_DEAD in tsk->state and calls
O
Oleg Nesterov 已提交
2581 2582
	 * schedule one last time. The schedule call will never return, and
	 * the scheduled task must drop that reference.
2583 2584 2585 2586 2587
	 *
	 * 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
2588
	 */
O
Oleg Nesterov 已提交
2589
	prev_state = prev->state;
2590
	vtime_task_switch(prev);
2591
	perf_event_task_sched_in(prev, current);
2592
	finish_lock_switch(rq, prev);
2593
	finish_arch_post_lock_switch();
S
Steven Rostedt 已提交
2594

2595
	fire_sched_in_preempt_notifiers(current);
L
Linus Torvalds 已提交
2596 2597
	if (mm)
		mmdrop(mm);
2598
	if (unlikely(prev_state == TASK_DEAD)) {
2599 2600 2601
		if (prev->sched_class->task_dead)
			prev->sched_class->task_dead(prev);

2602 2603 2604
		/*
		 * Remove function-return probe instances associated with this
		 * task and put them back on the free list.
I
Ingo Molnar 已提交
2605
		 */
2606
		kprobe_flush_task(prev);
L
Linus Torvalds 已提交
2607
		put_task_struct(prev);
2608
	}
2609

2610
	tick_nohz_task_switch();
2611
	return rq;
L
Linus Torvalds 已提交
2612 2613
}

2614 2615 2616
#ifdef CONFIG_SMP

/* rq->lock is NOT held, but preemption is disabled */
2617
static void __balance_callback(struct rq *rq)
2618
{
2619 2620 2621
	struct callback_head *head, *next;
	void (*func)(struct rq *rq);
	unsigned long flags;
2622

2623 2624 2625 2626 2627 2628 2629 2630
	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;
2631

2632
		func(rq);
2633
	}
2634 2635 2636 2637 2638 2639 2640
	raw_spin_unlock_irqrestore(&rq->lock, flags);
}

static inline void balance_callback(struct rq *rq)
{
	if (unlikely(rq->balance_callback))
		__balance_callback(rq);
2641 2642 2643
}

#else
2644

2645
static inline void balance_callback(struct rq *rq)
2646
{
L
Linus Torvalds 已提交
2647 2648
}

2649 2650
#endif

L
Linus Torvalds 已提交
2651 2652 2653 2654
/**
 * schedule_tail - first thing a freshly forked thread must call.
 * @prev: the thread we just switched away from.
 */
2655
asmlinkage __visible void schedule_tail(struct task_struct *prev)
L
Linus Torvalds 已提交
2656 2657
	__releases(rq->lock)
{
2658
	struct rq *rq;
2659

2660 2661 2662 2663 2664 2665 2666 2667 2668
	/*
	 * 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).
	 */

2669
	rq = finish_task_switch(prev);
2670
	balance_callback(rq);
2671
	preempt_enable();
2672

L
Linus Torvalds 已提交
2673
	if (current->set_child_tid)
2674
		put_user(task_pid_vnr(current), current->set_child_tid);
L
Linus Torvalds 已提交
2675 2676 2677
}

/*
2678
 * context_switch - switch to the new MM and the new thread's register state.
L
Linus Torvalds 已提交
2679
 */
2680
static inline struct rq *
2681
context_switch(struct rq *rq, struct task_struct *prev,
2682
	       struct task_struct *next)
L
Linus Torvalds 已提交
2683
{
I
Ingo Molnar 已提交
2684
	struct mm_struct *mm, *oldmm;
L
Linus Torvalds 已提交
2685

2686
	prepare_task_switch(rq, prev, next);
2687

I
Ingo Molnar 已提交
2688 2689
	mm = next->mm;
	oldmm = prev->active_mm;
2690 2691 2692 2693 2694
	/*
	 * For paravirt, this is coupled with an exit in switch_to to
	 * combine the page table reload and the switch backend into
	 * one hypercall.
	 */
2695
	arch_start_context_switch(prev);
2696

2697
	if (!mm) {
L
Linus Torvalds 已提交
2698 2699 2700 2701 2702 2703
		next->active_mm = oldmm;
		atomic_inc(&oldmm->mm_count);
		enter_lazy_tlb(oldmm, next);
	} else
		switch_mm(oldmm, mm, next);

2704
	if (!prev->mm) {
L
Linus Torvalds 已提交
2705 2706 2707
		prev->active_mm = NULL;
		rq->prev_mm = oldmm;
	}
2708 2709 2710 2711 2712 2713
	/*
	 * 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:
	 */
2714
	lockdep_unpin_lock(&rq->lock);
2715
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
L
Linus Torvalds 已提交
2716 2717 2718

	/* Here we just switch the register state and the stack. */
	switch_to(prev, next, prev);
I
Ingo Molnar 已提交
2719
	barrier();
2720 2721

	return finish_task_switch(prev);
L
Linus Torvalds 已提交
2722 2723 2724
}

/*
2725
 * nr_running and nr_context_switches:
L
Linus Torvalds 已提交
2726 2727
 *
 * externally visible scheduler statistics: current number of runnable
2728
 * threads, total number of context switches performed since bootup.
L
Linus Torvalds 已提交
2729 2730 2731 2732 2733 2734 2735 2736 2737
 */
unsigned long nr_running(void)
{
	unsigned long i, sum = 0;

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

	return sum;
2738
}
L
Linus Torvalds 已提交
2739

2740 2741
/*
 * Check if only the current task is running on the cpu.
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751
 *
 * 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)
2752 2753 2754
 */
bool single_task_running(void)
{
2755
	return raw_rq()->nr_running == 1;
2756 2757 2758
}
EXPORT_SYMBOL(single_task_running);

L
Linus Torvalds 已提交
2759
unsigned long long nr_context_switches(void)
2760
{
2761 2762
	int i;
	unsigned long long sum = 0;
2763

2764
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2765
		sum += cpu_rq(i)->nr_switches;
2766

L
Linus Torvalds 已提交
2767 2768
	return sum;
}
2769

L
Linus Torvalds 已提交
2770 2771 2772
unsigned long nr_iowait(void)
{
	unsigned long i, sum = 0;
2773

2774
	for_each_possible_cpu(i)
L
Linus Torvalds 已提交
2775
		sum += atomic_read(&cpu_rq(i)->nr_iowait);
2776

L
Linus Torvalds 已提交
2777 2778
	return sum;
}
2779

2780
unsigned long nr_iowait_cpu(int cpu)
2781
{
2782
	struct rq *this = cpu_rq(cpu);
2783 2784
	return atomic_read(&this->nr_iowait);
}
2785

2786 2787
void get_iowait_load(unsigned long *nr_waiters, unsigned long *load)
{
2788 2789 2790
	struct rq *rq = this_rq();
	*nr_waiters = atomic_read(&rq->nr_iowait);
	*load = rq->load.weight;
2791 2792
}

I
Ingo Molnar 已提交
2793
#ifdef CONFIG_SMP
2794

2795
/*
P
Peter Zijlstra 已提交
2796 2797
 * sched_exec - execve() is a valuable balancing opportunity, because at
 * this point the task has the smallest effective memory and cache footprint.
2798
 */
P
Peter Zijlstra 已提交
2799
void sched_exec(void)
2800
{
P
Peter Zijlstra 已提交
2801
	struct task_struct *p = current;
L
Linus Torvalds 已提交
2802
	unsigned long flags;
2803
	int dest_cpu;
2804

2805
	raw_spin_lock_irqsave(&p->pi_lock, flags);
2806
	dest_cpu = p->sched_class->select_task_rq(p, task_cpu(p), SD_BALANCE_EXEC, 0);
2807 2808
	if (dest_cpu == smp_processor_id())
		goto unlock;
P
Peter Zijlstra 已提交
2809

2810
	if (likely(cpu_active(dest_cpu))) {
2811
		struct migration_arg arg = { p, dest_cpu };
2812

2813 2814
		raw_spin_unlock_irqrestore(&p->pi_lock, flags);
		stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
L
Linus Torvalds 已提交
2815 2816
		return;
	}
2817
unlock:
2818
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
2819
}
I
Ingo Molnar 已提交
2820

L
Linus Torvalds 已提交
2821 2822 2823
#endif

DEFINE_PER_CPU(struct kernel_stat, kstat);
2824
DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
L
Linus Torvalds 已提交
2825 2826

EXPORT_PER_CPU_SYMBOL(kstat);
2827
EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
L
Linus Torvalds 已提交
2828

2829 2830 2831 2832 2833 2834 2835 2836 2837
/*
 * 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)
{
	unsigned long flags;
	struct rq *rq;
2838
	u64 ns;
2839

2840 2841 2842 2843 2844 2845 2846 2847 2848
#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.
2849 2850
	 * If we see ->on_cpu without ->on_rq, the task is leaving, and has
	 * been accounted, so we're correct here as well.
2851
	 */
2852
	if (!p->on_cpu || !task_on_rq_queued(p))
2853 2854 2855
		return p->se.sum_exec_runtime;
#endif

2856
	rq = task_rq_lock(p, &flags);
2857 2858 2859 2860 2861 2862 2863 2864 2865 2866
	/*
	 * 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;
2867
	task_rq_unlock(rq, p, &flags);
2868 2869 2870

	return ns;
}
2871

2872 2873 2874 2875 2876 2877 2878 2879
/*
 * 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
2880
	struct task_struct *curr = rq->curr;
2881 2882

	sched_clock_tick();
I
Ingo Molnar 已提交
2883

2884
	raw_spin_lock(&rq->lock);
2885
	update_rq_clock(rq);
P
Peter Zijlstra 已提交
2886
	curr->sched_class->task_tick(rq, curr, 0);
2887
	update_cpu_load_active(rq);
2888
	calc_global_load_tick(rq);
2889
	raw_spin_unlock(&rq->lock);
2890

2891
	perf_event_task_tick();
2892

2893
#ifdef CONFIG_SMP
2894
	rq->idle_balance = idle_cpu(cpu);
2895
	trigger_load_balance(rq);
2896
#endif
2897
	rq_last_tick_reset(rq);
L
Linus Torvalds 已提交
2898 2899
}

2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910
#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.
2911 2912
 *
 * Return: Maximum deferment in nanoseconds.
2913 2914 2915 2916
 */
u64 scheduler_tick_max_deferment(void)
{
	struct rq *rq = this_rq();
2917
	unsigned long next, now = READ_ONCE(jiffies);
2918 2919 2920 2921 2922 2923

	next = rq->last_sched_tick + HZ;

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

2924
	return jiffies_to_nsecs(next - now);
L
Linus Torvalds 已提交
2925
}
2926
#endif
L
Linus Torvalds 已提交
2927

2928
notrace unsigned long get_parent_ip(unsigned long addr)
2929 2930 2931 2932 2933 2934 2935 2936
{
	if (in_lock_functions(addr)) {
		addr = CALLER_ADDR2;
		if (in_lock_functions(addr))
			addr = CALLER_ADDR3;
	}
	return addr;
}
L
Linus Torvalds 已提交
2937

2938 2939 2940
#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
				defined(CONFIG_PREEMPT_TRACER))

2941
void preempt_count_add(int val)
L
Linus Torvalds 已提交
2942
{
2943
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2944 2945 2946
	/*
	 * Underflow?
	 */
2947 2948
	if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0)))
		return;
2949
#endif
2950
	__preempt_count_add(val);
2951
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2952 2953 2954
	/*
	 * Spinlock count overflowing soon?
	 */
2955 2956
	DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >=
				PREEMPT_MASK - 10);
2957
#endif
2958 2959 2960 2961 2962 2963 2964
	if (preempt_count() == val) {
		unsigned long ip = get_parent_ip(CALLER_ADDR1);
#ifdef CONFIG_DEBUG_PREEMPT
		current->preempt_disable_ip = ip;
#endif
		trace_preempt_off(CALLER_ADDR0, ip);
	}
L
Linus Torvalds 已提交
2965
}
2966
EXPORT_SYMBOL(preempt_count_add);
2967
NOKPROBE_SYMBOL(preempt_count_add);
L
Linus Torvalds 已提交
2968

2969
void preempt_count_sub(int val)
L
Linus Torvalds 已提交
2970
{
2971
#ifdef CONFIG_DEBUG_PREEMPT
L
Linus Torvalds 已提交
2972 2973 2974
	/*
	 * Underflow?
	 */
2975
	if (DEBUG_LOCKS_WARN_ON(val > preempt_count()))
2976
		return;
L
Linus Torvalds 已提交
2977 2978 2979
	/*
	 * Is the spinlock portion underflowing?
	 */
2980 2981 2982
	if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) &&
			!(preempt_count() & PREEMPT_MASK)))
		return;
2983
#endif
2984

2985 2986
	if (preempt_count() == val)
		trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
2987
	__preempt_count_sub(val);
L
Linus Torvalds 已提交
2988
}
2989
EXPORT_SYMBOL(preempt_count_sub);
2990
NOKPROBE_SYMBOL(preempt_count_sub);
L
Linus Torvalds 已提交
2991 2992 2993 2994

#endif

/*
I
Ingo Molnar 已提交
2995
 * Print scheduling while atomic bug:
L
Linus Torvalds 已提交
2996
 */
I
Ingo Molnar 已提交
2997
static noinline void __schedule_bug(struct task_struct *prev)
L
Linus Torvalds 已提交
2998
{
2999 3000 3001
	if (oops_in_progress)
		return;

P
Peter Zijlstra 已提交
3002 3003
	printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n",
		prev->comm, prev->pid, preempt_count());
3004

I
Ingo Molnar 已提交
3005
	debug_show_held_locks(prev);
3006
	print_modules();
I
Ingo Molnar 已提交
3007 3008
	if (irqs_disabled())
		print_irqtrace_events(prev);
3009 3010 3011 3012 3013 3014 3015
#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
3016
	dump_stack();
3017
	add_taint(TAINT_WARN, LOCKDEP_STILL_OK);
I
Ingo Molnar 已提交
3018
}
L
Linus Torvalds 已提交
3019

I
Ingo Molnar 已提交
3020 3021 3022 3023 3024
/*
 * Various schedule()-time debugging checks and statistics:
 */
static inline void schedule_debug(struct task_struct *prev)
{
3025
#ifdef CONFIG_SCHED_STACK_END_CHECK
3026
	BUG_ON(task_stack_end_corrupted(prev));
3027
#endif
3028

3029
	if (unlikely(in_atomic_preempt_off())) {
I
Ingo Molnar 已提交
3030
		__schedule_bug(prev);
3031 3032
		preempt_count_set(PREEMPT_DISABLED);
	}
3033
	rcu_sleep_check();
I
Ingo Molnar 已提交
3034

L
Linus Torvalds 已提交
3035 3036
	profile_hit(SCHED_PROFILING, __builtin_return_address(0));

3037
	schedstat_inc(this_rq(), sched_count);
I
Ingo Molnar 已提交
3038 3039 3040 3041 3042 3043
}

/*
 * Pick up the highest-prio task:
 */
static inline struct task_struct *
3044
pick_next_task(struct rq *rq, struct task_struct *prev)
I
Ingo Molnar 已提交
3045
{
3046
	const struct sched_class *class = &fair_sched_class;
I
Ingo Molnar 已提交
3047
	struct task_struct *p;
L
Linus Torvalds 已提交
3048 3049

	/*
I
Ingo Molnar 已提交
3050 3051
	 * Optimization: we know that if all tasks are in
	 * the fair class we can call that function directly:
L
Linus Torvalds 已提交
3052
	 */
3053
	if (likely(prev->sched_class == class &&
3054
		   rq->nr_running == rq->cfs.h_nr_running)) {
3055
		p = fair_sched_class.pick_next_task(rq, prev);
3056 3057 3058 3059 3060 3061 3062 3063
		if (unlikely(p == RETRY_TASK))
			goto again;

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

		return p;
L
Linus Torvalds 已提交
3064 3065
	}

3066
again:
3067
	for_each_class(class) {
3068
		p = class->pick_next_task(rq, prev);
3069 3070 3071
		if (p) {
			if (unlikely(p == RETRY_TASK))
				goto again;
I
Ingo Molnar 已提交
3072
			return p;
3073
		}
I
Ingo Molnar 已提交
3074
	}
3075 3076

	BUG(); /* the idle class will always have a runnable task */
I
Ingo Molnar 已提交
3077
}
L
Linus Torvalds 已提交
3078

I
Ingo Molnar 已提交
3079
/*
3080
 * __schedule() is the main scheduler function.
3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114
 *
 * 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
3115
 *
3116
 * WARNING: must be called with preemption disabled!
I
Ingo Molnar 已提交
3117
 */
3118
static void __sched notrace __schedule(bool preempt)
I
Ingo Molnar 已提交
3119 3120
{
	struct task_struct *prev, *next;
3121
	unsigned long *switch_count;
I
Ingo Molnar 已提交
3122
	struct rq *rq;
3123
	int cpu;
I
Ingo Molnar 已提交
3124 3125 3126

	cpu = smp_processor_id();
	rq = cpu_rq(cpu);
3127
	rcu_note_context_switch();
I
Ingo Molnar 已提交
3128 3129
	prev = rq->curr;

3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140
	/*
	 * 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
3141
	schedule_debug(prev);
L
Linus Torvalds 已提交
3142

3143
	if (sched_feat(HRTICK))
M
Mike Galbraith 已提交
3144
		hrtick_clear(rq);
P
Peter Zijlstra 已提交
3145

3146 3147 3148 3149 3150 3151
	/*
	 * 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();
3152
	raw_spin_lock_irq(&rq->lock);
3153
	lockdep_pin_lock(&rq->lock);
L
Linus Torvalds 已提交
3154

3155 3156
	rq->clock_skip_update <<= 1; /* promote REQ to ACT */

3157
	switch_count = &prev->nivcsw;
3158
	if (!preempt && prev->state) {
T
Tejun Heo 已提交
3159
		if (unlikely(signal_pending_state(prev->state, prev))) {
L
Linus Torvalds 已提交
3160
			prev->state = TASK_RUNNING;
T
Tejun Heo 已提交
3161
		} else {
3162 3163 3164
			deactivate_task(rq, prev, DEQUEUE_SLEEP);
			prev->on_rq = 0;

T
Tejun Heo 已提交
3165
			/*
3166 3167 3168
			 * 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
3169 3170 3171 3172 3173 3174 3175 3176 3177
			 */
			if (prev->flags & PF_WQ_WORKER) {
				struct task_struct *to_wakeup;

				to_wakeup = wq_worker_sleeping(prev, cpu);
				if (to_wakeup)
					try_to_wake_up_local(to_wakeup);
			}
		}
I
Ingo Molnar 已提交
3178
		switch_count = &prev->nvcsw;
L
Linus Torvalds 已提交
3179 3180
	}

3181
	if (task_on_rq_queued(prev))
3182 3183 3184
		update_rq_clock(rq);

	next = pick_next_task(rq, prev);
3185
	clear_tsk_need_resched(prev);
3186
	clear_preempt_need_resched();
3187
	rq->clock_skip_update = 0;
L
Linus Torvalds 已提交
3188 3189 3190 3191 3192 3193

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

3194
		trace_sched_switch(preempt, prev, next);
3195 3196
		rq = context_switch(rq, prev, next); /* unlocks the rq */
		cpu = cpu_of(rq);
3197 3198
	} else {
		lockdep_unpin_lock(&rq->lock);
3199
		raw_spin_unlock_irq(&rq->lock);
3200
	}
L
Linus Torvalds 已提交
3201

3202
	balance_callback(rq);
L
Linus Torvalds 已提交
3203
}
3204

3205 3206
static inline void sched_submit_work(struct task_struct *tsk)
{
3207
	if (!tsk->state || tsk_is_pi_blocked(tsk))
3208 3209 3210 3211 3212 3213 3214 3215 3216
		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);
}

3217
asmlinkage __visible void __sched schedule(void)
3218
{
3219 3220 3221
	struct task_struct *tsk = current;

	sched_submit_work(tsk);
3222
	do {
3223
		preempt_disable();
3224
		__schedule(false);
3225
		sched_preempt_enable_no_resched();
3226
	} while (need_resched());
3227
}
L
Linus Torvalds 已提交
3228 3229
EXPORT_SYMBOL(schedule);

3230
#ifdef CONFIG_CONTEXT_TRACKING
3231
asmlinkage __visible void __sched schedule_user(void)
3232 3233 3234 3235 3236 3237
{
	/*
	 * 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.
3238 3239
	 *
	 * NB: There are buggy callers of this function.  Ideally we
3240
	 * should warn if prev_state != CONTEXT_USER, but that will trigger
3241
	 * too frequently to make sense yet.
3242
	 */
3243
	enum ctx_state prev_state = exception_enter();
3244
	schedule();
3245
	exception_exit(prev_state);
3246 3247 3248
}
#endif

3249 3250 3251 3252 3253 3254 3255
/**
 * schedule_preempt_disabled - called with preemption disabled
 *
 * Returns with preemption disabled. Note: preempt_count must be 1
 */
void __sched schedule_preempt_disabled(void)
{
3256
	sched_preempt_enable_no_resched();
3257 3258 3259 3260
	schedule();
	preempt_disable();
}

3261
static void __sched notrace preempt_schedule_common(void)
3262 3263
{
	do {
3264
		preempt_disable_notrace();
3265
		__schedule(true);
3266
		preempt_enable_no_resched_notrace();
3267 3268 3269 3270 3271 3272 3273 3274

