tick-sched.c 28.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
/*
 *  linux/kernel/time/tick-sched.c
 *
 *  Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
 *  Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
 *  Copyright(C) 2006-2007  Timesys Corp., Thomas Gleixner
 *
 *  No idle tick implementation for low and high resolution timers
 *
 *  Started by: Thomas Gleixner and Ingo Molnar
 *
P
Pavel Machek 已提交
12
 *  Distribute under GPLv2.
13 14 15 16 17 18 19 20 21
 */
#include <linux/cpu.h>
#include <linux/err.h>
#include <linux/hrtimer.h>
#include <linux/interrupt.h>
#include <linux/kernel_stat.h>
#include <linux/percpu.h>
#include <linux/profile.h>
#include <linux/sched.h>
22
#include <linux/module.h>
23
#include <linux/irq_work.h>
24 25
#include <linux/posix-timers.h>
#include <linux/perf_event.h>
26
#include <linux/context_tracking.h>
27

28 29
#include <asm/irq_regs.h>

30 31
#include "tick-internal.h"

F
Frederic Weisbecker 已提交
32 33
#include <trace/events/timer.h>

34 35 36
/*
 * Per cpu nohz control structure
 */
37
static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched);
38 39

/*
40
 * The time, when the last jiffy update happened. Protected by jiffies_lock.
41 42 43
 */
static ktime_t last_jiffies_update;

44 45 46 47 48
struct tick_sched *tick_get_tick_sched(int cpu)
{
	return &per_cpu(tick_cpu_sched, cpu);
}

49 50 51 52 53 54 55 56
/*
 * Must be called with interrupts disabled !
 */
static void tick_do_update_jiffies64(ktime_t now)
{
	unsigned long ticks = 0;
	ktime_t delta;

57
	/*
58
	 * Do a quick check without holding jiffies_lock:
59 60 61 62 63
	 */
	delta = ktime_sub(now, last_jiffies_update);
	if (delta.tv64 < tick_period.tv64)
		return;

64 65
	/* Reevalute with jiffies_lock held */
	write_seqlock(&jiffies_lock);
66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83

	delta = ktime_sub(now, last_jiffies_update);
	if (delta.tv64 >= tick_period.tv64) {

		delta = ktime_sub(delta, tick_period);
		last_jiffies_update = ktime_add(last_jiffies_update,
						tick_period);

		/* Slow path for long timeouts */
		if (unlikely(delta.tv64 >= tick_period.tv64)) {
			s64 incr = ktime_to_ns(tick_period);

			ticks = ktime_divns(delta, incr);

			last_jiffies_update = ktime_add_ns(last_jiffies_update,
							   incr * ticks);
		}
		do_timer(++ticks);
84 85 86

		/* Keep the tick_next_period variable up to date */
		tick_next_period = ktime_add(last_jiffies_update, tick_period);
87 88 89
	} else {
		write_sequnlock(&jiffies_lock);
		return;
90
	}
91
	write_sequnlock(&jiffies_lock);
92
	update_wall_time();
93 94 95 96 97 98 99 100 101
}

/*
 * Initialize and return retrieve the jiffies update.
 */
static ktime_t tick_init_jiffy_update(void)
{
	ktime_t period;

102
	write_seqlock(&jiffies_lock);
103 104 105 106
	/* Did we start the jiffies update yet ? */
	if (last_jiffies_update.tv64 == 0)
		last_jiffies_update = tick_next_period;
	period = last_jiffies_update;
107
	write_sequnlock(&jiffies_lock);
108 109 110
	return period;
}

111 112 113 114 115

static void tick_sched_do_timer(ktime_t now)
{
	int cpu = smp_processor_id();

116
#ifdef CONFIG_NO_HZ_COMMON
117 118 119 120 121
	/*
	 * Check if the do_timer duty was dropped. We don't care about
	 * concurrency: This happens only when the cpu in charge went
	 * into a long sleep. If two cpus happen to assign themself to
	 * this duty, then the jiffies update is still serialized by
122
	 * jiffies_lock.
123
	 */
124
	if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE)
125
	    && !tick_nohz_full_cpu(cpu))
126 127 128 129 130 131 132 133
		tick_do_timer_cpu = cpu;
#endif

	/* Check, if the jiffies need an update */
	if (tick_do_timer_cpu == cpu)
		tick_do_update_jiffies64(now);
}

134 135
static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs)
{
136
#ifdef CONFIG_NO_HZ_COMMON
137 138 139 140 141 142 143 144 145 146 147 148 149
	/*
	 * When we are idle and the tick is stopped, we have to touch
	 * the watchdog as we might not schedule for a really long
	 * time. This happens on complete idle SMP systems while
	 * waiting on the login prompt. We also increment the "start of
	 * idle" jiffy stamp so the idle accounting adjustment we do
	 * when we go busy again does not account too much ticks.
	 */
	if (ts->tick_stopped) {
		touch_softlockup_watchdog();
		if (is_idle_task(current))
			ts->idle_jiffies++;
	}
150
#endif
151 152 153 154
	update_process_times(user_mode(regs));
	profile_tick(CPU_PROFILING);
}