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

L
Linus Torvalds 已提交
3275 3276
#ifdef CONFIG_PREEMPT
/*
3277
 * this is the entry point to schedule() from in-kernel preemption
I
Ingo Molnar 已提交
3278
 * off of preempt_enable. Kernel preemptions off return from interrupt
L
Linus Torvalds 已提交
3279 3280
 * occur there and call schedule directly.
 */
3281
asmlinkage __visible void __sched notrace preempt_schedule(void)
L
Linus Torvalds 已提交
3282 3283 3284
{
	/*
	 * If there is a non-zero preempt_count or interrupts are disabled,
I
Ingo Molnar 已提交
3285
	 * we do not want to preempt the current task. Just return..
L
Linus Torvalds 已提交
3286
	 */
3287
	if (likely(!preemptible()))
L
Linus Torvalds 已提交
3288 3289
		return;

3290
	preempt_schedule_common();
L
Linus Torvalds 已提交
3291
}
3292
NOKPROBE_SYMBOL(preempt_schedule);
L
Linus Torvalds 已提交
3293
EXPORT_SYMBOL(preempt_schedule);
3294 3295

/**
3296
 * preempt_schedule_notrace - preempt_schedule called by tracing
3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308
 *
 * 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.
 */
3309
asmlinkage __visible void __sched notrace preempt_schedule_notrace(void)
3310 3311 3312 3313 3314 3315 3316
{
	enum ctx_state prev_ctx;

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

	do {
3317
		preempt_disable_notrace();
3318 3319 3320 3321 3322 3323
		/*
		 * Needs preempt disabled in case user_exit() is traced
		 * and the tracer calls preempt_enable_notrace() causing
		 * an infinite recursion.
		 */
		prev_ctx = exception_enter();
3324
		__schedule(true);
3325 3326
		exception_exit(prev_ctx);

3327
		preempt_enable_no_resched_notrace();
3328 3329
	} while (need_resched());
}
3330
EXPORT_SYMBOL_GPL(preempt_schedule_notrace);
3331

3332
#endif /* CONFIG_PREEMPT */
L
Linus Torvalds 已提交
3333 3334

/*
3335
 * this is the entry point to schedule() from kernel preemption
L
Linus Torvalds 已提交
3336 3337 3338 3339
 * off of irq context.
 * Note, that this is called and return with irqs disabled. This will
 * protect us against recursive calling from irq.
 */
3340
asmlinkage __visible void __sched preempt_schedule_irq(void)
L
Linus Torvalds 已提交
3341
{
3342
	enum ctx_state prev_state;
3343

3344
	/* Catch callers which need to be fixed */
3345
	BUG_ON(preempt_count() || !irqs_disabled());
L
Linus Torvalds 已提交
3346

3347 3348
	prev_state = exception_enter();

3349
	do {
3350
		preempt_disable();
3351
		local_irq_enable();
3352
		__schedule(true);
3353
		local_irq_disable();
3354
		sched_preempt_enable_no_resched();
3355
	} while (need_resched());
3356 3357

	exception_exit(prev_state);
L
Linus Torvalds 已提交
3358 3359
}

P
Peter Zijlstra 已提交
3360
int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,
I
Ingo Molnar 已提交
3361
			  void *key)
L
Linus Torvalds 已提交
3362
{
P
Peter Zijlstra 已提交
3363
	return try_to_wake_up(curr->private, mode, wake_flags);
L
Linus Torvalds 已提交
3364 3365 3366
}
EXPORT_SYMBOL(default_wake_function);

3367 3368 3369 3370 3371 3372 3373 3374 3375 3376
#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().
 *
3377 3378
 * Used by the rt_mutex code to implement priority inheritance
 * logic. Call site only calls if the priority of the task changed.
3379
 */
3380
void rt_mutex_setprio(struct task_struct *p, int prio)
3381
{
3382
	int oldprio, queued, running, enqueue_flag = ENQUEUE_RESTORE;
3383
	struct rq *rq;
3384
	const struct sched_class *prev_class;
3385

3386
	BUG_ON(prio > MAX_PRIO);
3387

3388
	rq = __task_rq_lock(p);
3389

3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407
	/*
	 * 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;
	}

3408
	trace_sched_pi_setprio(p, prio);
3409
	oldprio = p->prio;
3410
	prev_class = p->sched_class;
3411
	queued = task_on_rq_queued(p);
3412
	running = task_current(rq, p);
3413
	if (queued)
3414
		dequeue_task(rq, p, DEQUEUE_SAVE);
3415
	if (running)
3416
		put_prev_task(rq, p);
I
Ingo Molnar 已提交
3417

3418 3419 3420 3421 3422 3423 3424 3425 3426 3427
	/*
	 * 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)) {
3428 3429 3430
		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))) {
3431
			p->dl.dl_boosted = 1;
3432
			enqueue_flag |= ENQUEUE_REPLENISH;
3433 3434
		} else
			p->dl.dl_boosted = 0;
3435
		p->sched_class = &dl_sched_class;
3436 3437 3438 3439
	} else if (rt_prio(prio)) {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
		if (oldprio < prio)
3440
			enqueue_flag |= ENQUEUE_HEAD;
I
Ingo Molnar 已提交
3441
		p->sched_class = &rt_sched_class;
3442 3443 3444
	} else {
		if (dl_prio(oldprio))
			p->dl.dl_boosted = 0;
3445 3446
		if (rt_prio(oldprio))
			p->rt.timeout = 0;
I
Ingo Molnar 已提交
3447
		p->sched_class = &fair_sched_class;
3448
	}
I
Ingo Molnar 已提交
3449

3450 3451
	p->prio = prio;

3452 3453
	if (running)
		p->sched_class->set_curr_task(rq);
3454
	if (queued)
3455
		enqueue_task(rq, p, enqueue_flag);
3456

P
Peter Zijlstra 已提交
3457
	check_class_changed(rq, p, prev_class, oldprio);
3458
out_unlock:
3459
	preempt_disable(); /* avoid rq from going away on us */
3460
	__task_rq_unlock(rq);
3461 3462 3463

	balance_callback(rq);
	preempt_enable();
3464 3465
}
#endif
3466

3467
void set_user_nice(struct task_struct *p, long nice)
L
Linus Torvalds 已提交
3468
{
3469
	int old_prio, delta, queued;
L
Linus Torvalds 已提交
3470
	unsigned long flags;
3471
	struct rq *rq;
L
Linus Torvalds 已提交
3472

3473
	if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE)
L
Linus Torvalds 已提交
3474 3475 3476 3477 3478 3479 3480 3481 3482 3483
		return;
	/*
	 * We have to be careful, if called from sys_setpriority(),
	 * the task might be in the middle of scheduling on another CPU.
	 */
	rq = task_rq_lock(p, &flags);
	/*
	 * 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
3484
	 * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR:
L
Linus Torvalds 已提交
3485
	 */
3486
	if (task_has_dl_policy(p) || task_has_rt_policy(p)) {
L
Linus Torvalds 已提交
3487 3488 3489
		p->static_prio = NICE_TO_PRIO(nice);
		goto out_unlock;
	}
3490 3491
	queued = task_on_rq_queued(p);
	if (queued)
3492
		dequeue_task(rq, p, DEQUEUE_SAVE);
L
Linus Torvalds 已提交
3493 3494

	p->static_prio = NICE_TO_PRIO(nice);
3495
	set_load_weight(p);
3496 3497 3498
	old_prio = p->prio;
	p->prio = effective_prio(p);
	delta = p->prio - old_prio;
L
Linus Torvalds 已提交
3499

3500
	if (queued) {
3501
		enqueue_task(rq, p, ENQUEUE_RESTORE);
L
Linus Torvalds 已提交
3502
		/*
3503 3504
		 * If the task increased its priority or is running and
		 * lowered its priority, then reschedule its CPU:
L
Linus Torvalds 已提交
3505
		 */
3506
		if (delta < 0 || (delta > 0 && task_running(rq, p)))
3507
			resched_curr(rq);
L
Linus Torvalds 已提交
3508 3509
	}
out_unlock:
3510
	task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
3511 3512 3513
}
EXPORT_SYMBOL(set_user_nice);

M
Matt Mackall 已提交
3514 3515 3516 3517 3518
/*
 * can_nice - check if a task can reduce its nice value
 * @p: task
 * @nice: nice value
 */
3519
int can_nice(const struct task_struct *p, const int nice)
M
Matt Mackall 已提交
3520
{
3521
	/* convert nice value [19,-20] to rlimit style value [1,40] */
3522
	int nice_rlim = nice_to_rlimit(nice);
3523

3524
	return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) ||
M
Matt Mackall 已提交
3525 3526 3527
		capable(CAP_SYS_NICE));
}

L
Linus Torvalds 已提交
3528 3529 3530 3531 3532 3533 3534 3535 3536
#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.
 */
3537
SYSCALL_DEFINE1(nice, int, increment)
L
Linus Torvalds 已提交
3538
{
3539
	long nice, retval;
L
Linus Torvalds 已提交
3540 3541 3542 3543 3544 3545

	/*
	 * 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.
	 */
3546
	increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH);
3547
	nice = task_nice(current) + increment;
L
Linus Torvalds 已提交
3548

3549
	nice = clamp_val(nice, MIN_NICE, MAX_NICE);
M
Matt Mackall 已提交
3550 3551 3552
	if (increment < 0 && !can_nice(current, nice))
		return -EPERM;

L
Linus Torvalds 已提交
3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566
	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.
 *
3567
 * Return: The priority value as seen by users in /proc.
L
Linus Torvalds 已提交
3568 3569 3570
 * RT tasks are offset by -200. Normal tasks are centered
 * around 0, value goes from -16 to +15.
 */
3571
int task_prio(const struct task_struct *p)
L
Linus Torvalds 已提交
3572 3573 3574 3575 3576 3577 3578
{
	return p->prio - MAX_RT_PRIO;
}

/**
 * idle_cpu - is a given cpu idle currently?
 * @cpu: the processor in question.
3579 3580
 *
 * Return: 1 if the CPU is currently idle. 0 otherwise.
L
Linus Torvalds 已提交
3581 3582 3583
 */
int idle_cpu(int cpu)
{
T
Thomas Gleixner 已提交
3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597
	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
3598 3599 3600 3601 3602
}

/**
 * idle_task - return the idle task for a given cpu.
 * @cpu: the processor in question.
3603 3604
 *
 * Return: The idle task for the cpu @cpu.
L
Linus Torvalds 已提交
3605
 */
3606
struct task_struct *idle_task(int cpu)
L
Linus Torvalds 已提交
3607 3608 3609 3610 3611 3612 3613
{
	return cpu_rq(cpu)->idle;
}

/**
 * find_process_by_pid - find a process with a matching PID value.
 * @pid: the pid in question.
3614 3615
 *
 * The task of @pid, if found. %NULL otherwise.
L
Linus Torvalds 已提交
3616
 */
A
Alexey Dobriyan 已提交
3617
static struct task_struct *find_process_by_pid(pid_t pid)
L
Linus Torvalds 已提交
3618
{
3619
	return pid ? find_task_by_vpid(pid) : current;
L
Linus Torvalds 已提交
3620 3621
}

3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636
/*
 * 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;
3637
	dl_se->dl_period = attr->sched_period ?: dl_se->dl_deadline;
3638
	dl_se->flags = attr->sched_flags;
3639
	dl_se->dl_bw = to_ratio(dl_se->dl_period, dl_se->dl_runtime);
3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659

	/*
	 * 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.
	 */
3660 3661
}

3662 3663 3664 3665 3666 3667
/*
 * sched_setparam() passes in -1 for its policy, to let the functions
 * it calls know not to change it.
 */
#define SETPARAM_POLICY	-1

3668 3669
static void __setscheduler_params(struct task_struct *p,
		const struct sched_attr *attr)
L
Linus Torvalds 已提交
3670
{
3671 3672
	int policy = attr->sched_policy;

3673
	if (policy == SETPARAM_POLICY)
3674 3675
		policy = p->policy;

L
Linus Torvalds 已提交
3676
	p->policy = policy;
3677

3678 3679
	if (dl_policy(policy))
		__setparam_dl(p, attr);
3680
	else if (fair_policy(policy))
3681 3682
		p->static_prio = NICE_TO_PRIO(attr->sched_nice);

3683 3684 3685 3686 3687 3688
	/*
	 * __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;
3689
	p->normal_prio = normal_prio(p);
3690 3691
	set_load_weight(p);
}
3692

3693 3694
/* Actually do priority change: must hold pi & rq lock. */
static void __setscheduler(struct rq *rq, struct task_struct *p,
3695
			   const struct sched_attr *attr, bool keep_boost)
3696 3697
{
	__setscheduler_params(p, attr);
3698

3699
	/*
3700 3701
	 * Keep a potential priority boosting if called from
	 * sched_setscheduler().
3702
	 */
3703 3704 3705 3706
	if (keep_boost)
		p->prio = rt_mutex_get_effective_prio(p, normal_prio(p));
	else
		p->prio = normal_prio(p);
3707

3708 3709 3710
	if (dl_prio(p->prio))
		p->sched_class = &dl_sched_class;
	else if (rt_prio(p->prio))
3711 3712 3713
		p->sched_class = &rt_sched_class;
	else
		p->sched_class = &fair_sched_class;
L
Linus Torvalds 已提交
3714
}
3715 3716 3717 3718 3719 3720 3721 3722 3723

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;
3724
	attr->sched_period = dl_se->dl_period;
3725 3726 3727 3728 3729 3730
	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
3731
 * than the runtime, as well as the period of being zero or
3732
 * greater than deadline. Furthermore, we have to be sure that
3733 3734 3735 3736
 * 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).
3737 3738 3739 3740
 */
static bool
__checkparam_dl(const struct sched_attr *attr)
{
3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766
	/* 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;
3767 3768
}

3769 3770 3771 3772 3773 3774 3775 3776 3777 3778
/*
 * 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);
3779 3780
	match = (uid_eq(cred->euid, pcred->euid) ||
		 uid_eq(cred->euid, pcred->uid));
3781 3782 3783 3784
	rcu_read_unlock();
	return match;
}

3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798
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;
}

3799 3800
static int __sched_setscheduler(struct task_struct *p,
				const struct sched_attr *attr,
3801
				bool user, bool pi)
L
Linus Torvalds 已提交
3802
{
3803 3804
	int newprio = dl_policy(attr->sched_policy) ? MAX_DL_PRIO - 1 :
		      MAX_RT_PRIO - 1 - attr->sched_priority;
3805
	int retval, oldprio, oldpolicy = -1, queued, running;
3806
	int new_effective_prio, policy = attr->sched_policy;
L
Linus Torvalds 已提交
3807
	unsigned long flags;
3808
	const struct sched_class *prev_class;
3809
	struct rq *rq;
3810
	int reset_on_fork;
L
Linus Torvalds 已提交
3811

3812 3813
	/* may grab non-irq protected spin_locks */
	BUG_ON(in_interrupt());
L
Linus Torvalds 已提交
3814 3815
recheck:
	/* double check policy once rq lock held */
3816 3817
	if (policy < 0) {
		reset_on_fork = p->sched_reset_on_fork;
L
Linus Torvalds 已提交
3818
		policy = oldpolicy = p->policy;
3819
	} else {
3820
		reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK);
3821

3822
		if (!valid_policy(policy))
3823 3824 3825
			return -EINVAL;
	}

3826 3827 3828
	if (attr->sched_flags & ~(SCHED_FLAG_RESET_ON_FORK))
		return -EINVAL;

L
Linus Torvalds 已提交
3829 3830
	/*
	 * Valid priorities for SCHED_FIFO and SCHED_RR are
I
Ingo Molnar 已提交
3831 3832
	 * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL,
	 * SCHED_BATCH and SCHED_IDLE is 0.
L
Linus Torvalds 已提交
3833
	 */
3834
	if ((p->mm && attr->sched_priority > MAX_USER_RT_PRIO-1) ||
3835
	    (!p->mm && attr->sched_priority > MAX_RT_PRIO-1))
L
Linus Torvalds 已提交
3836
		return -EINVAL;
3837 3838
	if ((dl_policy(policy) && !__checkparam_dl(attr)) ||
	    (rt_policy(policy) != (attr->sched_priority != 0)))
L
Linus Torvalds 已提交
3839 3840
		return -EINVAL;

3841 3842 3843
	/*
	 * Allow unprivileged RT tasks to decrease priority:
	 */
3844
	if (user && !capable(CAP_SYS_NICE)) {
3845
		if (fair_policy(policy)) {
3846
			if (attr->sched_nice < task_nice(p) &&
3847
			    !can_nice(p, attr->sched_nice))
3848 3849 3850
				return -EPERM;
		}

3851
		if (rt_policy(policy)) {
3852 3853
			unsigned long rlim_rtprio =
					task_rlimit(p, RLIMIT_RTPRIO);
3854 3855 3856 3857 3858 3859

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

			/* can't increase priority */
3860 3861
			if (attr->sched_priority > p->rt_priority &&
			    attr->sched_priority > rlim_rtprio)
3862 3863
				return -EPERM;
		}
3864

3865 3866 3867 3868 3869 3870 3871 3872 3873
		 /*
		  * 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
3874
		/*
3875 3876
		 * 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
3877
		 */
3878
		if (idle_policy(p->policy) && !idle_policy(policy)) {
3879
			if (!can_nice(p, task_nice(p)))
3880 3881
				return -EPERM;
		}
3882

3883
		/* can't change other user's priorities */
3884
		if (!check_same_owner(p))
3885
			return -EPERM;
3886 3887 3888 3889

		/* Normal users shall not reset the sched_reset_on_fork flag */
		if (p->sched_reset_on_fork && !reset_on_fork)
			return -EPERM;
3890
	}
L
Linus Torvalds 已提交
3891

3892
	if (user) {
3893
		retval = security_task_setscheduler(p);
3894 3895 3896 3897
		if (retval)
			return retval;
	}

3898 3899 3900
	/*
	 * make sure no PI-waiters arrive (or leave) while we are
	 * changing the priority of the task:
3901
	 *
L
Lucas De Marchi 已提交
3902
	 * To be able to change p->policy safely, the appropriate
L
Linus Torvalds 已提交
3903 3904
	 * runqueue lock must be held.
	 */
3905
	rq = task_rq_lock(p, &flags);
3906

3907 3908 3909 3910
	/*
	 * Changing the policy of the stop threads its a very bad idea
	 */
	if (p == rq->stop) {
3911
		task_rq_unlock(rq, p, &flags);
3912 3913 3914
		return -EINVAL;
	}

3915
	/*
3916 3917
	 * If not changing anything there's no need to proceed further,
	 * but store a possible modification of reset_on_fork.
3918
	 */
3919
	if (unlikely(policy == p->policy)) {
3920
		if (fair_policy(policy) && attr->sched_nice != task_nice(p))
3921 3922 3923
			goto change;
		if (rt_policy(policy) && attr->sched_priority != p->rt_priority)
			goto change;
3924
		if (dl_policy(policy) && dl_param_changed(p, attr))
3925
			goto change;
3926

3927
		p->sched_reset_on_fork = reset_on_fork;
3928
		task_rq_unlock(rq, p, &flags);
3929 3930
		return 0;
	}
3931
change:
3932

3933
	if (user) {
3934
#ifdef CONFIG_RT_GROUP_SCHED
3935 3936 3937 3938 3939
		/*
		 * Do not allow realtime tasks into groups that have no runtime
		 * assigned.
		 */
		if (rt_bandwidth_enabled() && rt_policy(policy) &&
3940 3941
				task_group(p)->rt_bandwidth.rt_runtime == 0 &&
				!task_group_is_autogroup(task_group(p))) {
3942
			task_rq_unlock(rq, p, &flags);
3943 3944 3945
			return -EPERM;
		}
#endif
3946 3947 3948 3949 3950 3951 3952 3953 3954
#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.
			 */
3955 3956
			if (!cpumask_subset(span, &p->cpus_allowed) ||
			    rq->rd->dl_bw.bw == 0) {
3957 3958 3959 3960 3961 3962
				task_rq_unlock(rq, p, &flags);
				return -EPERM;
			}
		}
#endif
	}
3963

L
Linus Torvalds 已提交
3964 3965 3966
	/* recheck policy now with rq lock held */
	if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) {
		policy = oldpolicy = -1;
3967
		task_rq_unlock(rq, p, &flags);
L
Linus Torvalds 已提交
3968 3969
		goto recheck;
	}
3970 3971 3972 3973 3974 3975

	/*
	 * If setscheduling to SCHED_DEADLINE (or changing the parameters
	 * of a SCHED_DEADLINE task) we need to check if enough bandwidth
	 * is available.
	 */
3976
	if ((dl_policy(policy) || dl_task(p)) && dl_overflow(p, policy, attr)) {
3977 3978 3979 3980
		task_rq_unlock(rq, p, &flags);
		return -EBUSY;
	}

3981 3982 3983
	p->sched_reset_on_fork = reset_on_fork;
	oldprio = p->prio;

3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997
	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);
		if (new_effective_prio == oldprio) {
			__setscheduler_params(p, attr);
			task_rq_unlock(rq, p, &flags);
			return 0;
		}
3998 3999
	}

4000
	queued = task_on_rq_queued(p);
4001
	running = task_current(rq, p);
4002
	if (queued)
4003
		dequeue_task(rq, p, DEQUEUE_SAVE);
4004
	if (running)
4005
		put_prev_task(rq, p);
4006

4007
	prev_class = p->sched_class;
4008
	__setscheduler(rq, p, attr, pi);
4009

4010 4011
	if (running)
		p->sched_class->set_curr_task(rq);
4012
	if (queued) {
4013
		int enqueue_flags = ENQUEUE_RESTORE;
4014 4015 4016 4017
		/*
		 * We enqueue to tail when the priority of a task is
		 * increased (user space view).
		 */
4018 4019 4020 4021
		if (oldprio <= p->prio)
			enqueue_flags |= ENQUEUE_HEAD;

		enqueue_task(rq, p, enqueue_flags);
4022
	}
4023

P
Peter Zijlstra 已提交
4024
	check_class_changed(rq, p, prev_class, oldprio);
4025
	preempt_disable(); /* avoid rq from going away on us */
4026
	task_rq_unlock(rq, p, &flags);
4027

4028 4029
	if (pi)
		rt_mutex_adjust_pi(p);
4030

4031 4032 4033 4034 4035
	/*
	 * Run balance callbacks after we've adjusted the PI chain.
	 */
	balance_callback(rq);
	preempt_enable();
4036

L
Linus Torvalds 已提交
4037 4038
	return 0;
}
4039

4040 4041 4042 4043 4044 4045 4046 4047 4048
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),
	};

4049 4050
	/* Fixup the legacy SCHED_RESET_ON_FORK hack. */
	if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) {
4051 4052 4053 4054 4055
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
		policy &= ~SCHED_RESET_ON_FORK;
		attr.sched_policy = policy;
	}

4056
	return __sched_setscheduler(p, &attr, check, true);
4057
}
4058 4059 4060 4061 4062 4063
/**
 * 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.
 *
4064 4065
 * Return: 0 on success. An error code otherwise.
 *
4066 4067 4068
 * NOTE that the task may be already dead.
 */
int sched_setscheduler(struct task_struct *p, int policy,
4069
		       const struct sched_param *param)
4070
{
4071
	return _sched_setscheduler(p, policy, param, true);
4072
}
L
Linus Torvalds 已提交
4073 4074
EXPORT_SYMBOL_GPL(sched_setscheduler);

4075 4076
int sched_setattr(struct task_struct *p, const struct sched_attr *attr)
{
4077
	return __sched_setscheduler(p, attr, true, true);
4078 4079 4080
}
EXPORT_SYMBOL_GPL(sched_setattr);

4081 4082 4083 4084 4085 4086 4087 4088 4089 4090
/**
 * 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.
4091 4092
 *
 * Return: 0 on success. An error code otherwise.
4093 4094
 */
int sched_setscheduler_nocheck(struct task_struct *p, int policy,
4095
			       const struct sched_param *param)
4096
{
4097
	return _sched_setscheduler(p, policy, param, false);
4098
}
4099
EXPORT_SYMBOL_GPL(sched_setscheduler_nocheck);
4100

I
Ingo Molnar 已提交
4101 4102
static int
do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param)
L
Linus Torvalds 已提交
4103 4104 4105
{
	struct sched_param lparam;
	struct task_struct *p;
4106
	int retval;
L
Linus Torvalds 已提交
4107 4108 4109 4110 4111

	if (!param || pid < 0)
		return -EINVAL;
	if (copy_from_user(&lparam, param, sizeof(struct sched_param)))
		return -EFAULT;
4112 4113 4114

	rcu_read_lock();
	retval = -ESRCH;
L
Linus Torvalds 已提交
4115
	p = find_process_by_pid(pid);
4116 4117 4118
	if (p != NULL)
		retval = sched_setscheduler(p, policy, &lparam);
	rcu_read_unlock();
4119

L
Linus Torvalds 已提交
4120 4121 4122
	return retval;
}

4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184
/*
 * 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?
	 */
4185
	attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE);
4186

4187
	return 0;
4188 4189 4190

err_size:
	put_user(sizeof(*attr), &uattr->size);
4191
	return -E2BIG;
4192 4193
}

L
Linus Torvalds 已提交
4194 4195 4196 4197 4198
/**
 * 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.
4199 4200
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4201
 */
4202 4203
SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy,
		struct sched_param __user *, param)
L
Linus Torvalds 已提交
4204
{
4205 4206 4207 4208
	/* negative values for policy are not valid */
	if (policy < 0)
		return -EINVAL;

L
Linus Torvalds 已提交
4209 4210 4211 4212 4213 4214 4215
	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.
4216 4217
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4218
 */
4219
SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4220
{
4221
	return do_sched_setscheduler(pid, SETPARAM_POLICY, param);
L
Linus Torvalds 已提交
4222 4223
}

4224 4225 4226
/**
 * sys_sched_setattr - same as above, but with extended sched_attr
 * @pid: the pid in question.
J
Juri Lelli 已提交
4227
 * @uattr: structure containing the extended parameters.
4228
 * @flags: for future extension.
4229
 */
4230 4231
SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
			       unsigned int, flags)
4232 4233 4234 4235 4236
{
	struct sched_attr attr;
	struct task_struct *p;
	int retval;

4237
	if (!uattr || pid < 0 || flags)
4238 4239
		return -EINVAL;

4240 4241 4242
	retval = sched_copy_attr(uattr, &attr);
	if (retval)
		return retval;
4243

4244
	if ((int)attr.sched_policy < 0)
4245
		return -EINVAL;
4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256

	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
4257 4258 4259
/**
 * sys_sched_getscheduler - get the policy (scheduling class) of a thread
 * @pid: the pid in question.
4260 4261 4262
 *
 * Return: On success, the policy of the thread. Otherwise, a negative error
 * code.
L
Linus Torvalds 已提交
4263
 */
4264
SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid)
L
Linus Torvalds 已提交
4265
{
4266
	struct task_struct *p;
4267
	int retval;
L
Linus Torvalds 已提交
4268 4269

	if (pid < 0)
4270
		return -EINVAL;
L
Linus Torvalds 已提交
4271 4272

	retval = -ESRCH;
4273
	rcu_read_lock();
L
Linus Torvalds 已提交
4274 4275 4276 4277
	p = find_process_by_pid(pid);
	if (p) {
		retval = security_task_getscheduler(p);
		if (!retval)
4278 4279
			retval = p->policy
				| (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0);
L
Linus Torvalds 已提交
4280
	}
4281
	rcu_read_unlock();
L
Linus Torvalds 已提交
4282 4283 4284 4285
	return retval;
}

/**
4286
 * sys_sched_getparam - get the RT priority of a thread
L
Linus Torvalds 已提交
4287 4288
 * @pid: the pid in question.
 * @param: structure containing the RT priority.
4289 4290 4291
 *
 * Return: On success, 0 and the RT priority is in @param. Otherwise, an error
 * code.
L
Linus Torvalds 已提交
4292
 */
4293
SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param)
L
Linus Torvalds 已提交
4294
{
4295
	struct sched_param lp = { .sched_priority = 0 };
4296
	struct task_struct *p;
4297
	int retval;
L
Linus Torvalds 已提交
4298 4299

	if (!param || pid < 0)
4300
		return -EINVAL;
L
Linus Torvalds 已提交
4301

4302
	rcu_read_lock();
L
Linus Torvalds 已提交
4303 4304 4305 4306 4307 4308 4309 4310 4311
	p = find_process_by_pid(pid);
	retval = -ESRCH;
	if (!p)
		goto out_unlock;

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

4312 4313
	if (task_has_rt_policy(p))
		lp.sched_priority = p->rt_priority;
4314
	rcu_read_unlock();
L
Linus Torvalds 已提交
4315 4316 4317 4318 4319 4320 4321 4322 4323

	/*
	 * 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:
4324
	rcu_read_unlock();
L
Linus Torvalds 已提交
4325 4326 4327
	return retval;
}

4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350
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)
4351
				return -EFBIG;
4352 4353 4354 4355 4356
		}

		attr->size = usize;
	}

4357
	ret = copy_to_user(uattr, attr, attr->size);
4358 4359 4360
	if (ret)
		return -EFAULT;

4361
	return 0;
4362 4363 4364
}

/**
4365
 * sys_sched_getattr - similar to sched_getparam, but with sched_attr
4366
 * @pid: the pid in question.
J
Juri Lelli 已提交
4367
 * @uattr: structure containing the extended parameters.
4368
 * @size: sizeof(attr) for fwd/bwd comp.
4369
 * @flags: for future extension.
4370
 */
4371 4372
SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr,
		unsigned int, size, unsigned int, flags)
4373 4374 4375 4376 4377 4378 4379 4380
{
	struct sched_attr attr = {
		.size = sizeof(struct sched_attr),
	};
	struct task_struct *p;
	int retval;

	if (!uattr || pid < 0 || size > PAGE_SIZE ||
4381
	    size < SCHED_ATTR_SIZE_VER0 || flags)
4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394
		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;
4395 4396
	if (p->sched_reset_on_fork)
		attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK;
4397 4398 4399
	if (task_has_dl_policy(p))
		__getparam_dl(p, &attr);
	else if (task_has_rt_policy(p))
4400 4401
		attr.sched_priority = p->rt_priority;
	else
4402
		attr.sched_nice = task_nice(p);
4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413

	rcu_read_unlock();

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

out_unlock:
	rcu_read_unlock();
	return retval;
}

4414
long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
L
Linus Torvalds 已提交
4415
{
4416
	cpumask_var_t cpus_allowed, new_mask;
4417 4418
	struct task_struct *p;
	int retval;
L
Linus Torvalds 已提交
4419

4420
	rcu_read_lock();
L
Linus Torvalds 已提交
4421 4422 4423

	p = find_process_by_pid(pid);
	if (!p) {
4424
		rcu_read_unlock();
L
Linus Torvalds 已提交
4425 4426 4427
		return -ESRCH;
	}

4428
	/* Prevent p going away */
L
Linus Torvalds 已提交
4429
	get_task_struct(p);
4430
	rcu_read_unlock();
L
Linus Torvalds 已提交
4431

4432 4433 4434 4435
	if (p->flags & PF_NO_SETAFFINITY) {
		retval = -EINVAL;
		goto out_put_task;
	}
4436 4437 4438 4439 4440 4441 4442 4443
	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
4444
	retval = -EPERM;
E
Eric W. Biederman 已提交
4445 4446 4447 4448
	if (!check_same_owner(p)) {
		rcu_read_lock();
		if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
			rcu_read_unlock();
4449
			goto out_free_new_mask;
E
Eric W. Biederman 已提交
4450 4451 4452
		}
		rcu_read_unlock();
	}
L
Linus Torvalds 已提交
4453

4454
	retval = security_task_setscheduler(p);
4455
	if (retval)
4456
		goto out_free_new_mask;
4457

4458 4459 4460 4461

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

4462 4463 4464 4465 4466 4467 4468
	/*
	 * 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
4469 4470 4471
	if (task_has_dl_policy(p) && dl_bandwidth_enabled()) {
		rcu_read_lock();
		if (!cpumask_subset(task_rq(p)->rd->span, new_mask)) {
4472
			retval = -EBUSY;
4473
			rcu_read_unlock();
4474
			goto out_free_new_mask;
4475
		}
4476
		rcu_read_unlock();
4477 4478
	}
#endif
P
Peter Zijlstra 已提交
4479
again:
4480
	retval = __set_cpus_allowed_ptr(p, new_mask, true);
L
Linus Torvalds 已提交
4481

P
Paul Menage 已提交
4482
	if (!retval) {
4483 4484
		cpuset_cpus_allowed(p, cpus_allowed);
		if (!cpumask_subset(new_mask, cpus_allowed)) {
P
Paul Menage 已提交
4485 4486 4487 4488 4489
			/*
			 * We must have raced with a concurrent cpuset
			 * update. Just reset the cpus_allowed to the
			 * cpuset's cpus_allowed
			 */
4490
			cpumask_copy(new_mask, cpus_allowed);
P
Paul Menage 已提交
4491 4492 4493
			goto again;
		}
	}
4494
out_free_new_mask:
4495 4496 4497 4498
	free_cpumask_var(new_mask);
out_free_cpus_allowed:
	free_cpumask_var(cpus_allowed);
out_put_task:
L
Linus Torvalds 已提交
4499 4500 4501 4502 4503
	put_task_struct(p);
	return retval;
}

static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len,
4504
			     struct cpumask *new_mask)
L
Linus Torvalds 已提交
4505
{
4506 4507 4508 4509 4510
	if (len < cpumask_size())
		cpumask_clear(new_mask);
	else if (len > cpumask_size())
		len = cpumask_size();

L
Linus Torvalds 已提交
4511 4512 4513 4514 4515 4516 4517 4518
	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
4519 4520
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4521
 */
4522 4523
SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4524
{
4525
	cpumask_var_t new_mask;
L
Linus Torvalds 已提交
4526 4527
	int retval;

4528 4529
	if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4530

4531 4532 4533 4534 4535
	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
4536 4537
}

4538
long sched_getaffinity(pid_t pid, struct cpumask *mask)
L
Linus Torvalds 已提交
4539
{
4540
	struct task_struct *p;
4541
	unsigned long flags;
L
Linus Torvalds 已提交
4542 4543
	int retval;

4544
	rcu_read_lock();
L
Linus Torvalds 已提交
4545 4546 4547 4548 4549 4550

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

4551 4552 4553 4554
	retval = security_task_getscheduler(p);
	if (retval)
		goto out_unlock;

4555
	raw_spin_lock_irqsave(&p->pi_lock, flags);
4556
	cpumask_and(mask, &p->cpus_allowed, cpu_active_mask);
4557
	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
L
Linus Torvalds 已提交
4558 4559

out_unlock:
4560
	rcu_read_unlock();
L
Linus Torvalds 已提交
4561

4562
	return retval;
L
Linus Torvalds 已提交
4563 4564 4565 4566 4567 4568 4569
}

/**
 * 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
4570 4571
 *
 * Return: 0 on success. An error code otherwise.
L
Linus Torvalds 已提交
4572
 */
4573 4574
SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len,
		unsigned long __user *, user_mask_ptr)
L
Linus Torvalds 已提交
4575 4576
{
	int ret;
4577
	cpumask_var_t mask;
L
Linus Torvalds 已提交
4578

A
Anton Blanchard 已提交
4579
	if ((len * BITS_PER_BYTE) < nr_cpu_ids)
4580 4581
		return -EINVAL;
	if (len & (sizeof(unsigned long)-1))
L
Linus Torvalds 已提交
4582 4583
		return -EINVAL;

4584 4585
	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
		return -ENOMEM;
L
Linus Torvalds 已提交
4586

4587 4588
	ret = sched_getaffinity(pid, mask);
	if (ret == 0) {
4589
		size_t retlen = min_t(size_t, len, cpumask_size());
4590 4591

		if (copy_to_user(user_mask_ptr, mask, retlen))
4592 4593
			ret = -EFAULT;
		else
4594
			ret = retlen;
4595 4596
	}
	free_cpumask_var(mask);
L
Linus Torvalds 已提交
4597

4598
	return ret;
L
Linus Torvalds 已提交
4599 4600 4601 4602 4603
}

/**
 * sys_sched_yield - yield the current processor to other threads.
 *
I
Ingo Molnar 已提交
4604 4605
 * This function yields the current CPU to other tasks. If there are no
 * other threads running on this CPU then this function will return.
4606 4607
 *
 * Return: 0.
L
Linus Torvalds 已提交
4608
 */
4609
SYSCALL_DEFINE0(sched_yield)
L
Linus Torvalds 已提交
4610
{
4611
	struct rq *rq = this_rq_lock();
L
Linus Torvalds 已提交
4612

4613
	schedstat_inc(rq, yld_count);
4614
	current->sched_class->yield_task(rq);
L
Linus Torvalds 已提交
4615 4616 4617 4618 4619 4620

	/*
	 * Since we are going to call schedule() anyway, there's
	 * no need to preempt or enable interrupts:
	 */
	__release(rq->lock);
4621
	spin_release(&rq->lock.dep_map, 1, _THIS_IP_);
4622
	do_raw_spin_unlock(&rq->lock);
4623
	sched_preempt_enable_no_resched();
L
Linus Torvalds 已提交
4624 4625 4626 4627 4628 4629

	schedule();

	return 0;
}

4630
int __sched _cond_resched(void)
L
Linus Torvalds 已提交
4631
{
4632
	if (should_resched(0)) {
4633
		preempt_schedule_common();
L
Linus Torvalds 已提交
4634 4635 4636 4637
		return 1;
	}
	return 0;
}
4638
EXPORT_SYMBOL(_cond_resched);
L
Linus Torvalds 已提交
4639 4640

/*
4641
 * __cond_resched_lock() - if a reschedule is pending, drop the given lock,
L
Linus Torvalds 已提交
4642 4643
 * call schedule, and on return reacquire the lock.
 *
I
Ingo Molnar 已提交
4644
 * This works OK both with and without CONFIG_PREEMPT. We do strange low-level
L
Linus Torvalds 已提交
4645 4646 4647
 * operations here to prevent schedule() from being called twice (once via
 * spin_unlock(), once by hand).
 */
4648
int __cond_resched_lock(spinlock_t *lock)
L
Linus Torvalds 已提交
4649
{
4650
	int resched = should_resched(PREEMPT_LOCK_OFFSET);
J
Jan Kara 已提交
4651 4652
	int ret = 0;

4653 4654
	lockdep_assert_held(lock);

4655
	if (spin_needbreak(lock) || resched) {
L
Linus Torvalds 已提交
4656
		spin_unlock(lock);
P
Peter Zijlstra 已提交
4657
		if (resched)
4658
			preempt_schedule_common();
N
Nick Piggin 已提交
4659 4660
		else
			cpu_relax();
J
Jan Kara 已提交
4661
		ret = 1;
L
Linus Torvalds 已提交
4662 4663
		spin_lock(lock);
	}
J
Jan Kara 已提交
4664
	return ret;
L
Linus Torvalds 已提交
4665
}
4666
EXPORT_SYMBOL(__cond_resched_lock);
L
Linus Torvalds 已提交
4667

4668
int __sched __cond_resched_softirq(void)
L
Linus Torvalds 已提交
4669 4670 4671
{
	BUG_ON(!in_softirq());

4672
	if (should_resched(SOFTIRQ_DISABLE_OFFSET)) {
4673
		local_bh_enable();
4674
		preempt_schedule_common();
L
Linus Torvalds 已提交
4675 4676 4677 4678 4679
		local_bh_disable();
		return 1;
	}
	return 0;
}
4680
EXPORT_SYMBOL(__cond_resched_softirq);
L
Linus Torvalds 已提交
4681 4682 4683 4684

/**
 * yield - yield the current processor to other threads.
 *
P
Peter Zijlstra 已提交
4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702
 * 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
4703 4704 4705 4706 4707 4708 4709 4710
 */
void __sched yield(void)
{
	set_current_state(TASK_RUNNING);
	sys_sched_yield();
}
EXPORT_SYMBOL(yield);

4711 4712 4713 4714
/**
 * 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
4715 4716
 * @p: target task
 * @preempt: whether task preemption is allowed or not
4717 4718 4719 4720
 *
 * 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.
 *
4721
 * Return:
4722 4723 4724
 *	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.
4725
 */
4726
int __sched yield_to(struct task_struct *p, bool preempt)
4727 4728 4729 4730
{
	struct task_struct *curr = current;
	struct rq *rq, *p_rq;
	unsigned long flags;
4731
	int yielded = 0;
4732 4733 4734 4735 4736 4737

	local_irq_save(flags);
	rq = this_rq();

again:
	p_rq = task_rq(p);
4738 4739 4740 4741 4742 4743 4744 4745 4746
	/*
	 * 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;
	}

4747
	double_rq_lock(rq, p_rq);
4748
	if (task_rq(p) != p_rq) {
4749 4750 4751 4752 4753
		double_rq_unlock(rq, p_rq);
		goto again;
	}

	if (!curr->sched_class->yield_to_task)
4754
		goto out_unlock;
4755 4756

	if (curr->sched_class != p->sched_class)
4757
		goto out_unlock;
4758 4759

	if (task_running(p_rq, p) || p->state)
4760
		goto out_unlock;
4761 4762

	yielded = curr->sched_class->yield_to_task(rq, p, preempt);
4763
	if (yielded) {
4764
		schedstat_inc(rq, yld_count);
4765 4766 4767 4768 4769
		/*
		 * Make p's CPU reschedule; pick_next_entity takes care of
		 * fairness.
		 */
		if (preempt && rq != p_rq)
4770
			resched_curr(p_rq);
4771
	}
4772

4773
out_unlock:
4774
	double_rq_unlock(rq, p_rq);
4775
out_irq:
4776 4777
	local_irq_restore(flags);

4778
	if (yielded > 0)
4779 4780 4781 4782 4783 4784
		schedule();

	return yielded;
}
EXPORT_SYMBOL_GPL(yield_to);

L
Linus Torvalds 已提交
4785
/*
I
Ingo Molnar 已提交
4786
 * This task is about to go to sleep on IO. Increment rq->nr_iowait so
L
Linus Torvalds 已提交
4787 4788 4789 4790
 * that process accounting knows that this is a task in IO wait state.
 */
long __sched io_schedule_timeout(long timeout)
{
4791 4792
	int old_iowait = current->in_iowait;
	struct rq *rq;
L
Linus Torvalds 已提交
4793 4794
	long ret;

4795
	current->in_iowait = 1;
4796
	blk_schedule_flush_plug(current);
4797

4798
	delayacct_blkio_start();
4799
	rq = raw_rq();
L
Linus Torvalds 已提交
4800 4801
	atomic_inc(&rq->nr_iowait);
	ret = schedule_timeout(timeout);
4802
	current->in_iowait = old_iowait;
L
Linus Torvalds 已提交
4803
	atomic_dec(&rq->nr_iowait);
4804
	delayacct_blkio_end();
4805