155
#ifdef CONFIG_NO_HZ_FULL
156
cpumask_var_t tick_nohz_full_mask;
157
cpumask_var_t housekeeping_mask;
158
bool tick_nohz_full_running;
159

160 161 162 163
static bool can_stop_full_tick(void)
{
	WARN_ON_ONCE(!irqs_disabled());

F
Frederic Weisbecker 已提交
164 165
	if (!sched_can_stop_tick()) {
		trace_tick_stop(0, "more than 1 task in runqueue\n");
166
		return false;
F
Frederic Weisbecker 已提交
167
	}
168

F
Frederic Weisbecker 已提交
169 170
	if (!posix_cpu_timers_can_stop_tick(current)) {
		trace_tick_stop(0, "posix timers running\n");
171
		return false;
F
Frederic Weisbecker 已提交
172
	}
173

F
Frederic Weisbecker 已提交
174 175
	if (!perf_event_can_stop_tick()) {
		trace_tick_stop(0, "perf events running\n");
176
		return false;
F
Frederic Weisbecker 已提交
177
	}
178 179 180 181 182 183 184

	/* sched_clock_tick() needs us? */
#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
	/*
	 * TODO: kick full dynticks CPUs when
	 * sched_clock_stable is set.
	 */
185
	if (!sched_clock_stable()) {
F
Frederic Weisbecker 已提交
186
		trace_tick_stop(0, "unstable sched clock\n");
187 188 189 190
		/*
		 * Don't allow the user to think they can get
		 * full NO_HZ with this machine.
		 */
191
		WARN_ONCE(tick_nohz_full_running,
192
			  "NO_HZ FULL will not work with unstable sched clock");
193
		return false;
F
Frederic Weisbecker 已提交
194
	}
195 196 197 198 199
#endif

	return true;
}

200 201
static void nohz_full_kick_work_func(struct irq_work *work)
{
202
	/* Empty, the tick restart happens on tick_nohz_irq_exit() */
203 204 205 206 207 208
}

static DEFINE_PER_CPU(struct irq_work, nohz_full_kick_work) = {
	.func = nohz_full_kick_work_func,
};

209 210 211 212 213 214 215 216 217 218 219
/*
 * Kick this CPU if it's full dynticks in order to force it to
 * re-evaluate its dependency on the tick and restart it if necessary.
 * This kick, unlike tick_nohz_full_kick_cpu() and tick_nohz_full_kick_all(),
 * is NMI safe.
 */
void tick_nohz_full_kick(void)
{
	if (!tick_nohz_full_cpu(smp_processor_id()))
		return;

220
	irq_work_queue(this_cpu_ptr(&nohz_full_kick_work));
221 222
}

223
/*
224
 * Kick the CPU if it's full dynticks in order to force it to
225 226
 * re-evaluate its dependency on the tick and restart it if necessary.
 */
227
void tick_nohz_full_kick_cpu(int cpu)
228
{
229 230 231 232
	if (!tick_nohz_full_cpu(cpu))
		return;

	irq_work_queue_on(&per_cpu(nohz_full_kick_work, cpu), cpu);
233 234 235 236
}

static void nohz_full_kick_ipi(void *info)
{
237
	/* Empty, the tick restart happens on tick_nohz_irq_exit() */
238 239 240 241 242 243 244 245
}

/*
 * Kick all full dynticks CPUs in order to force these to re-evaluate
 * their dependency on the tick and restart it if necessary.
 */
void tick_nohz_full_kick_all(void)
{
246
	if (!tick_nohz_full_running)
247 248 249
		return;

	preempt_disable();
250
	smp_call_function_many(tick_nohz_full_mask,
251
			       nohz_full_kick_ipi, NULL, false);
252
	tick_nohz_full_kick();
253 254 255
	preempt_enable();
}

256 257 258 259 260
/*
 * Re-evaluate the need for the tick as we switch the current task.
 * It might need the tick due to per task/process properties:
 * perf events, posix cpu timers, ...
 */
261
void __tick_nohz_task_switch(void)
262 263 264 265 266
{
	unsigned long flags;

	local_irq_save(flags);

267 268 269
	if (!tick_nohz_full_cpu(smp_processor_id()))
		goto out;

270 271 272
	if (tick_nohz_tick_stopped() && !can_stop_full_tick())
		tick_nohz_full_kick();