L
Linus Torvalds 已提交
4806 4807
	return ret;
}
4808
EXPORT_SYMBOL(io_schedule_timeout);
L
Linus Torvalds 已提交
4809 4810 4811 4812 4813

/**
 * sys_sched_get_priority_max - return maximum RT priority.
 * @policy: scheduling class.
 *
4814 4815 4816
 * 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
4817
 */
4818
SYSCALL_DEFINE1(sched_get_priority_max, int, policy)
L
Linus Torvalds 已提交
4819 4820 4821 4822 4823 4824 4825 4826
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = MAX_USER_RT_PRIO-1;
		break;
4827
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
4828
	case SCHED_NORMAL:
4829
	case SCHED_BATCH:
I
Ingo Molnar 已提交
4830
	case SCHED_IDLE:
L
Linus Torvalds 已提交
4831 4832 4833 4834 4835 4836 4837 4838 4839 4840
		ret = 0;
		break;
	}
	return ret;
}

/**
 * sys_sched_get_priority_min - return minimum RT priority.
 * @policy: scheduling class.
 *
4841 4842 4843
 * 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
4844
 */
4845
SYSCALL_DEFINE1(sched_get_priority_min, int, policy)
L
Linus Torvalds 已提交
4846 4847 4848 4849 4850 4851 4852 4853
{
	int ret = -EINVAL;

	switch (policy) {
	case SCHED_FIFO:
	case SCHED_RR:
		ret = 1;
		break;
4854
	case SCHED_DEADLINE:
L
Linus Torvalds 已提交
4855
	case SCHED_NORMAL:
4856
	case SCHED_BATCH:
I
Ingo Molnar 已提交
4857
	case SCHED_IDLE:
L
Linus Torvalds 已提交
4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869
		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.
4870 4871 4872
 *
 * Return: On success, 0 and the timeslice is in @interval. Otherwise,
 * an error code.
L
Linus Torvalds 已提交
4873
 */
4874
SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid,
4875
		struct timespec __user *, interval)
L
Linus Torvalds 已提交
4876
{
4877
	struct task_struct *p;
D
Dmitry Adamushko 已提交
4878
	unsigned int time_slice;
4879 4880
	unsigned long flags;
	struct rq *rq;
4881
	int retval;
L
Linus Torvalds 已提交
4882 4883 4884
	struct timespec t;

	if (pid < 0)
4885
		return -EINVAL;
L
Linus Torvalds 已提交
4886 4887

	retval = -ESRCH;
4888
	rcu_read_lock();
L
Linus Torvalds 已提交
4889 4890 4891 4892 4893 4894 4895 4896
	p = find_process_by_pid(pid);
	if (!p)
		goto out_unlock;

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

4897
	rq = task_rq_lock(p, &flags);
4898 4899 4900
	time_slice = 0;
	if (p->sched_class->get_rr_interval)
		time_slice = p->sched_class->get_rr_interval(rq, p);
4901
	task_rq_unlock(rq, p, &flags);
D
Dmitry Adamushko 已提交
4902

4903
	rcu_read_unlock();
D
Dmitry Adamushko 已提交
4904
	jiffies_to_timespec(time_slice, &t);
L
Linus Torvalds 已提交
4905 4906
	retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0;
	return retval;
4907

L
Linus Torvalds 已提交
4908
out_unlock:
4909
	rcu_read_unlock();
L
Linus Torvalds 已提交
4910 4911 4912
	return retval;
}

4913
static const char stat_nam[] = TASK_STATE_TO_CHAR_STR;
4914

4915
void sched_show_task(struct task_struct *p)
L
Linus Torvalds 已提交
4916 4917
{
	unsigned long free = 0;
4918
	int ppid;
4919
	unsigned long state = p->state;
L
Linus Torvalds 已提交
4920

4921 4922
	if (state)
		state = __ffs(state) + 1;
4923
	printk(KERN_INFO "%-15.15s %c", p->comm,
4924
		state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?');
4925
#if BITS_PER_LONG == 32
L
Linus Torvalds 已提交
4926
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
4927
		printk(KERN_CONT " running  ");
L
Linus Torvalds 已提交
4928
	else
P
Peter Zijlstra 已提交
4929
		printk(KERN_CONT " %08lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
4930 4931
#else
	if (state == TASK_RUNNING)
P
Peter Zijlstra 已提交
4932
		printk(KERN_CONT "  running task    ");
L
Linus Torvalds 已提交
4933
	else
P
Peter Zijlstra 已提交
4934
		printk(KERN_CONT " %016lx ", thread_saved_pc(p));
L
Linus Torvalds 已提交
4935 4936
#endif
#ifdef CONFIG_DEBUG_STACK_USAGE
4937
	free = stack_not_used(p);
L
Linus Torvalds 已提交
4938
#endif
4939
	ppid = 0;
4940
	rcu_read_lock();
4941 4942
	if (pid_alive(p))
		ppid = task_pid_nr(rcu_dereference(p->real_parent));
4943
	rcu_read_unlock();
P
Peter Zijlstra 已提交
4944
	printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free,
4945
		task_pid_nr(p), ppid,
4946
		(unsigned long)task_thread_info(p)->flags);
L
Linus Torvalds 已提交
4947

4948
	print_worker_info(KERN_INFO, p);
4949
	show_stack(p, NULL);
L
Linus Torvalds 已提交
4950 4951
}

I
Ingo Molnar 已提交
4952
void show_state_filter(unsigned long state_filter)
L
Linus Torvalds 已提交
4953
{
4954
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
4955

4956
#if BITS_PER_LONG == 32
P
Peter Zijlstra 已提交
4957 4958
	printk(KERN_INFO
		"  task                PC stack   pid father\n");
L
Linus Torvalds 已提交
4959
#else
P
Peter Zijlstra 已提交
4960 4961
	printk(KERN_INFO
		"  task                        PC stack   pid father\n");
L
Linus Torvalds 已提交
4962
#endif
4963
	rcu_read_lock();
4964
	for_each_process_thread(g, p) {
L
Linus Torvalds 已提交
4965 4966
		/*
		 * reset the NMI-timeout, listing all files on a slow
L
Lucas De Marchi 已提交
4967
		 * console might take a lot of time:
L
Linus Torvalds 已提交
4968 4969
		 */
		touch_nmi_watchdog();
I
Ingo Molnar 已提交
4970
		if (!state_filter || (p->state & state_filter))
4971
			sched_show_task(p);
4972
	}
L
Linus Torvalds 已提交
4973

4974 4975
	touch_all_softlockup_watchdogs();

I
Ingo Molnar 已提交
4976 4977 4978
#ifdef CONFIG_SCHED_DEBUG
	sysrq_sched_debug_show();
#endif
4979
	rcu_read_unlock();
I
Ingo Molnar 已提交
4980 4981 4982
	/*
	 * Only show locks if all tasks are dumped:
	 */
4983
	if (!state_filter)
I
Ingo Molnar 已提交
4984
		debug_show_all_locks();
L
Linus Torvalds 已提交
4985 4986
}

4987
void init_idle_bootup_task(struct task_struct *idle)
I
Ingo Molnar 已提交
4988
{
I
Ingo Molnar 已提交
4989
	idle->sched_class = &idle_sched_class;
I
Ingo Molnar 已提交
4990 4991
}

4992 4993 4994 4995 4996 4997 4998 4999
/**
 * 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.
 */
5000
void init_idle(struct task_struct *idle, int cpu)
L
Linus Torvalds 已提交
5001
{
5002
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5003 5004
	unsigned long flags;

5005 5006
	raw_spin_lock_irqsave(&idle->pi_lock, flags);
	raw_spin_lock(&rq->lock);
5007

5008
	__sched_fork(0, idle);
5009
	idle->state = TASK_RUNNING;
I
Ingo Molnar 已提交
5010 5011
	idle->se.exec_start = sched_clock();

5012 5013 5014 5015 5016 5017 5018 5019 5020
#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
5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031
	/*
	 * 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
5032
	__set_task_cpu(idle, cpu);
5033
	rcu_read_unlock();
L
Linus Torvalds 已提交
5034 5035

	rq->curr = rq->idle = idle;
5036
	idle->on_rq = TASK_ON_RQ_QUEUED;
5037
#ifdef CONFIG_SMP
P
Peter Zijlstra 已提交
5038
	idle->on_cpu = 1;
5039
#endif
5040 5041
	raw_spin_unlock(&rq->lock);
	raw_spin_unlock_irqrestore(&idle->pi_lock, flags);
L
Linus Torvalds 已提交
5042 5043

	/* Set the preempt count _outside_ the spinlocks! */
5044
	init_idle_preempt_count(idle, cpu);
5045

I
Ingo Molnar 已提交
5046 5047 5048 5049
	/*
	 * The idle tasks have their own, simple scheduling class:
	 */
	idle->sched_class = &idle_sched_class;
5050
	ftrace_graph_init_idle_task(idle, cpu);
5051
	vtime_init_idle(idle, cpu);
5052
#ifdef CONFIG_SMP
5053 5054
	sprintf(idle->comm, "%s/%d", INIT_TASK_COMM, cpu);
#endif
I
Ingo Molnar 已提交
5055 5056
}

5057 5058 5059 5060 5061 5062 5063
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;

5064 5065 5066
	if (!cpumask_weight(cur))
		return ret;

5067
	rcu_read_lock_sched();
5068 5069 5070 5071 5072 5073 5074 5075
	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);
5076
	rcu_read_unlock_sched();
5077 5078 5079 5080

	return ret;
}

5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104
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);
5105
		struct dl_bw *dl_b;
5106 5107 5108 5109
		bool overflow;
		int cpus;
		unsigned long flags;

5110 5111
		rcu_read_lock_sched();
		dl_b = dl_bw_of(dest_cpu);
5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126
		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);
5127
		rcu_read_unlock_sched();
5128 5129 5130 5131 5132 5133 5134

	}
#endif
out:
	return ret;
}

L
Linus Torvalds 已提交
5135 5136
#ifdef CONFIG_SMP

5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151
#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 */

5152
	trace_sched_move_numa(p, curr_cpu, target_cpu);
5153 5154
	return stop_one_cpu(curr_cpu, migration_cpu_stop, &arg);
}
5155 5156 5157 5158 5159 5160 5161 5162 5163

/*
 * 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)
{
	struct rq *rq;
	unsigned long flags;
5164
	bool queued, running;
5165 5166

	rq = task_rq_lock(p, &flags);
5167
	queued = task_on_rq_queued(p);
5168 5169
	running = task_current(rq, p);

5170
	if (queued)
5171
		dequeue_task(rq, p, DEQUEUE_SAVE);
5172
	if (running)
5173
		put_prev_task(rq, p);
5174 5175 5176 5177 5178

	p->numa_preferred_nid = nid;

	if (running)
		p->sched_class->set_curr_task(rq);
5179
	if (queued)
5180
		enqueue_task(rq, p, ENQUEUE_RESTORE);
5181 5182
	task_rq_unlock(rq, p, &flags);
}
P
Peter Zijlstra 已提交
5183
#endif /* CONFIG_NUMA_BALANCING */
5184

L
Linus Torvalds 已提交
5185
#ifdef CONFIG_HOTPLUG_CPU
5186
/*
5187 5188
 * Ensures that the idle task is using init_mm right before its cpu goes
 * offline.
5189
 */
5190
void idle_task_exit(void)
L
Linus Torvalds 已提交
5191
{
5192
	struct mm_struct *mm = current->active_mm;
5193

5194
	BUG_ON(cpu_online(smp_processor_id()));
5195

5196
	if (mm != &init_mm) {
5197
		switch_mm(mm, &init_mm, current);
5198 5199
		finish_arch_post_lock_switch();
	}
5200
	mmdrop(mm);
L
Linus Torvalds 已提交
5201 5202 5203
}

/*
5204 5205 5206 5207 5208
 * 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
5209
 */
5210
static void calc_load_migrate(struct rq *rq)
L
Linus Torvalds 已提交
5211
{
5212 5213 5214
	long delta = calc_load_fold_active(rq);
	if (delta)
		atomic_long_add(delta, &calc_load_tasks);
L
Linus Torvalds 已提交
5215 5216
}

5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232
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,
};

5233
/*
5234 5235 5236 5237 5238 5239
 * 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
5240
 */
5241
static void migrate_tasks(struct rq *dead_rq)
L
Linus Torvalds 已提交
5242
{
5243
	struct rq *rq = dead_rq;
5244 5245
	struct task_struct *next, *stop = rq->stop;
	int dest_cpu;
L
Linus Torvalds 已提交
5246 5247

	/*
5248 5249 5250 5251 5252 5253 5254
	 * 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
5255
	 */
5256
	rq->stop = NULL;
5257

5258 5259 5260 5261 5262 5263 5264
	/*
	 * 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);

5265
	for (;;) {
5266 5267 5268 5269 5270
		/*
		 * There's this thread running, bail when that's the only
		 * remaining thread.
		 */
		if (rq->nr_running == 1)
I
Ingo Molnar 已提交
5271
			break;
5272

5273
		/*
W
Wanpeng Li 已提交
5274
		 * pick_next_task assumes pinned rq->lock.
5275 5276
		 */
		lockdep_pin_lock(&rq->lock);
5277
		next = pick_next_task(rq, &fake_task);
5278
		BUG_ON(!next);
D
Dmitry Adamushko 已提交
5279
		next->sched_class->put_prev_task(rq, next);
5280

W
Wanpeng Li 已提交
5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304
		/*
		 * 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.
		 */
		lockdep_unpin_lock(&rq->lock);
		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;
		}

5305
		/* Find suitable destination for @next, with force if needed. */
5306
		dest_cpu = select_fallback_rq(dead_rq->cpu, next);
5307

5308 5309 5310 5311 5312 5313
		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
5314
		raw_spin_unlock(&next->pi_lock);
L
Linus Torvalds 已提交
5315
	}
5316

5317
	rq->stop = stop;
5318
}
L
Linus Torvalds 已提交
5319 5320
#endif /* CONFIG_HOTPLUG_CPU */

5321 5322 5323
#if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL)

static struct ctl_table sd_ctl_dir[] = {
5324 5325
	{
		.procname	= "sched_domain",
5326
		.mode		= 0555,
5327
	},
5328
	{}
5329 5330 5331
};

static struct ctl_table sd_ctl_root[] = {
5332 5333
	{
		.procname	= "kernel",
5334
		.mode		= 0555,
5335 5336
		.child		= sd_ctl_dir,
	},
5337
	{}
5338 5339 5340 5341 5342
};

static struct ctl_table *sd_alloc_ctl_entry(int n)
{
	struct ctl_table *entry =
5343
		kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL);
5344 5345 5346 5347

	return entry;
}

5348 5349
static void sd_free_ctl_entry(struct ctl_table **tablep)
{
5350
	struct ctl_table *entry;
5351

5352 5353 5354
	/*
	 * In the intermediate directories, both the child directory and
	 * procname are dynamically allocated and could fail but the mode
I
Ingo Molnar 已提交
5355
	 * will always be set. In the lowest directory the names are
5356 5357 5358
	 * static strings and all have proc handlers.
	 */
	for (entry = *tablep; entry->mode; entry++) {
5359 5360
		if (entry->child)
			sd_free_ctl_entry(&entry->child);
5361 5362 5363
		if (entry->proc_handler == NULL)
			kfree(entry->procname);
	}
5364 5365 5366 5367 5368

	kfree(*tablep);
	*tablep = NULL;
}

5369
static int min_load_idx = 0;
5370
static int max_load_idx = CPU_LOAD_IDX_MAX-1;
5371

5372
static void
5373
set_table_entry(struct ctl_table *entry,
5374
		const char *procname, void *data, int maxlen,
5375 5376
		umode_t mode, proc_handler *proc_handler,
		bool load_idx)
5377 5378 5379 5380 5381 5382
{
	entry->procname = procname;
	entry->data = data;
	entry->maxlen = maxlen;
	entry->mode = mode;
	entry->proc_handler = proc_handler;
5383 5384 5385 5386 5387

	if (load_idx) {
		entry->extra1 = &min_load_idx;
		entry->extra2 = &max_load_idx;
	}
5388 5389 5390 5391 5392
}

static struct ctl_table *
sd_alloc_ctl_domain_table(struct sched_domain *sd)
{
5393
	struct ctl_table *table = sd_alloc_ctl_entry(14);
5394

5395 5396 5397
	if (table == NULL)
		return NULL;

5398
	set_table_entry(&table[0], "min_interval", &sd->min_interval,
5399
		sizeof(long), 0644, proc_doulongvec_minmax, false);
5400
	set_table_entry(&table[1], "max_interval", &sd->max_interval,
5401
		sizeof(long), 0644, proc_doulongvec_minmax, false);
5402
	set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
5403
		sizeof(int), 0644, proc_dointvec_minmax, true);
5404
	set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
5405
		sizeof(int), 0644, proc_dointvec_minmax, true);
5406
	set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
5407
		sizeof(int), 0644, proc_dointvec_minmax, true);
5408
	set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
5409
		sizeof(int), 0644, proc_dointvec_minmax, true);
5410
	set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
5411
		sizeof(int), 0644, proc_dointvec_minmax, true);
5412
	set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
5413
		sizeof(int), 0644, proc_dointvec_minmax, false);
5414
	set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
5415
		sizeof(int), 0644, proc_dointvec_minmax, false);
5416
	set_table_entry(&table[9], "cache_nice_tries",
5417
		&sd->cache_nice_tries,
5418
		sizeof(int), 0644, proc_dointvec_minmax, false);
5419
	set_table_entry(&table[10], "flags", &sd->flags,
5420
		sizeof(int), 0644, proc_dointvec_minmax, false);
5421 5422 5423 5424
	set_table_entry(&table[11], "max_newidle_lb_cost",
		&sd->max_newidle_lb_cost,
		sizeof(long), 0644, proc_doulongvec_minmax, false);
	set_table_entry(&table[12], "name", sd->name,
5425
		CORENAME_MAX_SIZE, 0444, proc_dostring, false);
5426
	/* &table[13] is terminator */
5427 5428 5429 5430

	return table;
}

5431
static struct ctl_table *sd_alloc_ctl_cpu_table(int cpu)
5432 5433 5434 5435 5436 5437 5438 5439 5440
{
	struct ctl_table *entry, *table;
	struct sched_domain *sd;
	int domain_num = 0, i;
	char buf[32];

	for_each_domain(cpu, sd)
		domain_num++;
	entry = table = sd_alloc_ctl_entry(domain_num + 1);
5441 5442
	if (table == NULL)
		return NULL;
5443 5444 5445 5446 5447

	i = 0;
	for_each_domain(cpu, sd) {
		snprintf(buf, 32, "domain%d", i);
		entry->procname = kstrdup(buf, GFP_KERNEL);
5448
		entry->mode = 0555;
5449 5450 5451 5452 5453 5454 5455 5456
		entry->child = sd_alloc_ctl_domain_table(sd);
		entry++;
		i++;
	}
	return table;
}

static struct ctl_table_header *sd_sysctl_header;
5457
static void register_sched_domain_sysctl(void)
5458
{
5459
	int i, cpu_num = num_possible_cpus();
5460 5461 5462
	struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1);
	char buf[32];

5463 5464 5465
	WARN_ON(sd_ctl_dir[0].child);
	sd_ctl_dir[0].child = entry;

5466 5467 5468
	if (entry == NULL)
		return;

5469
	for_each_possible_cpu(i) {
5470 5471
		snprintf(buf, 32, "cpu%d", i);
		entry->procname = kstrdup(buf, GFP_KERNEL);
5472
		entry->mode = 0555;
5473
		entry->child = sd_alloc_ctl_cpu_table(i);
5474
		entry++;
5475
	}
5476 5477

	WARN_ON(sd_sysctl_header);
5478 5479
	sd_sysctl_header = register_sysctl_table(sd_ctl_root);
}
5480

5481
/* may be called multiple times per register */
5482 5483
static void unregister_sched_domain_sysctl(void)
{
5484
	unregister_sysctl_table(sd_sysctl_header);
5485
	sd_sysctl_header = NULL;
5486 5487
	if (sd_ctl_dir[0].child)
		sd_free_ctl_entry(&sd_ctl_dir[0].child);
5488
}
5489
#else
5490 5491 5492 5493
static void register_sched_domain_sysctl(void)
{
}
static void unregister_sched_domain_sysctl(void)
5494 5495
{
}
P
Peter Zijlstra 已提交
5496
#endif /* CONFIG_SCHED_DEBUG && CONFIG_SYSCTL */
5497

5498 5499 5500 5501 5502
static void set_rq_online(struct rq *rq)
{
	if (!rq->online) {
		const struct sched_class *class;

5503
		cpumask_set_cpu(rq->cpu, rq->rd->online);
5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522
		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);
		}

5523
		cpumask_clear_cpu(rq->cpu, rq->rd->online);
5524 5525 5526 5527
		rq->online = 0;
	}
}

L
Linus Torvalds 已提交
5528 5529 5530 5531
/*
 * migration_call - callback that gets triggered when a CPU is added.
 * Here we can start up the necessary migration thread for the new CPU.
 */
5532
static int
5533
migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
L
Linus Torvalds 已提交
5534
{
5535
	int cpu = (long)hcpu;
L
Linus Torvalds 已提交
5536
	unsigned long flags;
5537
	struct rq *rq = cpu_rq(cpu);
L
Linus Torvalds 已提交
5538

5539
	switch (action & ~CPU_TASKS_FROZEN) {
5540

L
Linus Torvalds 已提交
5541
	case CPU_UP_PREPARE:
5542
		rq->calc_load_update = calc_load_update;
L
Linus Torvalds 已提交
5543
		break;
5544

L
Linus Torvalds 已提交
5545
	case CPU_ONLINE:
5546
		/* Update our root-domain */
5547
		raw_spin_lock_irqsave(&rq->lock, flags);
5548
		if (rq->rd) {
5549
			BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5550 5551

			set_rq_online(rq);
5552
		}
5553
		raw_spin_unlock_irqrestore(&rq->lock, flags);
L
Linus Torvalds 已提交
5554
		break;
5555

L
Linus Torvalds 已提交
5556
#ifdef CONFIG_HOTPLUG_CPU
5557
	case CPU_DYING:
5558
		sched_ttwu_pending();
G
Gregory Haskins 已提交
5559
		/* Update our root-domain */
5560
		raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5561
		if (rq->rd) {
5562
			BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span));
5563
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5564
		}
5565
		migrate_tasks(rq);
5566
		BUG_ON(rq->nr_running != 1); /* the migration thread */
5567
		raw_spin_unlock_irqrestore(&rq->lock, flags);
5568
		break;
5569

5570
	case CPU_DEAD:
5571
		calc_load_migrate(rq);
G
Gregory Haskins 已提交
5572
		break;
L
Linus Torvalds 已提交
5573 5574
#endif
	}
5575 5576 5577

	update_max_interval();

L
Linus Torvalds 已提交
5578 5579 5580
	return NOTIFY_OK;
}

5581 5582 5583
/*
 * Register at high priority so that task migration (migrate_all_tasks)
 * happens before everything else.  This has to be lower priority than
5584
 * the notifier in the perf_event subsystem, though.
L
Linus Torvalds 已提交
5585
 */
5586
static struct notifier_block migration_notifier = {
L
Linus Torvalds 已提交
5587
	.notifier_call = migration_call,
5588
	.priority = CPU_PRI_MIGRATION,
L
Linus Torvalds 已提交
5589 5590
};

5591
static void set_cpu_rq_start_time(void)
5592 5593 5594 5595 5596 5597
{
	int cpu = smp_processor_id();
	struct rq *rq = cpu_rq(cpu);
	rq->age_stamp = sched_clock_cpu(cpu);
}

5598
static int sched_cpu_active(struct notifier_block *nfb,
5599 5600
				      unsigned long action, void *hcpu)
{
P
Peter Zijlstra 已提交
5601 5602
	int cpu = (long)hcpu;

5603
	switch (action & ~CPU_TASKS_FROZEN) {
5604 5605 5606
	case CPU_STARTING:
		set_cpu_rq_start_time();
		return NOTIFY_OK;
P
Peter Zijlstra 已提交
5607

5608 5609 5610 5611 5612 5613
	case CPU_ONLINE:
		/*
		 * At this point a starting CPU has marked itself as online via
		 * set_cpu_online(). But it might not yet have marked itself
		 * as active, which is essential from here on.
		 */
P
Peter Zijlstra 已提交
5614 5615 5616 5617
		set_cpu_active(cpu, true);
		stop_machine_unpark(cpu);
		return NOTIFY_OK;

5618
	case CPU_DOWN_FAILED:
P
Peter Zijlstra 已提交
5619
		set_cpu_active(cpu, true);
5620
		return NOTIFY_OK;
P
Peter Zijlstra 已提交
5621

5622 5623 5624 5625 5626
	default:
		return NOTIFY_DONE;
	}
}

5627
static int sched_cpu_inactive(struct notifier_block *nfb,
5628 5629 5630 5631
					unsigned long action, void *hcpu)
{
	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DOWN_PREPARE:
5632
		set_cpu_active((long)hcpu, false);
5633
		return NOTIFY_OK;
5634 5635
	default:
		return NOTIFY_DONE;
5636 5637 5638
	}
}

5639
static int __init migration_init(void)
L
Linus Torvalds 已提交
5640 5641
{
	void *cpu = (void *)(long)smp_processor_id();
5642
	int err;
5643

5644
	/* Initialize migration for the boot CPU */
5645 5646
	err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu);
	BUG_ON(err == NOTIFY_BAD);
L
Linus Torvalds 已提交
5647 5648
	migration_call(&migration_notifier, CPU_ONLINE, cpu);
	register_cpu_notifier(&migration_notifier);
5649

5650 5651 5652 5653
	/* Register cpu active notifiers */
	cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE);
	cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE);

5654
	return 0;
L
Linus Torvalds 已提交
5655
}
5656
early_initcall(migration_init);
5657

5658 5659
static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */

5660
#ifdef CONFIG_SCHED_DEBUG
I
Ingo Molnar 已提交
5661

5662
static __read_mostly int sched_debug_enabled;
5663

5664
static int __init sched_debug_setup(char *str)
5665
{
5666
	sched_debug_enabled = 1;
5667 5668 5669

	return 0;
}
5670 5671 5672 5673 5674 5675
early_param("sched_debug", sched_debug_setup);

static inline bool sched_debug(void)
{
	return sched_debug_enabled;
}
5676

5677
static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
5678
				  struct cpumask *groupmask)
L
Linus Torvalds 已提交
5679
{
I
Ingo Molnar 已提交
5680
	struct sched_group *group = sd->groups;
L
Linus Torvalds 已提交
5681

5682
	cpumask_clear(groupmask);
I
Ingo Molnar 已提交
5683 5684 5685 5686

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

	if (!(sd->flags & SD_LOAD_BALANCE)) {
P
Peter Zijlstra 已提交
5687
		printk("does not load-balance\n");
I
Ingo Molnar 已提交
5688
		if (sd->parent)
P
Peter Zijlstra 已提交
5689 5690
			printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain"
					" has parent");
I
Ingo Molnar 已提交
5691
		return -1;
N
Nick Piggin 已提交
5692 5693
	}

5694 5695
	printk(KERN_CONT "span %*pbl level %s\n",
	       cpumask_pr_args(sched_domain_span(sd)), sd->name);
I
Ingo Molnar 已提交
5696

5697
	if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) {
P
Peter Zijlstra 已提交
5698 5699
		printk(KERN_ERR "ERROR: domain->span does not contain "
				"CPU%d\n", cpu);
I
Ingo Molnar 已提交
5700
	}
5701
	if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5702 5703
		printk(KERN_ERR "ERROR: domain->groups does not contain"
				" CPU%d\n", cpu);
I
Ingo Molnar 已提交
5704
	}
L
Linus Torvalds 已提交
5705

I
Ingo Molnar 已提交
5706
	printk(KERN_DEBUG "%*s groups:", level + 1, "");
L
Linus Torvalds 已提交
5707
	do {
I
Ingo Molnar 已提交
5708
		if (!group) {
P
Peter Zijlstra 已提交
5709 5710
			printk("\n");
			printk(KERN_ERR "ERROR: group is NULL\n");
L
Linus Torvalds 已提交
5711 5712 5713
			break;
		}

5714
		if (!cpumask_weight(sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5715 5716
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: empty group\n");
I
Ingo Molnar 已提交
5717 5718
			break;
		}
L
Linus Torvalds 已提交
5719

5720 5721
		if (!(sd->flags & SD_OVERLAP) &&
		    cpumask_intersects(groupmask, sched_group_cpus(group))) {
P
Peter Zijlstra 已提交
5722 5723
			printk(KERN_CONT "\n");
			printk(KERN_ERR "ERROR: repeated CPUs\n");
I
Ingo Molnar 已提交
5724 5725
			break;
		}
L
Linus Torvalds 已提交
5726

5727
		cpumask_or(groupmask, groupmask, sched_group_cpus(group));
L
Linus Torvalds 已提交
5728

5729 5730
		printk(KERN_CONT " %*pbl",
		       cpumask_pr_args(sched_group_cpus(group)));
5731
		if (group->sgc->capacity != SCHED_CAPACITY_SCALE) {
5732 5733
			printk(KERN_CONT " (cpu_capacity = %d)",
				group->sgc->capacity);
5734
		}
L
Linus Torvalds 已提交
5735

I
Ingo Molnar 已提交
5736 5737
		group = group->next;
	} while (group != sd->groups);
P
Peter Zijlstra 已提交
5738
	printk(KERN_CONT "\n");
L
Linus Torvalds 已提交
5739

5740
	if (!cpumask_equal(sched_domain_span(sd), groupmask))
P
Peter Zijlstra 已提交
5741
		printk(KERN_ERR "ERROR: groups don't span domain->span\n");
L
Linus Torvalds 已提交
5742

5743 5744
	if (sd->parent &&
	    !cpumask_subset(groupmask, sched_domain_span(sd->parent)))
P
Peter Zijlstra 已提交
5745 5746
		printk(KERN_ERR "ERROR: parent span is not a superset "
			"of domain->span\n");
I
Ingo Molnar 已提交
5747 5748
	return 0;
}
L
Linus Torvalds 已提交
5749

I
Ingo Molnar 已提交
5750 5751 5752
static void sched_domain_debug(struct sched_domain *sd, int cpu)
{
	int level = 0;
L
Linus Torvalds 已提交
5753

5754
	if (!sched_debug_enabled)
5755 5756
		return;

I
Ingo Molnar 已提交
5757 5758 5759 5760
	if (!sd) {
		printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu);
		return;
	}
L
Linus Torvalds 已提交
5761

I
Ingo Molnar 已提交
5762 5763 5764
	printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu);

	for (;;) {
5765
		if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask))
I
Ingo Molnar 已提交
5766
			break;
L
Linus Torvalds 已提交
5767 5768
		level++;
		sd = sd->parent;
5769
		if (!sd)
I
Ingo Molnar 已提交
5770 5771
			break;
	}
L
Linus Torvalds 已提交
5772
}
5773
#else /* !CONFIG_SCHED_DEBUG */
5774
# define sched_domain_debug(sd, cpu) do { } while (0)
5775 5776 5777 5778
static inline bool sched_debug(void)
{
	return false;
}
5779
#endif /* CONFIG_SCHED_DEBUG */
L
Linus Torvalds 已提交
5780

5781
static int sd_degenerate(struct sched_domain *sd)
5782
{
5783
	if (cpumask_weight(sched_domain_span(sd)) == 1)
5784 5785 5786 5787 5788 5789
		return 1;

	/* Following flags need at least 2 groups */
	if (sd->flags & (SD_LOAD_BALANCE |
			 SD_BALANCE_NEWIDLE |
			 SD_BALANCE_FORK |
5790
			 SD_BALANCE_EXEC |
5791
			 SD_SHARE_CPUCAPACITY |
5792 5793
			 SD_SHARE_PKG_RESOURCES |
			 SD_SHARE_POWERDOMAIN)) {
5794 5795 5796 5797 5798
		if (sd->groups != sd->groups->next)
			return 0;
	}

	/* Following flags don't use groups */
5799
	if (sd->flags & (SD_WAKE_AFFINE))
5800 5801 5802 5803 5804
		return 0;

	return 1;
}

5805 5806
static int
sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
5807 5808 5809 5810 5811 5812
{
	unsigned long cflags = sd->flags, pflags = parent->flags;

	if (sd_degenerate(parent))
		return 1;

5813
	if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent)))
5814 5815 5816 5817 5818 5819 5820
		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 |
5821
				SD_BALANCE_EXEC |
5822
				SD_SHARE_CPUCAPACITY |
5823
				SD_SHARE_PKG_RESOURCES |
5824 5825
				SD_PREFER_SIBLING |
				SD_SHARE_POWERDOMAIN);
5826 5827
		if (nr_node_ids == 1)
			pflags &= ~SD_SERIALIZE;
5828 5829 5830 5831 5832 5833 5834
	}
	if (~cflags & pflags)
		return 0;

	return 1;
}

5835
static void free_rootdomain(struct rcu_head *rcu)
5836
{
5837
	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
5838

5839
	cpupri_cleanup(&rd->cpupri);
5840
	cpudl_cleanup(&rd->cpudl);
5841
	free_cpumask_var(rd->dlo_mask);
5842 5843 5844 5845 5846 5847
	free_cpumask_var(rd->rto_mask);
	free_cpumask_var(rd->online);
	free_cpumask_var(rd->span);
	kfree(rd);
}

G
Gregory Haskins 已提交
5848 5849
static void rq_attach_root(struct rq *rq, struct root_domain *rd)
{
I
Ingo Molnar 已提交
5850
	struct root_domain *old_rd = NULL;
G
Gregory Haskins 已提交
5851 5852
	unsigned long flags;

5853
	raw_spin_lock_irqsave(&rq->lock, flags);
G
Gregory Haskins 已提交
5854 5855

	if (rq->rd) {
I
Ingo Molnar 已提交
5856
		old_rd = rq->rd;
G
Gregory Haskins 已提交
5857

5858
		if (cpumask_test_cpu(rq->cpu, old_rd->online))
5859
			set_rq_offline(rq);
G
Gregory Haskins 已提交
5860

5861
		cpumask_clear_cpu(rq->cpu, old_rd->span);
5862

I
Ingo Molnar 已提交
5863
		/*
5864
		 * If we dont want to free the old_rd yet then
I
Ingo Molnar 已提交
5865 5866 5867 5868 5869
		 * 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
5870 5871 5872 5873 5874
	}

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

5875
	cpumask_set_cpu(rq->cpu, rd->span);
5876
	if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
5877
		set_rq_online(rq);
G
Gregory Haskins 已提交
5878

5879
	raw_spin_unlock_irqrestore(&rq->lock, flags);
I
Ingo Molnar 已提交
5880 5881

	if (old_rd)
5882
		call_rcu_sched(&old_rd->rcu, free_rootdomain);
G
Gregory Haskins 已提交
5883 5884
}

5885
static int init_rootdomain(struct root_domain *rd)
G
Gregory Haskins 已提交
5886 5887 5888
{
	memset(rd, 0, sizeof(*rd));

5889
	if (!zalloc_cpumask_var(&rd->span, GFP_KERNEL))
5890
		goto out;
5891
	if (!zalloc_cpumask_var(&rd->online, GFP_KERNEL))
5892
		goto free_span;
5893
	if (!zalloc_cpumask_var(&rd->dlo_mask, GFP_KERNEL))
5894
		goto free_online;
5895
	if (!zalloc_cpumask_var(&rd->rto_mask, GFP_KERNEL))
5896
		goto free_dlo_mask;
5897

5898
	init_dl_bw(&rd->dl_bw);
5899 5900
	if (cpudl_init(&rd->cpudl) != 0)
		goto free_dlo_mask;
5901

5902
	if (cpupri_init(&rd->cpupri) != 0)
5903
		goto free_rto_mask;
5904
	return 0;
5905

5906 5907
free_rto_mask:
	free_cpumask_var(rd->rto_mask);
5908 5909
free_dlo_mask:
	free_cpumask_var(rd->dlo_mask);
5910 5911 5912 5913
free_online:
	free_cpumask_var(rd->online);
free_span:
	free_cpumask_var(rd->span);
5914
out:
5915
	return -ENOMEM;
G
Gregory Haskins 已提交
5916 5917
}

5918 5919 5920 5921 5922 5923
/*
 * 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
5924 5925
static void init_defrootdomain(void)
{
5926
	init_rootdomain(&def_root_domain);
5927

G
Gregory Haskins 已提交
5928 5929 5930
	atomic_set(&def_root_domain.refcount, 1);
}

5931
static struct root_domain *alloc_rootdomain(void)
G
Gregory Haskins 已提交
5932 5933 5934 5935 5936 5937 5938
{
	struct root_domain *rd;

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

5939
	if (init_rootdomain(rd) != 0) {
5940 5941 5942
		kfree(rd);
		return NULL;
	}
G
Gregory Haskins 已提交
5943 5944 5945 5946

	return rd;
}

5947
static void free_sched_groups(struct sched_group *sg, int free_sgc)
5948 5949 5950 5951 5952 5953 5954 5955 5956 5957
{
	struct sched_group *tmp, *first;

	if (!sg)
		return;

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

5958 5959
		if (free_sgc && atomic_dec_and_test(&sg->sgc->ref))
			kfree(sg->sgc);
5960 5961 5962 5963 5964 5965

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

5966 5967 5968
static void free_sched_domain(struct rcu_head *rcu)
{
	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
5969 5970 5971 5972 5973 5974 5975 5976

	/*
	 * 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)) {
5977
		kfree(sd->groups->sgc);
5978
		kfree(sd->groups);
5979
	}
5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993
	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);
}

5994 5995 5996 5997 5998 5999 6000
/*
 * 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
6001
 * two cpus are in the same cache domain, see cpus_share_cache().
6002 6003
 */
DEFINE_PER_CPU(struct sched_domain *, sd_llc);
6004
DEFINE_PER_CPU(int, sd_llc_size);
6005
DEFINE_PER_CPU(int, sd_llc_id);
6006
DEFINE_PER_CPU(struct sched_domain *, sd_numa);
6007 6008
DEFINE_PER_CPU(struct sched_domain *, sd_busy);
DEFINE_PER_CPU(struct sched_domain *, sd_asym);
6009 6010 6011 6012

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

	sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
6018
	if (sd) {
6019
		id = cpumask_first(sched_domain_span(sd));
6020
		size = cpumask_weight(sched_domain_span(sd));
6021
		busy_sd = sd->parent; /* sd_busy */
6022
	}
6023
	rcu_assign_pointer(per_cpu(sd_busy, cpu), busy_sd);
6024 6025

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

	sd = lowest_flag_domain(cpu, SD_NUMA);
	rcu_assign_pointer(per_cpu(sd_numa, cpu), sd);
6031 6032 6033

	sd = highest_flag_domain(cpu, SD_ASYM_PACKING);
	rcu_assign_pointer(per_cpu(sd_asym, cpu), sd);
6034 6035
}

L
Linus Torvalds 已提交
6036
/*
I
Ingo Molnar 已提交
6037
 * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
L
Linus Torvalds 已提交
6038 6039
 * hold the hotplug lock.
 */
I
Ingo Molnar 已提交
6040 6041
static void
cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu)
L
Linus Torvalds 已提交
6042
{
6043
	struct rq *rq = cpu_rq(cpu);
6044 6045 6046
	struct sched_domain *tmp;

	/* Remove the sched domains which do not contribute to scheduling. */
6047
	for (tmp = sd; tmp; ) {
6048 6049 6050
		struct sched_domain *parent = tmp->parent;
		if (!parent)
			break;
6051

6052
		if (sd_parent_degenerate(tmp, parent)) {
6053
			tmp->parent = parent->parent;
6054 6055
			if (parent->parent)
				parent->parent->child = tmp;
6056 6057 6058 6059 6060 6061 6062
			/*
			 * 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;
6063
			destroy_sched_domain(parent, cpu);
6064 6065
		} else
			tmp = tmp->parent;
6066 6067
	}

6068
	if (sd && sd_degenerate(sd)) {
6069
		tmp = sd;
6070
		sd = sd->parent;
6071
		destroy_sched_domain(tmp, cpu);
6072 6073 6074
		if (sd)
			sd->child = NULL;
	}
L
Linus Torvalds 已提交
6075

6076
	sched_domain_debug(sd, cpu);
L
Linus Torvalds 已提交
6077

G
Gregory Haskins 已提交
6078
	rq_attach_root(rq, rd);
6079
	tmp = rq->sd;
N
Nick Piggin 已提交
6080
	rcu_assign_pointer(rq->sd, sd);
6081
	destroy_sched_domains(tmp, cpu);
6082 6083

	update_top_cache_domain(cpu);
L
Linus Torvalds 已提交
6084 6085 6086 6087 6088
}

/* Setup the mask of cpus configured for isolated domains */
static int __init isolated_cpu_setup(char *str)
{
R
Rusty Russell 已提交
6089
	alloc_bootmem_cpumask_var(&cpu_isolated_map);
R
Rusty Russell 已提交
6090
	cpulist_parse(str, cpu_isolated_map);
L
Linus Torvalds 已提交
6091 6092 6093
	return 1;
}

I
Ingo Molnar 已提交
6094
__setup("isolcpus=", isolated_cpu_setup);
L
Linus Torvalds 已提交
6095

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

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

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

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

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

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

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

	cpumask_clear(covered);

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

		if (cpumask_test_cpu(i, covered))
			continue;

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

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

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

		if (!sg)
			goto fail;

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

		cpumask_or(covered, covered, sg_span);

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

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

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

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

	return 0;

fail:
	free_sched_groups(first, 0);

	return -ENOMEM;
}

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

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

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

	return cpu;
6236 6237
}

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

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

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

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

6263
	cpumask_clear(covered);
6264

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

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

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

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

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

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

	return 0;
6292
}
6293

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

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

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

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

6318 6319
	update_group_capacity(sd, cpu);
	atomic_set(&sg->sgc->nr_busy_cpus, sg->group_weight);
6320 6321
}

6322 6323 6324 6325 6326
/*
 * Initializers for schedule domains
 * Non-inlined to reduce accumulated stack pressure in build_sched_domains()
 */

6327
static int default_relax_domain_level = -1;
6328
int sched_domain_level_max;
6329 6330 6331

static int __init setup_relax_domain_level(char *str)
{
6332 6333
	if (kstrtoint(str, 0, &default_relax_domain_level))
		pr_warn("Unable to set relax_domain_level\n");
6334

6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352
	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 */
6353
		sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6354 6355
	} else {
		/* turn on idle balance on this domain */
6356
		sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE);
6357 6358 6359
	}
}