273
out:
274 275 276
	local_irq_restore(flags);
}

277
/* Parse the boot-time nohz CPU list from the kernel parameters. */
278
static int __init tick_nohz_full_setup(char *str)
279
{
280 281
	alloc_bootmem_cpumask_var(&tick_nohz_full_mask);
	if (cpulist_parse(str, tick_nohz_full_mask) < 0) {
282
		pr_warning("NOHZ: Incorrect nohz_full cpumask\n");
283
		free_bootmem_cpumask_var(tick_nohz_full_mask);
284 285
		return 1;
	}
286
	tick_nohz_full_running = true;
287

288 289
	return 1;
}
290
__setup("nohz_full=", tick_nohz_full_setup);
291

292
static int tick_nohz_cpu_down_callback(struct notifier_block *nfb,
293 294
				       unsigned long action,
				       void *hcpu)
295 296 297 298 299 300
{
	unsigned int cpu = (unsigned long)hcpu;

	switch (action & ~CPU_TASKS_FROZEN) {
	case CPU_DOWN_PREPARE:
		/*
301 302 303
		 * The boot CPU handles housekeeping duty (unbound timers,
		 * workqueues, timekeeping, ...) on behalf of full dynticks
		 * CPUs. It must remain online when nohz full is enabled.
304
		 */
305
		if (tick_nohz_full_running && tick_do_timer_cpu == cpu)
306
			return NOTIFY_BAD;
307 308 309 310 311
		break;
	}
	return NOTIFY_OK;
}

312 313 314 315 316
static int tick_nohz_init_all(void)
{
	int err = -1;

#ifdef CONFIG_NO_HZ_FULL_ALL
317
	if (!alloc_cpumask_var(&tick_nohz_full_mask, GFP_KERNEL)) {
318
		WARN(1, "NO_HZ: Can't allocate full dynticks cpumask\n");
319 320
		return err;
	}
321
	err = 0;
322 323
	cpumask_setall(tick_nohz_full_mask);
	tick_nohz_full_running = true;
324 325 326 327
#endif
	return err;
}

328
void __init tick_nohz_init(void)
329
{
330 331
	int cpu;

332
	if (!tick_nohz_full_running) {
333 334 335
		if (tick_nohz_init_all() < 0)
			return;
	}
336

337 338 339 340 341 342 343
	if (!alloc_cpumask_var(&housekeeping_mask, GFP_KERNEL)) {
		WARN(1, "NO_HZ: Can't allocate not-full dynticks cpumask\n");
		cpumask_clear(tick_nohz_full_mask);
		tick_nohz_full_running = false;
		return;
	}

344 345 346 347 348 349 350 351 352 353 354 355 356 357
	/*
	 * Full dynticks uses irq work to drive the tick rescheduling on safe
	 * locking contexts. But then we need irq work to raise its own
	 * interrupts to avoid circular dependency on the tick
	 */
	if (!arch_irq_work_has_interrupt()) {
		pr_warning("NO_HZ: Can't run full dynticks because arch doesn't "
			   "support irq work self-IPIs\n");
		cpumask_clear(tick_nohz_full_mask);
		cpumask_copy(housekeeping_mask, cpu_possible_mask);
		tick_nohz_full_running = false;
		return;
	}

358 359 360 361 362 363 364 365 366 367
	cpu = smp_processor_id();

	if (cpumask_test_cpu(cpu, tick_nohz_full_mask)) {
		pr_warning("NO_HZ: Clearing %d from nohz_full range for timekeeping\n", cpu);
		cpumask_clear_cpu(cpu, tick_nohz_full_mask);
	}

	cpumask_andnot(housekeeping_mask,
		       cpu_possible_mask, tick_nohz_full_mask);

368
	for_each_cpu(cpu, tick_nohz_full_mask)
369 370
		context_tracking_cpu_set(cpu);

371
	cpu_notifier(tick_nohz_cpu_down_callback, 0);
372 373
	pr_info("NO_HZ: Full dynticks CPUs: %*pbl.\n",
		cpumask_pr_args(tick_nohz_full_mask));
374 375 376 377 378 379

	/*
	 * We need at least one CPU to handle housekeeping work such
	 * as timekeeping, unbound timers, workqueues, ...
	 */
	WARN_ON_ONCE(cpumask_empty(housekeeping_mask));
380 381 382
}
#endif

383 384 385
/*
 * NOHZ - aka dynamic tick functionality
 */
386
#ifdef CONFIG_NO_HZ_COMMON
387 388 389
/*
 * NO HZ enabled ?
 */
390
static int tick_nohz_enabled __read_mostly  = 1;
391
unsigned long tick_nohz_active  __read_mostly;
392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407
/*
 * Enable / Disable tickless mode
 */
static int __init setup_tick_nohz(char *str)
{
	if (!strcmp(str, "off"))
		tick_nohz_enabled = 0;
	else if (!strcmp(str, "on"))
		tick_nohz_enabled = 1;
	else
		return 0;
	return 1;
}