6360 6361 6362
static void __sdt_free(const struct cpumask *cpu_map);
static int __sdt_alloc(const struct cpumask *cpu_map);

6363 6364 6365 6366 6367
static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
				 const struct cpumask *cpu_map)
{
	switch (what) {
	case sa_rootdomain:
6368 6369
		if (!atomic_read(&d->rd->refcount))
			free_rootdomain(&d->rd->rcu); /* fall through */
6370 6371
	case sa_sd:
		free_percpu(d->sd); /* fall through */
6372
	case sa_sd_storage:
6373
		__sdt_free(cpu_map); /* fall through */
6374 6375 6376 6377
	case sa_none:
		break;
	}
}
6378

6379 6380 6381
static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
						   const struct cpumask *cpu_map)
{
6382 6383
	memset(d, 0, sizeof(*d));

6384 6385
	if (__sdt_alloc(cpu_map))
		return sa_sd_storage;
6386 6387 6388
	d->sd = alloc_percpu(struct sched_domain *);
	if (!d->sd)
		return sa_sd_storage;
6389
	d->rd = alloc_rootdomain();
6390
	if (!d->rd)
6391
		return sa_sd;
6392 6393
	return sa_rootdomain;
}
G
Gregory Haskins 已提交
6394

6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406
/*
 * 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;

6407
	if (atomic_read(&(*per_cpu_ptr(sdd->sg, cpu))->ref))
6408
		*per_cpu_ptr(sdd->sg, cpu) = NULL;
6409

6410 6411
	if (atomic_read(&(*per_cpu_ptr(sdd->sgc, cpu))->ref))
		*per_cpu_ptr(sdd->sgc, cpu) = NULL;
6412 6413
}

6414 6415
#ifdef CONFIG_NUMA
static int sched_domains_numa_levels;
6416
enum numa_topology_type sched_numa_topology_type;
6417
static int *sched_domains_numa_distance;
6418
int sched_max_numa_distance;
6419 6420
static struct cpumask ***sched_domains_numa_masks;
static int sched_domains_curr_level;
6421
#endif
6422

6423 6424 6425
/*
 * SD_flags allowed in topology descriptions.
 *
6426
 * SD_SHARE_CPUCAPACITY      - describes SMT topologies
6427 6428
 * SD_SHARE_PKG_RESOURCES - describes shared caches
 * SD_NUMA                - describes NUMA topologies
6429
 * SD_SHARE_POWERDOMAIN   - describes shared power domain
6430 6431 6432 6433 6434
 *
 * Odd one out:
 * SD_ASYM_PACKING        - describes SMT quirks
 */
#define TOPOLOGY_SD_FLAGS		\
6435
	(SD_SHARE_CPUCAPACITY |		\
6436 6437
	 SD_SHARE_PKG_RESOURCES |	\
	 SD_NUMA |			\
6438 6439
	 SD_ASYM_PACKING |		\
	 SD_SHARE_POWERDOMAIN)
6440 6441

static struct sched_domain *
6442
sd_init(struct sched_domain_topology_level *tl, int cpu)
6443 6444
{
	struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu);
6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460
	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;
6461 6462 6463 6464 6465

	*sd = (struct sched_domain){
		.min_interval		= sd_weight,
		.max_interval		= 2*sd_weight,
		.busy_factor		= 32,
6466
		.imbalance_pct		= 125,
6467 6468 6469 6470

		.cache_nice_tries	= 0,
		.busy_idx		= 0,
		.idle_idx		= 0,
6471 6472 6473 6474 6475 6476
		.newidle_idx		= 0,
		.wake_idx		= 0,
		.forkexec_idx		= 0,

		.flags			= 1*SD_LOAD_BALANCE
					| 1*SD_BALANCE_NEWIDLE
6477 6478
					| 1*SD_BALANCE_EXEC
					| 1*SD_BALANCE_FORK
6479
					| 0*SD_BALANCE_WAKE
6480
					| 1*SD_WAKE_AFFINE
6481
					| 0*SD_SHARE_CPUCAPACITY
6482
					| 0*SD_SHARE_PKG_RESOURCES
6483
					| 0*SD_SERIALIZE
6484
					| 0*SD_PREFER_SIBLING
6485 6486
					| 0*SD_NUMA
					| sd_flags
6487
					,
6488

6489 6490
		.last_balance		= jiffies,
		.balance_interval	= sd_weight,
6491
		.smt_gain		= 0,
6492 6493
		.max_newidle_lb_cost	= 0,
		.next_decay_max_lb_cost	= jiffies,
6494 6495 6496
#ifdef CONFIG_SCHED_DEBUG
		.name			= tl->name,
#endif
6497 6498 6499
	};

	/*
6500
	 * Convert topological properties into behaviour.
6501
	 */
6502

6503
	if (sd->flags & SD_SHARE_CPUCAPACITY) {
6504
		sd->flags |= SD_PREFER_SIBLING;
6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529 6530 6531 6532 6533 6534
		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;
6535 6536 6537 6538

	return sd;
}

6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552
/*
 * 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, },
};

6553 6554
static struct sched_domain_topology_level *sched_domain_topology =
	default_topology;
6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565

#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

6566 6567 6568 6569 6570
static const struct cpumask *sd_numa_mask(int cpu)
{
	return sched_domains_numa_masks[sched_domains_curr_level][cpu_to_node(cpu)];
}

6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591
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");
}

6592
bool find_numa_distance(int distance)
6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606
{
	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;
}

6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631
/*
 * 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;

6632
	if (sched_domains_numa_levels <= 1) {
6633
		sched_numa_topology_type = NUMA_DIRECT;
6634 6635
		return;
	}
6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658

	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;
		}
	}
}

6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679
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++) {
6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703
			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;
6704
		}
6705 6706 6707 6708 6709 6710

		/*
		 * In case of sched_debug() we verify the above assumption.
		 */
		if (!sched_debug())
			break;
6711
	}
6712 6713 6714 6715

	if (!level)
		return;

6716 6717 6718 6719
	/*
	 * 'level' contains the number of unique distances, excluding the
	 * identity distance node_distance(i,i).
	 *
V
Viresh Kumar 已提交
6720
	 * The sched_domains_numa_distance[] array includes the actual distance
6721 6722 6723
	 * numbers.
	 */

6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734
	/*
	 * 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;

6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749
	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++) {
6750
			struct cpumask *mask = kzalloc(cpumask_size(), GFP_KERNEL);
6751 6752 6753 6754 6755 6756
			if (!mask)
				return;

			sched_domains_numa_masks[i][j] = mask;

			for (k = 0; k < nr_node_ids; k++) {
6757
				if (node_distance(j, k) > sched_domains_numa_distance[i])
6758 6759 6760 6761 6762 6763 6764
					continue;

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

6765 6766 6767
	/* Compute default topology size */
	for (i = 0; sched_domain_topology[i].mask; i++);

6768
	tl = kzalloc((i + level + 1) *
6769 6770 6771 6772 6773 6774 6775
			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
	if (!tl)
		return;

	/*
	 * Copy the default topology bits..
	 */
6776 6777
	for (i = 0; sched_domain_topology[i].mask; i++)
		tl[i] = sched_domain_topology[i];
6778 6779 6780 6781 6782 6783 6784

	/*
	 * .. 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,
6785
			.sd_flags = cpu_numa_flags,
6786 6787
			.flags = SDTL_OVERLAP,
			.numa_level = j,
6788
			SD_INIT_NAME(NUMA)
6789 6790 6791 6792
		};
	}

	sched_domain_topology = tl;
6793 6794

	sched_domains_numa_levels = level;
6795
	sched_max_numa_distance = sched_domains_numa_distance[level - 1];
6796 6797

	init_numa_topology_type();
6798
}
6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845

static void sched_domains_numa_masks_set(int cpu)
{
	int i, j;
	int node = cpu_to_node(cpu);

	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]);
		}
	}
}

static void sched_domains_numa_masks_clear(int cpu)
{
	int i, j;
	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]);
	}
}

/*
 * Update sched_domains_numa_masks[level][node] array when new cpus
 * are onlined.
 */
static int sched_domains_numa_masks_update(struct notifier_block *nfb,
					   unsigned long action,
					   void *hcpu)
{
	int cpu = (long)hcpu;

	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_ONLINE:
		sched_domains_numa_masks_set(cpu);
		break;

	case CPU_DEAD:
		sched_domains_numa_masks_clear(cpu);
		break;

	default:
		return NOTIFY_DONE;
	}

	return NOTIFY_OK;
6846 6847 6848 6849 6850
}
#else
static inline void sched_init_numa(void)
{
}
6851 6852 6853 6854 6855 6856 6857

static int sched_domains_numa_masks_update(struct notifier_block *nfb,
					   unsigned long action,
					   void *hcpu)
{
	return 0;
}
6858 6859
#endif /* CONFIG_NUMA */

6860 6861 6862 6863 6864
static int __sdt_alloc(const struct cpumask *cpu_map)
{
	struct sched_domain_topology_level *tl;
	int j;

6865
	for_each_sd_topology(tl) {
6866 6867 6868 6869 6870 6871 6872 6873 6874 6875
		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;

6876 6877
		sdd->sgc = alloc_percpu(struct sched_group_capacity *);
		if (!sdd->sgc)
6878 6879
			return -ENOMEM;

6880 6881 6882
		for_each_cpu(j, cpu_map) {
			struct sched_domain *sd;
			struct sched_group *sg;
6883
			struct sched_group_capacity *sgc;
6884

P
Peter Zijlstra 已提交
6885
			sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896
					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;

6897 6898
			sg->next = sg;

6899
			*per_cpu_ptr(sdd->sg, j) = sg;
6900

6901
			sgc = kzalloc_node(sizeof(struct sched_group_capacity) + cpumask_size(),
6902
					GFP_KERNEL, cpu_to_node(j));
6903
			if (!sgc)
6904 6905
				return -ENOMEM;

6906
			*per_cpu_ptr(sdd->sgc, j) = sgc;
6907 6908 6909 6910 6911 6912 6913 6914 6915 6916 6917
		}
	}

	return 0;
}

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

6918
	for_each_sd_topology(tl) {
6919 6920 6921
		struct sd_data *sdd = &tl->data;

		for_each_cpu(j, cpu_map) {
6922 6923 6924 6925 6926 6927 6928 6929 6930 6931 6932
			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));
6933 6934
			if (sdd->sgc)
				kfree(*per_cpu_ptr(sdd->sgc, j));
6935 6936
		}
		free_percpu(sdd->sd);
6937
		sdd->sd = NULL;
6938
		free_percpu(sdd->sg);
6939
		sdd->sg = NULL;
6940 6941
		free_percpu(sdd->sgc);
		sdd->sgc = NULL;
6942 6943 6944
	}
}

6945
struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl,
6946 6947
		const struct cpumask *cpu_map, struct sched_domain_attr *attr,
		struct sched_domain *child, int cpu)
6948
{
6949
	struct sched_domain *sd = sd_init(tl, cpu);
6950
	if (!sd)
6951
		return child;
6952 6953

	cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu));
6954 6955 6956
	if (child) {
		sd->level = child->level + 1;
		sched_domain_level_max = max(sched_domain_level_max, sd->level);
6957
		child->parent = sd;
6958
		sd->child = child;
P
Peter Zijlstra 已提交
6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972

		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));
		}

6973
	}
6974
	set_domain_attribute(sd, attr);
6975 6976 6977 6978

	return sd;
}

6979 6980 6981 6982
/*
 * Build sched domains for a given set of cpus and attach the sched domains
 * to the individual cpus
 */
6983 6984
static int build_sched_domains(const struct cpumask *cpu_map,
			       struct sched_domain_attr *attr)
6985
{
6986
	enum s_alloc alloc_state;
6987
	struct sched_domain *sd;
6988
	struct s_data d;
6989
	int i, ret = -ENOMEM;
6990

6991 6992 6993
	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
	if (alloc_state != sa_rootdomain)
		goto error;
6994

6995
	/* Set up domains for cpus specified by the cpu_map. */
6996
	for_each_cpu(i, cpu_map) {
6997 6998
		struct sched_domain_topology_level *tl;

6999
		sd = NULL;
7000
		for_each_sd_topology(tl) {
7001
			sd = build_sched_domain(tl, cpu_map, attr, sd, i);
7002 7003
			if (tl == sched_domain_topology)
				*per_cpu_ptr(d.sd, i) = sd;
7004 7005
			if (tl->flags & SDTL_OVERLAP || sched_feat(FORCE_SD_OVERLAP))
				sd->flags |= SD_OVERLAP;
7006 7007
			if (cpumask_equal(cpu_map, sched_domain_span(sd)))
				break;
7008
		}
7009 7010 7011 7012 7013 7014
	}

	/* 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));
7015 7016 7017 7018 7019 7020 7021
			if (sd->flags & SD_OVERLAP) {
				if (build_overlap_sched_groups(sd, i))
					goto error;
			} else {
				if (build_sched_groups(sd, i))
					goto error;
			}
7022
		}
7023
	}
7024

7025
	/* Calculate CPU capacity for physical packages and nodes */
7026 7027 7028
	for (i = nr_cpumask_bits-1; i >= 0; i--) {
		if (!cpumask_test_cpu(i, cpu_map))
			continue;
7029

7030 7031
		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
			claim_allocations(i, sd);
7032
			init_sched_groups_capacity(i, sd);
7033
		}
7034
	}
7035

L
Linus Torvalds 已提交
7036
	/* Attach the domains */
7037
	rcu_read_lock();
7038
	for_each_cpu(i, cpu_map) {
7039
		sd = *per_cpu_ptr(d.sd, i);
7040
		cpu_attach_domain(sd, d.rd, i);
L
Linus Torvalds 已提交
7041
	}
7042
	rcu_read_unlock();
7043

7044
	ret = 0;
7045
error:
7046
	__free_domain_allocs(&d, alloc_state, cpu_map);
7047
	return ret;
L
Linus Torvalds 已提交
7048
}
P
Paul Jackson 已提交
7049

7050
static cpumask_var_t *doms_cur;	/* current sched domains */
P
Paul Jackson 已提交
7051
static int ndoms_cur;		/* number of sched domains in 'doms_cur' */
I
Ingo Molnar 已提交
7052 7053
static struct sched_domain_attr *dattr_cur;
				/* attribues of custom domains in 'doms_cur' */
P
Paul Jackson 已提交
7054 7055 7056

/*
 * Special case: If a kmalloc of a doms_cur partition (array of
7057 7058
 * cpumask) fails, then fallback to a single sched domain,
 * as determined by the single cpumask fallback_doms.
P
Paul Jackson 已提交
7059
 */
7060
static cpumask_var_t fallback_doms;
P
Paul Jackson 已提交
7061

7062 7063 7064 7065 7066
/*
 * 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.
 */
7067
int __weak arch_update_cpu_topology(void)
7068
{
7069
	return 0;
7070 7071
}

7072 7073 7074 7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095 7096
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);
}

7097
/*
I
Ingo Molnar 已提交
7098
 * Set up scheduler domains and groups. Callers must hold the hotplug lock.
P
Paul Jackson 已提交
7099 7100
 * For now this just excludes isolated cpus, but could be used to
 * exclude other special cases in the future.
7101
 */
7102
static int init_sched_domains(const struct cpumask *cpu_map)
7103
{
7104 7105
	int err;

7106
	arch_update_cpu_topology();
P
Paul Jackson 已提交
7107
	ndoms_cur = 1;
7108
	doms_cur = alloc_sched_domains(ndoms_cur);
P
Paul Jackson 已提交
7109
	if (!doms_cur)
7110 7111
		doms_cur = &fallback_doms;
	cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map);
7112
	err = build_sched_domains(doms_cur[0], NULL);
7113
	register_sched_domain_sysctl();
7114 7115

	return err;
7116 7117 7118 7119 7120 7121
}

/*
 * Detach sched domains from a group of cpus specified in cpu_map
 * These cpus will now be attached to the NULL domain
 */
7122
static void detach_destroy_domains(const struct cpumask *cpu_map)
7123 7124 7125
{
	int i;

7126
	rcu_read_lock();
7127
	for_each_cpu(i, cpu_map)
G
Gregory Haskins 已提交
7128
		cpu_attach_domain(NULL, &def_root_domain, i);
7129
	rcu_read_unlock();
7130 7131
}

7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147
/* 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
7148 7149
/*
 * Partition sched domains as specified by the 'ndoms_new'
I
Ingo Molnar 已提交
7150
 * cpumasks in the array doms_new[] of cpumasks. This compares
P
Paul Jackson 已提交
7151 7152 7153
 * doms_new[] to the current sched domain partitioning, doms_cur[].
 * It destroys each deleted domain and builds each new domain.
 *
7154
 * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'.
I
Ingo Molnar 已提交
7155 7156 7157
 * 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
7158 7159 7160
 * current 'doms_cur' domains and in the new 'doms_new', we can leave
 * it as it is.
 *
7161 7162 7163 7164 7165 7166
 * 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
7167
 *
7168
 * If doms_new == NULL it will be replaced with cpu_online_mask.
7169 7170
 * ndoms_new == 0 is a special case for destroying existing domains,
 * and it will not create the default domain.
7171
 *
P
Paul Jackson 已提交
7172 7173
 * Call with hotplug lock held
 */
7174
void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[],
7175
			     struct sched_domain_attr *dattr_new)
P
Paul Jackson 已提交
7176
{
7177
	int i, j, n;
7178
	int new_topology;
P
Paul Jackson 已提交
7179

7180
	mutex_lock(&sched_domains_mutex);
7181

7182 7183 7184
	/* always unregister in case we don't destroy any domains */
	unregister_sched_domain_sysctl();

7185 7186 7187
	/* Let architecture update cpu core mappings. */
	new_topology = arch_update_cpu_topology();

7188
	n = doms_new ? ndoms_new : 0;
P
Paul Jackson 已提交
7189 7190 7191

	/* Destroy deleted domains */
	for (i = 0; i < ndoms_cur; i++) {
7192
		for (j = 0; j < n && !new_topology; j++) {
7193
			if (cpumask_equal(doms_cur[i], doms_new[j])
7194
			    && dattrs_equal(dattr_cur, i, dattr_new, j))
P
Paul Jackson 已提交
7195 7196 7197
				goto match1;
		}
		/* no match - a current sched domain not in new doms_new[] */
7198
		detach_destroy_domains(doms_cur[i]);
P
Paul Jackson 已提交
7199 7200 7201 7202
match1:
		;
	}

7203
	n = ndoms_cur;
7204
	if (doms_new == NULL) {
7205
		n = 0;
7206
		doms_new = &fallback_doms;
7207
		cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map);
7208
		WARN_ON_ONCE(dattr_new);
7209 7210
	}

P
Paul Jackson 已提交
7211 7212
	/* Build new domains */
	for (i = 0; i < ndoms_new; i++) {
7213
		for (j = 0; j < n && !new_topology; j++) {
7214
			if (cpumask_equal(doms_new[i], doms_cur[j])
7215
			    && dattrs_equal(dattr_new, i, dattr_cur, j))
P
Paul Jackson 已提交
7216 7217 7218
				goto match2;
		}
		/* no match - add a new doms_new */
7219
		build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL);
P
Paul Jackson 已提交
7220 7221 7222 7223 7224
match2:
		;
	}

	/* Remember the new sched domains */
7225 7226
	if (doms_cur != &fallback_doms)
		free_sched_domains(doms_cur, ndoms_cur);
7227
	kfree(dattr_cur);	/* kfree(NULL) is safe */
P
Paul Jackson 已提交
7228
	doms_cur = doms_new;
7229
	dattr_cur = dattr_new;
P
Paul Jackson 已提交
7230
	ndoms_cur = ndoms_new;
7231 7232

	register_sched_domain_sysctl();
7233

7234
	mutex_unlock(&sched_domains_mutex);
P
Paul Jackson 已提交
7235 7236
}

7237 7238
static int num_cpus_frozen;	/* used to mark begin/end of suspend/resume */

L
Linus Torvalds 已提交
7239
/*
7240 7241 7242
 * Update cpusets according to cpu_active mask.  If cpusets are
 * disabled, cpuset_update_active_cpus() becomes a simple wrapper
 * around partition_sched_domains().
7243 7244 7245
 *
 * 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
7246
 */
7247 7248
static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action,
			     void *hcpu)
7249
{
7250 7251 7252 7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271
	switch (action) {
	case CPU_ONLINE_FROZEN:
	case CPU_DOWN_FAILED_FROZEN:

		/*
		 * 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);
			break;
		}

		/*
		 * This is the last CPU online operation. So fall through and
		 * restore the original sched domains by considering the
		 * cpuset configurations.
		 */

7272
	case CPU_ONLINE:
7273
		cpuset_update_active_cpus(true);
7274
		break;
7275 7276 7277
	default:
		return NOTIFY_DONE;
	}
7278
	return NOTIFY_OK;
7279
}
7280

7281 7282
static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action,
			       void *hcpu)
7283
{
7284 7285 7286
	unsigned long flags;
	long cpu = (long)hcpu;
	struct dl_bw *dl_b;
7287 7288
	bool overflow;
	int cpus;
7289

7290
	switch (action) {
7291
	case CPU_DOWN_PREPARE:
7292 7293
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
7294

7295 7296 7297 7298
		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);
7299

7300
		rcu_read_unlock_sched();
7301

7302 7303
		if (overflow)
			return notifier_from_errno(-EBUSY);
7304
		cpuset_update_active_cpus(false);
7305 7306 7307 7308 7309
		break;
	case CPU_DOWN_PREPARE_FROZEN:
		num_cpus_frozen++;
		partition_sched_domains(1, NULL, NULL);
		break;
7310 7311 7312
	default:
		return NOTIFY_DONE;
	}
7313
	return NOTIFY_OK;
7314 7315
}