__setup("nohz=", setup_tick_nohz);

408 409 410 411 412
int tick_nohz_tick_stopped(void)
{
	return __this_cpu_read(tick_cpu_sched.tick_stopped);
}

413 414 415 416 417 418 419 420 421 422
/**
 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
 *
 * Called from interrupt entry when the CPU was idle
 *
 * In case the sched_tick was stopped on this CPU, we have to check if jiffies
 * must be updated. Otherwise an interrupt handler could use a stale jiffy
 * value. We do this unconditionally on any cpu, as we don't know whether the
 * cpu, which has the update task assigned is in a long sleep.
 */
423
static void tick_nohz_update_jiffies(ktime_t now)
424 425 426
{
	unsigned long flags;

427
	__this_cpu_write(tick_cpu_sched.idle_waketime, now);
428 429 430 431

	local_irq_save(flags);
	tick_do_update_jiffies64(now);
	local_irq_restore(flags);
432 433

	touch_softlockup_watchdog();
434 435
}

436 437 438
/*
 * Updates the per cpu time idle statistics counters
 */
439
static void
440
update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time)
441
{
442
	ktime_t delta;
443

444 445
	if (ts->idle_active) {
		delta = ktime_sub(now, ts->idle_entrytime);
446
		if (nr_iowait_cpu(cpu) > 0)
447
			ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
448 449
		else
			ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
450
		ts->idle_entrytime = now;
451
	}
452

453
	if (last_update_time)
454 455
		*last_update_time = ktime_to_us(now);

456 457
}

458
static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
459
{
460
	update_ts_time_stats(smp_processor_id(), ts, now, NULL);
461
	ts->idle_active = 0;
462

463
	sched_clock_idle_wakeup_event(0);
464 465
}

466
static ktime_t tick_nohz_start_idle(struct tick_sched *ts)
467
{
468
	ktime_t now = ktime_get();
469

470 471
	ts->idle_entrytime = now;
	ts->idle_active = 1;
472
	sched_clock_idle_sleep_event();
473 474 475
	return now;
}

476 477 478
/**
 * get_cpu_idle_time_us - get the total idle time of a cpu
 * @cpu: CPU number to query
479 480
 * @last_update_time: variable to store update time in. Do not update
 * counters if NULL.
481 482
 *
 * Return the cummulative idle time (since boot) for a given
483
 * CPU, in microseconds.
484 485 486 487 488 489
 *
 * This time is measured via accounting rather than sampling,
 * and is as accurate as ktime_get() is.
 *
 * This function returns -1 if NOHZ is not enabled.
 */
490 491 492
u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
{
	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
493
	ktime_t now, idle;
494

495
	if (!tick_nohz_active)
496 497
		return -1;

498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
	now = ktime_get();
	if (last_update_time) {
		update_ts_time_stats(cpu, ts, now, last_update_time);
		idle = ts->idle_sleeptime;
	} else {
		if (ts->idle_active && !nr_iowait_cpu(cpu)) {
			ktime_t delta = ktime_sub(now, ts->idle_entrytime);

			idle = ktime_add(ts->idle_sleeptime, delta);
		} else {
			idle = ts->idle_sleeptime;
		}
	}

	return ktime_to_us(idle);
513

514
}
515
EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
516

517
/**
518 519
 * get_cpu_iowait_time_us - get the total iowait time of a cpu
 * @cpu: CPU number to query
520 521
 * @last_update_time: variable to store update time in. Do not update
 * counters if NULL.
522 523 524 525 526 527 528 529 530 531 532 533
 *
 * Return the cummulative iowait time (since boot) for a given
 * CPU, in microseconds.
 *
 * This time is measured via accounting rather than sampling,
 * and is as accurate as ktime_get() is.
 *
 * This function returns -1 if NOHZ is not enabled.
 */
u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
{
	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
534
	ktime_t now, iowait;
535

536
	if (!tick_nohz_active)
537 538
		return -1;

539 540 541 542 543 544 545
	now = ktime_get();
	if (last_update_time) {
		update_ts_time_stats(cpu, ts, now, last_update_time);
		iowait = ts->iowait_sleeptime;
	} else {
		if (ts->idle_active && nr_iowait_cpu(cpu) > 0) {
			ktime_t delta = ktime_sub(now, ts->idle_entrytime);
546

547 548 549 550 551
			iowait = ktime_add(ts->iowait_sleeptime, delta);
		} else {
			iowait = ts->iowait_sleeptime;
		}
	}
552

553
	return ktime_to_us(iowait);
554 555 556
}
EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);

557 558 559 560 561 562 563 564 565 566 567 568 569 570
static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
{
	hrtimer_cancel(&ts->sched_timer);
	hrtimer_set_expires(&ts->sched_timer, ts->last_tick);

	/* Forward the time to expire in the future */
	hrtimer_forward(&ts->sched_timer, now, tick_period);

	if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
		hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED);
	else
		tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
}

571 572
static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
					 ktime_t now, int cpu)
573
{
574
	struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
575 576 577
	u64 basemono, next_tick, next_tmr, next_rcu, delta, expires;
	unsigned long seq, basejiff;
	ktime_t	tick;
578

579 580
	/* Read jiffies and the time when jiffies were updated last */
	do {
581
		seq = read_seqbegin(&jiffies_lock);
582 583
		basemono = last_jiffies_update.tv64;
		basejiff = jiffies;
584
	} while (read_seqretry(&jiffies_lock, seq));
585
	ts->last_jiffies = basejiff;
586

587
	if (rcu_needs_cpu(basemono, &next_rcu) ||
588
	    arch_needs_cpu() || irq_work_needs_cpu()) {
589
		next_tick = basemono + TICK_NSEC;
590
	} else {
591 592 593 594 595 596 597 598 599 600 601
		/*
		 * Get the next pending timer. If high resolution
		 * timers are enabled this only takes the timer wheel
		 * timers into account. If high resolution timers are
		 * disabled this also looks at the next expiring
		 * hrtimer.
		 */
		next_tmr = get_next_timer_interrupt(basejiff, basemono);
		ts->next_timer = next_tmr;
		/* Take the next rcu event into account */
		next_tick = next_rcu < next_tmr ? next_rcu : next_tmr;
602
	}
603

604 605 606 607 608 609 610
	/*
	 * If the tick is due in the next period, keep it ticking or
	 * restart it proper.
	 */
	delta = next_tick - basemono;
	if (delta <= (u64)TICK_NSEC) {
		tick.tv64 = 0;
T
Thomas Gleixner 已提交
611 612
		if (!ts->tick_stopped)
			goto out;
613
		if (delta == 0) {
T
Thomas Gleixner 已提交
614 615 616 617 618 619
			/* Tick is stopped, but required now. Enforce it */
			tick_nohz_restart(ts, now);
			goto out;
		}
	}

620
	/*
T
Thomas Gleixner 已提交
621 622 623 624 625 626 627
	 * If this cpu is the one which updates jiffies, then give up
	 * the assignment and let it be taken by the cpu which runs
	 * the tick timer next, which might be this cpu as well. If we
	 * don't drop this here the jiffies might be stale and
	 * do_timer() never invoked. Keep track of the fact that it
	 * was the one which had the do_timer() duty last. If this cpu
	 * is the one which had the do_timer() duty last, we limit the
628 629
	 * sleep time to the timekeeping max_deferement value.
	 * Otherwise we can sleep as long as we want.
630
	 */
631
	delta = timekeeping_max_deferment();
T
Thomas Gleixner 已提交
632 633 634 635
	if (cpu == tick_do_timer_cpu) {
		tick_do_timer_cpu = TICK_DO_TIMER_NONE;
		ts->do_timer_last = 1;
	} else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) {
636
		delta = KTIME_MAX;
T
Thomas Gleixner 已提交
637 638
		ts->do_timer_last = 0;
	} else if (!ts->do_timer_last) {
639
		delta = KTIME_MAX;
T
Thomas Gleixner 已提交
640
	}
T
Thomas Gleixner 已提交
641

642
#ifdef CONFIG_NO_HZ_FULL
643
	/* Limit the tick delta to the maximum scheduler deferment */
T
Thomas Gleixner 已提交
644
	if (!ts->inidle)
645
		delta = min(delta, scheduler_tick_max_deferment());
646 647
#endif

648 649 650
	/* Calculate the next expiry time */
	if (delta < (KTIME_MAX - basemono))
		expires = basemono + delta;
T
Thomas Gleixner 已提交
651
	else
652 653 654 655
		expires = KTIME_MAX;

	expires = min_t(u64, expires, next_tick);
	tick.tv64 = expires;
656

T
Thomas Gleixner 已提交
657
	/* Skip reprogram of event if its not changed */
658
	if (ts->tick_stopped && (expires == dev->next_event.tv64))
T
Thomas Gleixner 已提交
659
		goto out;
660

T
Thomas Gleixner 已提交
661 662 663 664 665 666 667 668 669 670
	/*
	 * nohz_stop_sched_tick can be called several times before
	 * the nohz_restart_sched_tick is called. This happens when
	 * interrupts arrive which do not cause a reschedule. In the
	 * first call we save the current tick time, so we can restart
	 * the scheduler tick in nohz_restart_sched_tick.
	 */
	if (!ts->tick_stopped) {
		nohz_balance_enter_idle(cpu);
		calc_load_enter_idle();
671