L
Linus Torvalds 已提交
7316 7317
void __init sched_init_smp(void)
{
7318 7319 7320
	cpumask_var_t non_isolated_cpus;

	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
7321
	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
7322

7323 7324
	sched_init_numa();

7325 7326 7327 7328 7329
	/*
	 * 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.
	 */
7330
	mutex_lock(&sched_domains_mutex);
7331
	init_sched_domains(cpu_active_mask);
7332 7333 7334
	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);
7335
	mutex_unlock(&sched_domains_mutex);
7336

7337
	hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE);
7338 7339
	hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
	hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
7340

7341
	init_hrtick();
7342 7343

	/* Move init over to a non-isolated CPU */
7344
	if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0)
7345
		BUG();
I
Ingo Molnar 已提交
7346
	sched_init_granularity();
7347
	free_cpumask_var(non_isolated_cpus);
7348

7349
	init_sched_rt_class();
7350
	init_sched_dl_class();
L
Linus Torvalds 已提交
7351 7352 7353 7354
}
#else
void __init sched_init_smp(void)
{
I
Ingo Molnar 已提交
7355
	sched_init_granularity();
L
Linus Torvalds 已提交
7356 7357 7358 7359 7360 7361 7362 7363 7364 7365
}
#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);
}

7366
#ifdef CONFIG_CGROUP_SCHED
7367 7368 7369 7370
/*
 * Default task group.
 * Every task in system belongs to this group at bootup.
 */
7371
struct task_group root_task_group;
7372
LIST_HEAD(task_groups);
7373 7374 7375

/* Cacheline aligned slab cache for task_group */
static struct kmem_cache *task_group_cache __read_mostly;
7376
#endif
P
Peter Zijlstra 已提交
7377

7378
DECLARE_PER_CPU(cpumask_var_t, load_balance_mask);
P
Peter Zijlstra 已提交
7379

L
Linus Torvalds 已提交
7380 7381
void __init sched_init(void)
{
I
Ingo Molnar 已提交
7382
	int i, j;
7383 7384 7385 7386 7387 7388 7389 7390 7391
	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) {
7392
		ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT);
7393 7394

#ifdef CONFIG_FAIR_GROUP_SCHED
7395
		root_task_group.se = (struct sched_entity **)ptr;
7396 7397
		ptr += nr_cpu_ids * sizeof(void **);

7398
		root_task_group.cfs_rq = (struct cfs_rq **)ptr;
7399
		ptr += nr_cpu_ids * sizeof(void **);
7400

7401
#endif /* CONFIG_FAIR_GROUP_SCHED */
7402
#ifdef CONFIG_RT_GROUP_SCHED
7403
		root_task_group.rt_se = (struct sched_rt_entity **)ptr;
7404 7405
		ptr += nr_cpu_ids * sizeof(void **);

7406
		root_task_group.rt_rq = (struct rt_rq **)ptr;
7407 7408
		ptr += nr_cpu_ids * sizeof(void **);

7409
#endif /* CONFIG_RT_GROUP_SCHED */
7410
	}
7411
#ifdef CONFIG_CPUMASK_OFFSTACK
7412 7413 7414
	for_each_possible_cpu(i) {
		per_cpu(load_balance_mask, i) = (cpumask_var_t)kzalloc_node(
			cpumask_size(), GFP_KERNEL, cpu_to_node(i));
7415
	}
7416
#endif /* CONFIG_CPUMASK_OFFSTACK */
I
Ingo Molnar 已提交
7417

7418 7419 7420
	init_rt_bandwidth(&def_rt_bandwidth,
			global_rt_period(), global_rt_runtime());
	init_dl_bandwidth(&def_dl_bandwidth,
7421
			global_rt_period(), global_rt_runtime());
7422

G
Gregory Haskins 已提交
7423 7424 7425 7426
#ifdef CONFIG_SMP
	init_defrootdomain();
#endif

7427
#ifdef CONFIG_RT_GROUP_SCHED
7428
	init_rt_bandwidth(&root_task_group.rt_bandwidth,
7429
			global_rt_period(), global_rt_runtime());
7430
#endif /* CONFIG_RT_GROUP_SCHED */
7431

D
Dhaval Giani 已提交
7432
#ifdef CONFIG_CGROUP_SCHED
7433 7434
	task_group_cache = KMEM_CACHE(task_group, 0);

7435 7436
	list_add(&root_task_group.list, &task_groups);
	INIT_LIST_HEAD(&root_task_group.children);
7437
	INIT_LIST_HEAD(&root_task_group.siblings);
7438
	autogroup_init(&init_task);
D
Dhaval Giani 已提交
7439
#endif /* CONFIG_CGROUP_SCHED */
P
Peter Zijlstra 已提交
7440

7441
	for_each_possible_cpu(i) {
7442
		struct rq *rq;
L
Linus Torvalds 已提交
7443 7444

		rq = cpu_rq(i);
7445
		raw_spin_lock_init(&rq->lock);
N
Nick Piggin 已提交
7446
		rq->nr_running = 0;
7447 7448
		rq->calc_load_active = 0;
		rq->calc_load_update = jiffies + LOAD_FREQ;
7449
		init_cfs_rq(&rq->cfs);
7450 7451
		init_rt_rq(&rq->rt);
		init_dl_rq(&rq->dl);
I
Ingo Molnar 已提交
7452
#ifdef CONFIG_FAIR_GROUP_SCHED
7453
		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
P
Peter Zijlstra 已提交
7454
		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
D
Dhaval Giani 已提交
7455
		/*
7456
		 * How much cpu bandwidth does root_task_group get?
D
Dhaval Giani 已提交
7457 7458 7459 7460
		 *
		 * 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
7461
		 * root_task_group and its child task-groups in a fair manner,
D
Dhaval Giani 已提交
7462 7463 7464
		 * based on each entity's (task or task-group's) weight
		 * (se->load.weight).
		 *
7465
		 * In other words, if root_task_group has 10 tasks of weight
D
Dhaval Giani 已提交
7466 7467 7468
		 * 1024) and two child groups A0 and A1 (of weight 1024 each),
		 * then A0's share of the cpu resource is:
		 *
7469
		 *	A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33%
D
Dhaval Giani 已提交
7470
		 *
7471 7472
		 * 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
7473
		 */
7474
		init_cfs_bandwidth(&root_task_group.cfs_bandwidth);
7475
		init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL);
D
Dhaval Giani 已提交
7476 7477 7478
#endif /* CONFIG_FAIR_GROUP_SCHED */

		rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime;
7479
#ifdef CONFIG_RT_GROUP_SCHED
7480
		init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL);
I
Ingo Molnar 已提交
7481
#endif
L
Linus Torvalds 已提交
7482

I
Ingo Molnar 已提交
7483 7484
		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
			rq->cpu_load[j] = 0;
7485 7486 7487

		rq->last_load_update_tick = jiffies;

L
Linus Torvalds 已提交
7488
#ifdef CONFIG_SMP
N
Nick Piggin 已提交
7489
		rq->sd = NULL;
G
Gregory Haskins 已提交
7490
		rq->rd = NULL;
7491
		rq->cpu_capacity = rq->cpu_capacity_orig = SCHED_CAPACITY_SCALE;
7492
		rq->balance_callback = NULL;
L
Linus Torvalds 已提交
7493
		rq->active_balance = 0;
I
Ingo Molnar 已提交
7494
		rq->next_balance = jiffies;
L
Linus Torvalds 已提交
7495
		rq->push_cpu = 0;
7496
		rq->cpu = i;
7497
		rq->online = 0;
7498 7499
		rq->idle_stamp = 0;
		rq->avg_idle = 2*sysctl_sched_migration_cost;
7500
		rq->max_idle_balance_cost = sysctl_sched_migration_cost;
7501 7502 7503

		INIT_LIST_HEAD(&rq->cfs_tasks);

7504
		rq_attach_root(rq, &def_root_domain);
7505
#ifdef CONFIG_NO_HZ_COMMON
7506
		rq->nohz_flags = 0;
7507
#endif
7508 7509 7510
#ifdef CONFIG_NO_HZ_FULL
		rq->last_sched_tick = 0;
#endif
L
Linus Torvalds 已提交
7511
#endif
P
Peter Zijlstra 已提交
7512
		init_rq_hrtick(rq);
L
Linus Torvalds 已提交
7513 7514 7515
		atomic_set(&rq->nr_iowait, 0);
	}

7516
	set_load_weight(&init_task);
7517

7518 7519 7520 7521
#ifdef CONFIG_PREEMPT_NOTIFIERS
	INIT_HLIST_HEAD(&init_task.preempt_notifiers);
#endif

L
Linus Torvalds 已提交
7522 7523 7524 7525 7526 7527
	/*
	 * The boot idle thread does lazy MMU switching as well:
	 */
	atomic_inc(&init_mm.mm_count);
	enter_lazy_tlb(&init_mm, current);

7528 7529 7530 7531 7532
	/*
	 * During early bootup we pretend to be a normal task:
	 */
	current->sched_class = &fair_sched_class;

L
Linus Torvalds 已提交
7533 7534 7535 7536 7537 7538 7539
	/*
	 * 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());
7540 7541 7542

	calc_load_update = jiffies + LOAD_FREQ;

7543
#ifdef CONFIG_SMP
7544
	zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT);
R
Rusty Russell 已提交
7545 7546 7547
	/* May be allocated at isolcpus cmdline parse time */
	if (cpu_isolated_map == NULL)
		zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT);
7548
	idle_thread_set_boot_cpu();
7549
	set_cpu_rq_start_time();
7550 7551
#endif
	init_sched_fair_class();
7552

7553
	scheduler_running = 1;
L
Linus Torvalds 已提交
7554 7555
}

7556
#ifdef CONFIG_DEBUG_ATOMIC_SLEEP
7557 7558
static inline int preempt_count_equals(int preempt_offset)
{
7559
	int nested = preempt_count() + rcu_preempt_depth();
7560

A
Arnd Bergmann 已提交
7561
	return (nested == preempt_offset);
7562 7563
}

7564
void __might_sleep(const char *file, int line, int preempt_offset)
L
Linus Torvalds 已提交
7565
{
P
Peter Zijlstra 已提交
7566 7567 7568 7569 7570
	/*
	 * 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.
	 */
7571
	WARN_ONCE(current->state != TASK_RUNNING && current->task_state_change,
P
Peter Zijlstra 已提交
7572 7573 7574 7575
			"do not call blocking ops when !TASK_RUNNING; "
			"state=%lx set at [<%p>] %pS\n",
			current->state,
			(void *)current->task_state_change,
7576
			(void *)current->task_state_change);
P
Peter Zijlstra 已提交
7577

7578 7579 7580 7581 7582
	___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
7583 7584 7585
{
	static unsigned long prev_jiffy;	/* ratelimiting */

7586
	rcu_sleep_check(); /* WARN_ON_ONCE() by default, no rate limit reqd. */
7587 7588
	if ((preempt_count_equals(preempt_offset) && !irqs_disabled() &&
	     !is_idle_task(current)) ||
7589
	    system_state != SYSTEM_RUNNING || oops_in_progress)
I
Ingo Molnar 已提交
7590 7591 7592 7593 7594
		return;
	if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy)
		return;
	prev_jiffy = jiffies;

P
Peter Zijlstra 已提交
7595 7596 7597 7598 7599 7600 7601
	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
7602

7603 7604 7605
	if (task_stack_end_corrupted(current))
		printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");

I
Ingo Molnar 已提交
7606 7607 7608
	debug_show_held_locks(current);
	if (irqs_disabled())
		print_irqtrace_events(current);
7609 7610 7611 7612 7613 7614 7615
#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
7616
	dump_stack();
L
Linus Torvalds 已提交
7617
}
7618
EXPORT_SYMBOL(___might_sleep);
L
Linus Torvalds 已提交
7619 7620 7621
#endif

#ifdef CONFIG_MAGIC_SYSRQ
7622
void normalize_rt_tasks(void)
7623
{
7624
	struct task_struct *g, *p;
7625 7626 7627
	struct sched_attr attr = {
		.sched_policy = SCHED_NORMAL,
	};
L
Linus Torvalds 已提交
7628

7629
	read_lock(&tasklist_lock);
7630
	for_each_process_thread(g, p) {
7631 7632 7633
		/*
		 * Only normalize user tasks:
		 */
7634
		if (p->flags & PF_KTHREAD)
7635 7636
			continue;

I
Ingo Molnar 已提交
7637 7638
		p->se.exec_start		= 0;
#ifdef CONFIG_SCHEDSTATS
7639 7640 7641
		p->se.statistics.wait_start	= 0;
		p->se.statistics.sleep_start	= 0;
		p->se.statistics.block_start	= 0;
I
Ingo Molnar 已提交
7642
#endif
I
Ingo Molnar 已提交
7643

7644
		if (!dl_task(p) && !rt_task(p)) {
I
Ingo Molnar 已提交
7645 7646 7647 7648
			/*
			 * Renice negative nice level userspace
			 * tasks back to 0:
			 */
7649
			if (task_nice(p) < 0)
I
Ingo Molnar 已提交
7650
				set_user_nice(p, 0);
L
Linus Torvalds 已提交
7651
			continue;
I
Ingo Molnar 已提交
7652
		}
L
Linus Torvalds 已提交
7653

7654
		__sched_setscheduler(p, &attr, false, false);
7655
	}
7656
	read_unlock(&tasklist_lock);
L
Linus Torvalds 已提交
7657 7658 7659
}

#endif /* CONFIG_MAGIC_SYSRQ */
7660

7661
#if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB)
7662
/*
7663
 * These functions are only useful for the IA64 MCA handling, or kdb.
7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674 7675 7676
 *
 * 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!
7677 7678
 *
 * Return: The current task for @cpu.
7679
 */
7680
struct task_struct *curr_task(int cpu)
7681 7682 7683 7684
{
	return cpu_curr(cpu);
}

7685 7686 7687
#endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */

#ifdef CONFIG_IA64
7688 7689 7690 7691 7692 7693
/**
 * 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
7694 7695
 * 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
7696 7697 7698 7699 7700 7701 7702
 * 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!
 */
7703
void set_curr_task(int cpu, struct task_struct *p)
7704 7705 7706 7707 7708
{
	cpu_curr(cpu) = p;
}

#endif
S
Srivatsa Vaddagiri 已提交
7709

D
Dhaval Giani 已提交
7710
#ifdef CONFIG_CGROUP_SCHED
7711 7712 7713
/* task_group_lock serializes the addition/removal of task groups */
static DEFINE_SPINLOCK(task_group_lock);

7714 7715 7716 7717
static void free_sched_group(struct task_group *tg)
{
	free_fair_sched_group(tg);
	free_rt_sched_group(tg);
7718
	autogroup_free(tg);
7719
	kmem_cache_free(task_group_cache, tg);
7720 7721 7722
}

/* allocate runqueue etc for a new task group */
7723
struct task_group *sched_create_group(struct task_group *parent)
7724 7725 7726
{
	struct task_group *tg;

7727
	tg = kmem_cache_alloc(task_group_cache, GFP_KERNEL | __GFP_ZERO);
7728 7729 7730
	if (!tg)
		return ERR_PTR(-ENOMEM);

7731
	if (!alloc_fair_sched_group(tg, parent))
7732 7733
		goto err;

7734
	if (!alloc_rt_sched_group(tg, parent))
7735 7736
		goto err;

7737 7738 7739 7740 7741 7742 7743 7744 7745 7746 7747
	return tg;

err:
	free_sched_group(tg);
	return ERR_PTR(-ENOMEM);
}

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

7748
	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7749
	list_add_rcu(&tg->list, &task_groups);
P
Peter Zijlstra 已提交
7750 7751 7752 7753 7754

	WARN_ON(!parent); /* root should already exist */

	tg->parent = parent;
	INIT_LIST_HEAD(&tg->children);
7755
	list_add_rcu(&tg->siblings, &parent->children);
7756
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7757 7758
}

7759
/* rcu callback to free various structures associated with a task group */
P
Peter Zijlstra 已提交
7760
static void free_sched_group_rcu(struct rcu_head *rhp)
S
Srivatsa Vaddagiri 已提交
7761 7762
{
	/* now it should be safe to free those cfs_rqs */
P
Peter Zijlstra 已提交
7763
	free_sched_group(container_of(rhp, struct task_group, rcu));
S
Srivatsa Vaddagiri 已提交
7764 7765
}

7766
/* Destroy runqueue etc associated with a task group */
7767
void sched_destroy_group(struct task_group *tg)
7768 7769 7770 7771 7772 7773
{
	/* wait for possible concurrent references to cfs_rqs complete */
	call_rcu(&tg->rcu, free_sched_group_rcu);
}

void sched_offline_group(struct task_group *tg)
S
Srivatsa Vaddagiri 已提交
7774
{
7775
	unsigned long flags;
7776
	int i;
S
Srivatsa Vaddagiri 已提交
7777

7778 7779
	/* end participation in shares distribution */
	for_each_possible_cpu(i)
7780
		unregister_fair_sched_group(tg, i);
7781 7782

	spin_lock_irqsave(&task_group_lock, flags);
P
Peter Zijlstra 已提交
7783
	list_del_rcu(&tg->list);
P
Peter Zijlstra 已提交
7784
	list_del_rcu(&tg->siblings);
7785
	spin_unlock_irqrestore(&task_group_lock, flags);
S
Srivatsa Vaddagiri 已提交
7786 7787
}

7788
/* change task's runqueue when it moves between groups.
I
Ingo Molnar 已提交
7789 7790 7791
 *	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.
7792 7793
 */
void sched_move_task(struct task_struct *tsk)
S
Srivatsa Vaddagiri 已提交
7794
{
P
Peter Zijlstra 已提交
7795
	struct task_group *tg;
7796
	int queued, running;
S
Srivatsa Vaddagiri 已提交
7797 7798 7799 7800 7801
	unsigned long flags;
	struct rq *rq;

	rq = task_rq_lock(tsk, &flags);

7802
	running = task_current(rq, tsk);
7803
	queued = task_on_rq_queued(tsk);
S
Srivatsa Vaddagiri 已提交
7804

7805
	if (queued)
7806
		dequeue_task(rq, tsk, DEQUEUE_SAVE);
7807
	if (unlikely(running))
7808
		put_prev_task(rq, tsk);
S
Srivatsa Vaddagiri 已提交
7809

7810 7811 7812 7813 7814 7815
	/*
	 * 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
7816 7817 7818 7819
			  struct task_group, css);
	tg = autogroup_task_group(tsk, tg);
	tsk->sched_task_group = tg;

P
Peter Zijlstra 已提交
7820
#ifdef CONFIG_FAIR_GROUP_SCHED
7821
	if (tsk->sched_class->task_move_group)
7822
		tsk->sched_class->task_move_group(tsk);
7823
	else
P
Peter Zijlstra 已提交
7824
#endif
7825
		set_task_rq(tsk, task_cpu(tsk));
P
Peter Zijlstra 已提交
7826

7827 7828
	if (unlikely(running))
		tsk->sched_class->set_curr_task(rq);
7829
	if (queued)
7830
		enqueue_task(rq, tsk, ENQUEUE_RESTORE);
S
Srivatsa Vaddagiri 已提交
7831

7832
	task_rq_unlock(rq, tsk, &flags);
S
Srivatsa Vaddagiri 已提交
7833
}
D
Dhaval Giani 已提交
7834
#endif /* CONFIG_CGROUP_SCHED */
S
Srivatsa Vaddagiri 已提交
7835

7836 7837 7838 7839 7840
#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
7841

P
Peter Zijlstra 已提交
7842 7843
/* Must be called with tasklist_lock held */
static inline int tg_has_rt_tasks(struct task_group *tg)
7844
{
P
Peter Zijlstra 已提交
7845
	struct task_struct *g, *p;
7846

7847 7848 7849 7850 7851 7852
	/*
	 * Autogroups do not have RT tasks; see autogroup_create().
	 */
	if (task_group_is_autogroup(tg))
		return 0;

7853
	for_each_process_thread(g, p) {
7854
		if (rt_task(p) && task_group(p) == tg)
P
Peter Zijlstra 已提交
7855
			return 1;
7856
	}
7857

P
Peter Zijlstra 已提交
7858 7859
	return 0;
}
7860

P
Peter Zijlstra 已提交
7861 7862 7863 7864 7865
struct rt_schedulable_data {
	struct task_group *tg;
	u64 rt_period;
	u64 rt_runtime;
};
7866

7867
static int tg_rt_schedulable(struct task_group *tg, void *data)
P
Peter Zijlstra 已提交
7868 7869 7870 7871 7872
{
	struct rt_schedulable_data *d = data;
	struct task_group *child;
	unsigned long total, sum = 0;
	u64 period, runtime;
7873

P
Peter Zijlstra 已提交
7874 7875
	period = ktime_to_ns(tg->rt_bandwidth.rt_period);
	runtime = tg->rt_bandwidth.rt_runtime;
7876

P
Peter Zijlstra 已提交
7877 7878 7879
	if (tg == d->tg) {
		period = d->rt_period;
		runtime = d->rt_runtime;
7880 7881
	}

7882 7883 7884 7885 7886
	/*
	 * Cannot have more runtime than the period.
	 */
	if (runtime > period && runtime != RUNTIME_INF)
		return -EINVAL;
P
Peter Zijlstra 已提交
7887