T
Thomas Gleixner 已提交
672 673 674 675
		ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
		ts->tick_stopped = 1;
		trace_tick_stop(1, " ");
	}
676

T
Thomas Gleixner 已提交
677
	/*
678 679
	 * If the expiration time == KTIME_MAX, then we simply stop
	 * the tick timer.
T
Thomas Gleixner 已提交
680
	 */
681
	if (unlikely(expires == KTIME_MAX)) {
T
Thomas Gleixner 已提交
682 683 684
		if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
			hrtimer_cancel(&ts->sched_timer);
		goto out;
685
	}
686

T
Thomas Gleixner 已提交
687
	if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
688
		hrtimer_start(&ts->sched_timer, tick, HRTIMER_MODE_ABS_PINNED);
T
Thomas Gleixner 已提交
689
	else
690
		tick_program_event(tick, 1);
691
out:
692
	/* Update the estimated sleep length */
693
	ts->sleep_length = ktime_sub(dev->next_event, now);
694
	return tick;
695 696
}

697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712
static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now)
{
	/* Update jiffies first */
	tick_do_update_jiffies64(now);
	update_cpu_load_nohz();

	calc_load_exit_idle();
	touch_softlockup_watchdog();
	/*
	 * Cancel the scheduled timer and restore the tick
	 */
	ts->tick_stopped  = 0;
	ts->idle_exittime = now;

	tick_nohz_restart(ts, now);
}
713 714

static void tick_nohz_full_update_tick(struct tick_sched *ts)
715 716
{
#ifdef CONFIG_NO_HZ_FULL
717
	int cpu = smp_processor_id();
718

719
	if (!tick_nohz_full_cpu(cpu))
720
		return;
721

722 723
	if (!ts->tick_stopped && ts->nohz_mode == NOHZ_MODE_INACTIVE)
		return;
724

725 726 727 728
	if (can_stop_full_tick())
		tick_nohz_stop_sched_tick(ts, ktime_get(), cpu);
	else if (ts->tick_stopped)
		tick_nohz_restart_sched_tick(ts, ktime_get());
729 730 731
#endif
}

732 733 734 735 736 737 738 739 740 741 742 743
static bool can_stop_idle_tick(int cpu, struct tick_sched *ts)
{
	/*
	 * If this cpu is offline and it is the one which updates
	 * jiffies, then give up the assignment and let it be taken by
	 * the cpu which runs the tick timer next. If we don't drop
	 * this here the jiffies might be stale and do_timer() never
	 * invoked.
	 */
	if (unlikely(!cpu_online(cpu))) {
		if (cpu == tick_do_timer_cpu)
			tick_do_timer_cpu = TICK_DO_TIMER_NONE;
744
		return false;
745 746
	}

747 748
	if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE)) {
		ts->sleep_length = (ktime_t) { .tv64 = NSEC_PER_SEC/HZ };
749
		return false;
750
	}
751 752 753 754 755 756 757

	if (need_resched())
		return false;

	if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
		static int ratelimit;

758 759
		if (ratelimit < 10 &&
		    (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
760 761
			pr_warn("NOHZ: local_softirq_pending %02x\n",
				(unsigned int) local_softirq_pending());
762 763 764 765 766
			ratelimit++;
		}
		return false;
	}

767
	if (tick_nohz_full_enabled()) {
768 769 770 771 772 773 774 775 776 777 778 779 780 781
		/*
		 * Keep the tick alive to guarantee timekeeping progression
		 * if there are full dynticks CPUs around
		 */
		if (tick_do_timer_cpu == cpu)
			return false;
		/*
		 * Boot safety: make sure the timekeeping duty has been
		 * assigned before entering dyntick-idle mode,
		 */
		if (tick_do_timer_cpu == TICK_DO_TIMER_NONE)
			return false;
	}

782 783 784
	return true;
}

785 786
static void __tick_nohz_idle_enter(struct tick_sched *ts)
{
787
	ktime_t now, expires;
788
	int cpu = smp_processor_id();
789

790
	now = tick_nohz_start_idle(ts);
791

792 793 794 795
	if (can_stop_idle_tick(cpu, ts)) {
		int was_stopped = ts->tick_stopped;

		ts->idle_calls++;
796 797 798 799 800 801

		expires = tick_nohz_stop_sched_tick(ts, now, cpu);
		if (expires.tv64 > 0LL) {
			ts->idle_sleeps++;
			ts->idle_expires = expires;
		}
802 803 804 805

		if (!was_stopped && ts->tick_stopped)
			ts->idle_jiffies = ts->last_jiffies;
	}
806 807 808 809 810 811 812
}

/**
 * tick_nohz_idle_enter - stop the idle tick from the idle task
 *
 * When the next event is more than a tick into the future, stop the idle tick
 * Called when we start the idle loop.
813
 *
814
 * The arch is responsible of calling:
815 816 817 818
 *
 * - rcu_idle_enter() after its last use of RCU before the CPU is put
 *  to sleep.
 * - rcu_idle_exit() before the first use of RCU after the CPU is woken up.
819
 */
820
void tick_nohz_idle_enter(void)
821 822 823
{
	struct tick_sched *ts;

824 825
	WARN_ON_ONCE(irqs_disabled());

826 827 828 829 830 831 832 833
	/*
 	 * Update the idle state in the scheduler domain hierarchy
 	 * when tick_nohz_stop_sched_tick() is called from the idle loop.
 	 * State will be updated to busy during the first busy tick after
 	 * exiting idle.
 	 */
	set_cpu_sd_state_idle();

834 835
	local_irq_disable();

836
	ts = this_cpu_ptr(&tick_cpu_sched);
837
	ts->inidle = 1;
838
	__tick_nohz_idle_enter(ts);
839 840

	local_irq_enable();
841 842 843 844 845 846 847 848 849 850 851 852
}

/**
 * tick_nohz_irq_exit - update next tick event from interrupt exit
 *
 * When an interrupt fires while we are idle and it doesn't cause
 * a reschedule, it may still add, modify or delete a timer, enqueue
 * an RCU callback, etc...
 * So we need to re-calculate and reprogram the next tick event.
 */
void tick_nohz_irq_exit(void)
{
853
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
854

855
	if (ts->inidle)
856
		__tick_nohz_idle_enter(ts);
857
	else
858
		tick_nohz_full_update_tick(ts);
859 860
}

861 862 863 864 865 866 867
/**
 * tick_nohz_get_sleep_length - return the length of the current sleep
 *
 * Called from power state control code with interrupts disabled
 */
ktime_t tick_nohz_get_sleep_length(void)
{
868
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
869 870 871 872

	return ts->sleep_length;
}

873 874
static void tick_nohz_account_idle_ticks(struct tick_sched *ts)
{
875
#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
876
	unsigned long ticks;
877 878 879

	if (vtime_accounting_enabled())
		return;
880 881 882 883 884 885 886 887 888
	/*
	 * We stopped the tick in idle. Update process times would miss the
	 * time we slept as update_process_times does only a 1 tick
	 * accounting. Enforce that this is accounted to idle !
	 */
	ticks = jiffies - ts->idle_jiffies;
	/*
	 * We might be one off. Do not randomly account a huge number of ticks!
	 */
889 890 891
	if (ticks && ticks < LONG_MAX)
		account_idle_ticks(ticks);
#endif
892 893
}

894
/**
895
 * tick_nohz_idle_exit - restart the idle tick from the idle task
896 897
 *
 * Restart the idle tick when the CPU is woken up from idle
898 899
 * This also exit the RCU extended quiescent state. The CPU
 * can use RCU again after this function is called.
900
 */
901
void tick_nohz_idle_exit(void)
902
{
903
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
904
	ktime_t now;
905

906
	local_irq_disable();
907

908 909 910 911 912
	WARN_ON_ONCE(!ts->inidle);

	ts->inidle = 0;

	if (ts->idle_active || ts->tick_stopped)
913 914 915
		now = ktime_get();

	if (ts->idle_active)
916
		tick_nohz_stop_idle(ts, now);
917

918
	if (ts->tick_stopped) {
919
		tick_nohz_restart_sched_tick(ts, now);
920
		tick_nohz_account_idle_ticks(ts);
921
	}
922 923 924 925 926 927 928 929 930

	local_irq_enable();
}

/*
 * The nohz low res interrupt handler
 */
static void tick_nohz_handler(struct clock_event_device *dev)
{
931
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
932 933 934 935 936
	struct pt_regs *regs = get_irq_regs();
	ktime_t now = ktime_get();

	dev->next_event.tv64 = KTIME_MAX;

937
	tick_sched_do_timer(now);
938
	tick_sched_handle(ts, regs);
939

940 941 942 943
	/* No need to reprogram if we are running tickless  */
	if (unlikely(ts->tick_stopped))
		return;

944 945
	hrtimer_forward(&ts->sched_timer, now, tick_period);
	tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
946 947
}

948 949 950 951 952 953 954
static inline void tick_nohz_activate(struct tick_sched *ts, int mode)
{
	if (!tick_nohz_enabled)
		return;
	ts->nohz_mode = mode;
	/* One update is enough */
	if (!test_and_set_bit(0, &tick_nohz_active))
955
		timers_update_migration(true);
956 957
}