7888 7889 7890
	/*
	 * Ensure we don't starve existing RT tasks.
	 */
P
Peter Zijlstra 已提交
7891 7892
	if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg))
		return -EBUSY;
P
Peter Zijlstra 已提交
7893

P
Peter Zijlstra 已提交
7894
	total = to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7895

7896 7897 7898 7899 7900
	/*
	 * 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
7901

7902 7903 7904
	/*
	 * The sum of our children's runtime should not exceed our own.
	 */
P
Peter Zijlstra 已提交
7905 7906 7907
	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
7908

P
Peter Zijlstra 已提交
7909 7910 7911 7912
		if (child == d->tg) {
			period = d->rt_period;
			runtime = d->rt_runtime;
		}
P
Peter Zijlstra 已提交
7913

P
Peter Zijlstra 已提交
7914
		sum += to_ratio(period, runtime);
P
Peter Zijlstra 已提交
7915
	}
P
Peter Zijlstra 已提交
7916

P
Peter Zijlstra 已提交
7917 7918 7919 7920
	if (sum > total)
		return -EINVAL;

	return 0;
P
Peter Zijlstra 已提交
7921 7922
}

P
Peter Zijlstra 已提交
7923
static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
7924
{
7925 7926
	int ret;

P
Peter Zijlstra 已提交
7927 7928 7929 7930 7931 7932
	struct rt_schedulable_data data = {
		.tg = tg,
		.rt_period = period,
		.rt_runtime = runtime,
	};

7933 7934 7935 7936 7937
	rcu_read_lock();
	ret = walk_tg_tree(tg_rt_schedulable, tg_nop, &data);
	rcu_read_unlock();

	return ret;
7938 7939
}

7940
static int tg_set_rt_bandwidth(struct task_group *tg,
7941
		u64 rt_period, u64 rt_runtime)
P
Peter Zijlstra 已提交
7942
{
P
Peter Zijlstra 已提交
7943
	int i, err = 0;
P
Peter Zijlstra 已提交
7944

7945 7946 7947 7948 7949 7950 7951 7952 7953 7954 7955
	/*
	 * 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
7956
	mutex_lock(&rt_constraints_mutex);
7957
	read_lock(&tasklist_lock);
P
Peter Zijlstra 已提交
7958 7959
	err = __rt_schedulable(tg, rt_period, rt_runtime);
	if (err)
P
Peter Zijlstra 已提交
7960
		goto unlock;
P
Peter Zijlstra 已提交
7961

7962
	raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock);
7963 7964
	tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
	tg->rt_bandwidth.rt_runtime = rt_runtime;
P
Peter Zijlstra 已提交
7965 7966 7967 7968

	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = tg->rt_rq[i];

7969
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7970
		rt_rq->rt_runtime = rt_runtime;
7971
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
7972
	}
7973
	raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
P
Peter Zijlstra 已提交
7974
unlock:
7975
	read_unlock(&tasklist_lock);
P
Peter Zijlstra 已提交
7976 7977 7978
	mutex_unlock(&rt_constraints_mutex);

	return err;
P
Peter Zijlstra 已提交
7979 7980
}

7981
static int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
7982 7983 7984 7985 7986 7987 7988 7989
{
	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;

7990
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
7991 7992
}

7993
static long sched_group_rt_runtime(struct task_group *tg)
P
Peter Zijlstra 已提交
7994 7995 7996
{
	u64 rt_runtime_us;

7997
	if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
P
Peter Zijlstra 已提交
7998 7999
		return -1;

8000
	rt_runtime_us = tg->rt_bandwidth.rt_runtime;
P
Peter Zijlstra 已提交
8001 8002 8003
	do_div(rt_runtime_us, NSEC_PER_USEC);
	return rt_runtime_us;
}
8004

8005
static int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us)
8006 8007 8008
{
	u64 rt_runtime, rt_period;

8009
	rt_period = rt_period_us * NSEC_PER_USEC;
8010 8011
	rt_runtime = tg->rt_bandwidth.rt_runtime;

8012
	return tg_set_rt_bandwidth(tg, rt_period, rt_runtime);
8013 8014
}

8015
static long sched_group_rt_period(struct task_group *tg)
8016 8017 8018 8019 8020 8021 8022
{
	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;
}
8023
#endif /* CONFIG_RT_GROUP_SCHED */
8024

8025
#ifdef CONFIG_RT_GROUP_SCHED
8026 8027 8028 8029 8030
static int sched_rt_global_constraints(void)
{
	int ret = 0;

	mutex_lock(&rt_constraints_mutex);
P
Peter Zijlstra 已提交
8031
	read_lock(&tasklist_lock);
8032
	ret = __rt_schedulable(NULL, 0, 0);
P
Peter Zijlstra 已提交
8033
	read_unlock(&tasklist_lock);
8034 8035 8036 8037
	mutex_unlock(&rt_constraints_mutex);

	return ret;
}
8038

8039
static int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk)
8040 8041 8042 8043 8044 8045 8046 8047
{
	/* 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;
}

8048
#else /* !CONFIG_RT_GROUP_SCHED */
8049 8050
static int sched_rt_global_constraints(void)
{
P
Peter Zijlstra 已提交
8051
	unsigned long flags;
8052
	int i, ret = 0;
8053

8054
	raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8055 8056 8057
	for_each_possible_cpu(i) {
		struct rt_rq *rt_rq = &cpu_rq(i)->rt;

8058
		raw_spin_lock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8059
		rt_rq->rt_runtime = global_rt_runtime();
8060
		raw_spin_unlock(&rt_rq->rt_runtime_lock);
P
Peter Zijlstra 已提交
8061
	}
8062
	raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags);
P
Peter Zijlstra 已提交
8063

8064
	return ret;
8065
}
8066
#endif /* CONFIG_RT_GROUP_SCHED */
8067

8068
static int sched_dl_global_validate(void)
8069
{
8070 8071
	u64 runtime = global_rt_runtime();
	u64 period = global_rt_period();
8072
	u64 new_bw = to_ratio(period, runtime);
8073
	struct dl_bw *dl_b;
8074
	int cpu, ret = 0;
8075
	unsigned long flags;
8076 8077 8078 8079 8080 8081 8082 8083 8084 8085

	/*
	 * 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!
	 */
8086
	for_each_possible_cpu(cpu) {
8087 8088
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8089

8090
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8091 8092
		if (new_bw < dl_b->total_bw)
			ret = -EBUSY;
8093
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8094

8095 8096
		rcu_read_unlock_sched();

8097 8098
		if (ret)
			break;
8099 8100
	}

8101
	return ret;
8102 8103
}

8104
static void sched_dl_do_global(void)
8105
{
8106
	u64 new_bw = -1;
8107
	struct dl_bw *dl_b;
8108
	int cpu;
8109
	unsigned long flags;
8110

8111 8112 8113 8114 8115 8116 8117 8118 8119 8120
	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) {
8121 8122
		rcu_read_lock_sched();
		dl_b = dl_bw_of(cpu);
8123

8124
		raw_spin_lock_irqsave(&dl_b->lock, flags);
8125
		dl_b->bw = new_bw;
8126
		raw_spin_unlock_irqrestore(&dl_b->lock, flags);
8127 8128

		rcu_read_unlock_sched();
8129
	}
8130 8131 8132 8133 8134 8135 8136
}

static int sched_rt_global_validate(void)
{
	if (sysctl_sched_rt_period <= 0)
		return -EINVAL;

8137 8138
	if ((sysctl_sched_rt_runtime != RUNTIME_INF) &&
		(sysctl_sched_rt_runtime > sysctl_sched_rt_period))
8139 8140 8141 8142 8143 8144 8145 8146 8147
		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());
8148 8149
}

8150
int sched_rt_handler(struct ctl_table *table, int write,
8151
		void __user *buffer, size_t *lenp,
8152 8153 8154 8155
		loff_t *ppos)
{
	int old_period, old_runtime;
	static DEFINE_MUTEX(mutex);
8156
	int ret;
8157 8158 8159 8160 8161

	mutex_lock(&mutex);
	old_period = sysctl_sched_rt_period;
	old_runtime = sysctl_sched_rt_runtime;

8162
	ret = proc_dointvec(table, write, buffer, lenp, ppos);
8163 8164

	if (!ret && write) {
8165 8166 8167 8168
		ret = sched_rt_global_validate();
		if (ret)
			goto undo;

8169
		ret = sched_dl_global_validate();
8170 8171 8172
		if (ret)
			goto undo;

8173
		ret = sched_rt_global_constraints();
8174 8175 8176 8177 8178 8179 8180 8181 8182 8183
		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;
8184 8185 8186 8187 8188
	}
	mutex_unlock(&mutex);

	return ret;
}
8189

8190
int sched_rr_handler(struct ctl_table *table, int write,
8191 8192 8193 8194 8195 8196 8197 8198
		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);
8199 8200
	/* make sure that internally we keep jiffies */
	/* also, writing zero resets timeslice to default */
8201
	if (!ret && write) {
8202 8203
		sched_rr_timeslice = sched_rr_timeslice <= 0 ?
			RR_TIMESLICE : msecs_to_jiffies(sched_rr_timeslice);
8204 8205 8206 8207 8208
	}
	mutex_unlock(&mutex);
	return ret;
}

8209
#ifdef CONFIG_CGROUP_SCHED
8210

8211
static inline struct task_group *css_tg(struct cgroup_subsys_state *css)
8212
{
8213
	return css ? container_of(css, struct task_group, css) : NULL;
8214 8215
}

8216 8217
static struct cgroup_subsys_state *
cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
8218
{
8219 8220
	struct task_group *parent = css_tg(parent_css);
	struct task_group *tg;
8221

8222
	if (!parent) {
8223
		/* This is early initialization for the top cgroup */
8224
		return &root_task_group.css;
8225 8226
	}

8227
	tg = sched_create_group(parent);
8228 8229 8230 8231 8232 8233
	if (IS_ERR(tg))
		return ERR_PTR(-ENOMEM);

	return &tg->css;
}

8234
static int cpu_cgroup_css_online(struct cgroup_subsys_state *css)
8235
{
8236
	struct task_group *tg = css_tg(css);
T
Tejun Heo 已提交
8237
	struct task_group *parent = css_tg(css->parent);
8238

T
Tejun Heo 已提交
8239 8240
	if (parent)
		sched_online_group(tg, parent);
8241 8242 8243
	return 0;
}

8244
static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
8245
{
8246
	struct task_group *tg = css_tg(css);
8247 8248 8249 8250

	sched_destroy_group(tg);
}

8251
static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css)
8252
{
8253
	struct task_group *tg = css_tg(css);
8254 8255 8256 8257

	sched_offline_group(tg);
}

8258
static void cpu_cgroup_fork(struct task_struct *task, void *private)
8259 8260 8261 8262
{
	sched_move_task(task);
}

8263
static int cpu_cgroup_can_attach(struct cgroup_subsys_state *css,
8264
				 struct cgroup_taskset *tset)
8265
{
8266 8267
	struct task_struct *task;

8268
	cgroup_taskset_for_each(task, tset) {
8269
#ifdef CONFIG_RT_GROUP_SCHED
8270
		if (!sched_rt_can_attach(css_tg(css), task))
8271
			return -EINVAL;
8272
#else
8273 8274 8275
		/* We don't support RT-tasks being in separate groups */
		if (task->sched_class != &fair_sched_class)
			return -EINVAL;
8276
#endif
8277
	}
8278 8279
	return 0;
}
8280

8281
static void cpu_cgroup_attach(struct cgroup_subsys_state *css,
8282
			      struct cgroup_taskset *tset)
8283
{
8284 8285
	struct task_struct *task;

8286
	cgroup_taskset_for_each(task, tset)
8287
		sched_move_task(task);
8288 8289
}

8290
#ifdef CONFIG_FAIR_GROUP_SCHED
8291 8292
static int cpu_shares_write_u64(struct cgroup_subsys_state *css,
				struct cftype *cftype, u64 shareval)
8293
{
8294
	return sched_group_set_shares(css_tg(css), scale_load(shareval));
8295 8296
}

8297 8298
static u64 cpu_shares_read_u64(struct cgroup_subsys_state *css,
			       struct cftype *cft)
8299
{
8300
	struct task_group *tg = css_tg(css);
8301

8302
	return (u64) scale_load_down(tg->shares);
8303
}
8304 8305

#ifdef CONFIG_CFS_BANDWIDTH
8306 8307
static DEFINE_MUTEX(cfs_constraints_mutex);

8308 8309 8310
const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */

8311 8312
static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);

8313 8314
static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
{
8315
	int i, ret = 0, runtime_enabled, runtime_was_enabled;
8316
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8317 8318 8319 8320 8321 8322 8323 8324 8325 8326 8327 8328 8329 8330 8331 8332 8333 8334 8335 8336

	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;

8337 8338 8339 8340 8341
	/*
	 * Prevent race between setting of cfs_rq->runtime_enabled and
	 * unthrottle_offline_cfs_rqs().
	 */
	get_online_cpus();
8342 8343 8344 8345 8346
	mutex_lock(&cfs_constraints_mutex);
	ret = __cfs_schedulable(tg, period, quota);
	if (ret)
		goto out_unlock;

8347
	runtime_enabled = quota != RUNTIME_INF;
8348
	runtime_was_enabled = cfs_b->quota != RUNTIME_INF;
8349 8350 8351 8352 8353 8354
	/*
	 * 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();
8355 8356 8357
	raw_spin_lock_irq(&cfs_b->lock);
	cfs_b->period = ns_to_ktime(period);
	cfs_b->quota = quota;
8358

P
Paul Turner 已提交
8359
	__refill_cfs_bandwidth_runtime(cfs_b);
8360
	/* restart the period timer (if active) to handle new period expiry */
P
Peter Zijlstra 已提交
8361 8362
	if (runtime_enabled)
		start_cfs_bandwidth(cfs_b);
8363 8364
	raw_spin_unlock_irq(&cfs_b->lock);

8365
	for_each_online_cpu(i) {
8366
		struct cfs_rq *cfs_rq = tg->cfs_rq[i];
8367
		struct rq *rq = cfs_rq->rq;
8368 8369

		raw_spin_lock_irq(&rq->lock);
8370
		cfs_rq->runtime_enabled = runtime_enabled;
8371
		cfs_rq->runtime_remaining = 0;
8372

8373
		if (cfs_rq->throttled)
8374
			unthrottle_cfs_rq(cfs_rq);
8375 8376
		raw_spin_unlock_irq(&rq->lock);
	}
8377 8378
	if (runtime_was_enabled && !runtime_enabled)
		cfs_bandwidth_usage_dec();
8379 8380
out_unlock:
	mutex_unlock(&cfs_constraints_mutex);
8381
	put_online_cpus();
8382

8383
	return ret;
8384 8385 8386 8387 8388 8389
}

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

8390
	period = ktime_to_ns(tg->cfs_bandwidth.period);
8391 8392 8393 8394 8395 8396 8397 8398 8399 8400 8401 8402
	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;

8403
	if (tg->cfs_bandwidth.quota == RUNTIME_INF)
8404 8405
		return -1;

8406
	quota_us = tg->cfs_bandwidth.quota;
8407 8408 8409 8410 8411 8412 8413 8414 8415 8416
	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;
8417
	quota = tg->cfs_bandwidth.quota;
8418 8419 8420 8421 8422 8423 8424 8425

	return tg_set_cfs_bandwidth(tg, period, quota);
}

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

8426
	cfs_period_us = ktime_to_ns(tg->cfs_bandwidth.period);
8427 8428 8429 8430 8431
	do_div(cfs_period_us, NSEC_PER_USEC);

	return cfs_period_us;
}

8432 8433
static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
				  struct cftype *cft)
8434
{
8435
	return tg_get_cfs_quota(css_tg(css));
8436 8437
}

8438 8439
static int cpu_cfs_quota_write_s64(struct cgroup_subsys_state *css,
				   struct cftype *cftype, s64 cfs_quota_us)
8440
{
8441
	return tg_set_cfs_quota(css_tg(css), cfs_quota_us);
8442 8443
}

8444 8445
static u64 cpu_cfs_period_read_u64(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8446
{
8447
	return tg_get_cfs_period(css_tg(css));
8448 8449
}

8450 8451
static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 cfs_period_us)
8452
{
8453
	return tg_set_cfs_period(css_tg(css), cfs_period_us);
8454 8455
}

8456 8457 8458 8459 8460 8461 8462 8463 8464 8465 8466 8467 8468 8469 8470 8471 8472 8473 8474 8475 8476 8477 8478 8479 8480 8481 8482 8483 8484 8485 8486 8487
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;
8488
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8489 8490 8491 8492 8493
	s64 quota = 0, parent_quota = -1;

	if (!tg->parent) {
		quota = RUNTIME_INF;
	} else {
8494
		struct cfs_bandwidth *parent_b = &tg->parent->cfs_bandwidth;
8495 8496

		quota = normalize_cfs_quota(tg, d);
8497
		parent_quota = parent_b->hierarchical_quota;
8498 8499 8500 8501 8502 8503 8504 8505 8506 8507

		/*
		 * 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;
	}
8508
	cfs_b->hierarchical_quota = quota;
8509 8510 8511 8512 8513 8514

	return 0;
}

static int __cfs_schedulable(struct task_group *tg, u64 period, u64 quota)
{
8515
	int ret;
8516 8517 8518 8519 8520 8521 8522 8523 8524 8525 8526
	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);
	}

8527 8528 8529 8530 8531
	rcu_read_lock();
	ret = walk_tg_tree(tg_cfs_schedulable_down, tg_nop, &data);
	rcu_read_unlock();

	return ret;
8532
}
8533

8534
static int cpu_stats_show(struct seq_file *sf, void *v)
8535
{
8536
	struct task_group *tg = css_tg(seq_css(sf));
8537
	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
8538

8539 8540 8541
	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);
8542 8543 8544

	return 0;
}
8545
#endif /* CONFIG_CFS_BANDWIDTH */
8546
#endif /* CONFIG_FAIR_GROUP_SCHED */
8547

8548
#ifdef CONFIG_RT_GROUP_SCHED
8549 8550
static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
				struct cftype *cft, s64 val)
P
Peter Zijlstra 已提交
8551
{
8552
	return sched_group_set_rt_runtime(css_tg(css), val);
P
Peter Zijlstra 已提交
8553 8554
}

8555 8556
static s64 cpu_rt_runtime_read(struct cgroup_subsys_state *css,
			       struct cftype *cft)
P
Peter Zijlstra 已提交
8557
{
8558
	return sched_group_rt_runtime(css_tg(css));
P
Peter Zijlstra 已提交
8559
}
8560

8561 8562
static int cpu_rt_period_write_uint(struct cgroup_subsys_state *css,
				    struct cftype *cftype, u64 rt_period_us)
8563
{
8564
	return sched_group_set_rt_period(css_tg(css), rt_period_us);
8565 8566
}

8567 8568
static u64 cpu_rt_period_read_uint(struct cgroup_subsys_state *css,
				   struct cftype *cft)
8569
{
8570
	return sched_group_rt_period(css_tg(css));
8571
}
8572
#endif /* CONFIG_RT_GROUP_SCHED */
P
Peter Zijlstra 已提交
8573

8574
static struct cftype cpu_files[] = {
8575
#ifdef CONFIG_FAIR_GROUP_SCHED
8576 8577
	{
		.name = "shares",
8578 8579
		.read_u64 = cpu_shares_read_u64,
		.write_u64 = cpu_shares_write_u64,
8580
	},
8581
#endif
8582 8583 8584 8585 8586 8587 8588 8589 8590 8591 8592
#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,
	},
8593 8594
	{
		.name = "stat",
8595
		.seq_show = cpu_stats_show,
8596
	},
8597
#endif
8598
#ifdef CONFIG_RT_GROUP_SCHED
P
Peter Zijlstra 已提交
8599
	{
P
Peter Zijlstra 已提交
8600
		.name = "rt_runtime_us",
8601 8602
		.read_s64 = cpu_rt_runtime_read,
		.write_s64 = cpu_rt_runtime_write,
P
Peter Zijlstra 已提交
8603
	},
8604 8605
	{
		.name = "rt_period_us",
8606 8607
		.read_u64 = cpu_rt_period_read_uint,
		.write_u64 = cpu_rt_period_write_uint,
8608
	},
8609
#endif
8610
	{ }	/* terminate */
8611 8612
};

8613
struct cgroup_subsys cpu_cgrp_subsys = {
8614 8615
	.css_alloc	= cpu_cgroup_css_alloc,
	.css_free	= cpu_cgroup_css_free,
8616 8617
	.css_online	= cpu_cgroup_css_online,
	.css_offline	= cpu_cgroup_css_offline,
8618
	.fork		= cpu_cgroup_fork,
8619 8620
	.can_attach	= cpu_cgroup_can_attach,
	.attach		= cpu_cgroup_attach,
8621
	.legacy_cftypes	= cpu_files,
8622 8623 8624
	.early_init	= 1,
};

8625
#endif	/* CONFIG_CGROUP_SCHED */
8626

8627 8628 8629 8630 8631
void dump_cpu_task(int cpu)
{
	pr_info("Task dump for CPU %d:\n", cpu);
	sched_show_task(cpu_curr(cpu));
}
8632 8633 8634 8635 8636 8637 8638 8639 8640 8641 8642 8643 8644 8645 8646 8647 8648 8649 8650 8651 8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664 8665 8666 8667 8668 8669 8670 8671 8672

/*
 * 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,
};