958 959 960 961 962
/**
 * tick_nohz_switch_to_nohz - switch to nohz mode
 */
static void tick_nohz_switch_to_nohz(void)
{
963
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
964 965
	ktime_t next;

966
	if (!tick_nohz_enabled)
967 968
		return;

969
	if (tick_switch_to_oneshot(tick_nohz_handler))
970
		return;
971

972 973 974 975 976 977 978 979
	/*
	 * Recycle the hrtimer in ts, so we can share the
	 * hrtimer_forward with the highres code.
	 */
	hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
	/* Get the next period */
	next = tick_init_jiffy_update();

980 981 982
	hrtimer_forward_now(&ts->sched_timer, tick_period);
	hrtimer_set_expires(&ts->sched_timer, next);
	tick_program_event(next, 1);
983
	tick_nohz_activate(ts, NOHZ_MODE_LOWRES);
984 985
}

986 987 988 989 990 991 992 993 994 995 996
/*
 * When NOHZ is enabled and the tick is stopped, we need to kick the
 * tick timer from irq_enter() so that the jiffies update is kept
 * alive during long running softirqs. That's ugly as hell, but
 * correctness is key even if we need to fix the offending softirq in
 * the first place.
 *
 * Note, this is different to tick_nohz_restart. We just kick the
 * timer and do not touch the other magic bits which need to be done
 * when idle is left.
 */
997
static void tick_nohz_kick_tick(struct tick_sched *ts, ktime_t now)
998
{
999 1000
#if 0
	/* Switch back to 2.6.27 behaviour */
1001
	ktime_t delta;
1002

1003 1004 1005 1006
	/*
	 * Do not touch the tick device, when the next expiry is either
	 * already reached or less/equal than the tick period.
	 */
1007
	delta =	ktime_sub(hrtimer_get_expires(&ts->sched_timer), now);
1008 1009 1010 1011
	if (delta.tv64 <= tick_period.tv64)
		return;

	tick_nohz_restart(ts, now);
1012
#endif
1013 1014
}

1015
static inline void tick_nohz_irq_enter(void)
1016
{
1017
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1018 1019 1020 1021 1022 1023
	ktime_t now;

	if (!ts->idle_active && !ts->tick_stopped)
		return;
	now = ktime_get();
	if (ts->idle_active)
1024
		tick_nohz_stop_idle(ts, now);
1025 1026
	if (ts->tick_stopped) {
		tick_nohz_update_jiffies(now);
1027
		tick_nohz_kick_tick(ts, now);
1028 1029 1030
	}
}

1031 1032 1033
#else

static inline void tick_nohz_switch_to_nohz(void) { }
1034
static inline void tick_nohz_irq_enter(void) { }
1035
static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { }
1036

1037
#endif /* CONFIG_NO_HZ_COMMON */
1038

1039 1040 1041
/*
 * Called from irq_enter to notify about the possible interruption of idle()
 */
1042
void tick_irq_enter(void)
1043
{
1044
	tick_check_oneshot_broadcast_this_cpu();
1045
	tick_nohz_irq_enter();
1046 1047
}

1048 1049 1050 1051 1052
/*
 * High resolution timer specific code
 */
#ifdef CONFIG_HIGH_RES_TIMERS
/*
P
Pavel Machek 已提交
1053
 * We rearm the timer until we get disabled by the idle code.
1054
 * Called with interrupts disabled.
1055 1056 1057 1058 1059 1060 1061
 */
static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
{
	struct tick_sched *ts =
		container_of(timer, struct tick_sched, sched_timer);
	struct pt_regs *regs = get_irq_regs();
	ktime_t now = ktime_get();
1062

1063
	tick_sched_do_timer(now);
1064 1065 1066 1067 1068

	/*
	 * Do not call, when we are not in irq context and have
	 * no valid regs pointer
	 */
1069 1070
	if (regs)
		tick_sched_handle(ts, regs);
1071

1072 1073 1074 1075
	/* No need to reprogram if we are in idle or full dynticks mode */
	if (unlikely(ts->tick_stopped))
		return HRTIMER_NORESTART;

1076 1077 1078 1079 1080
	hrtimer_forward(timer, now, tick_period);

	return HRTIMER_RESTART;
}

M
Mike Galbraith 已提交
1081 1082
static int sched_skew_tick;

1083 1084 1085 1086 1087 1088 1089 1090
static int __init skew_tick(char *str)
{
	get_option(&str, &sched_skew_tick);

	return 0;
}
early_param("skew_tick", skew_tick);

1091 1092 1093 1094 1095
/**
 * tick_setup_sched_timer - setup the tick emulation timer
 */
void tick_setup_sched_timer(void)
{
1096
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1097 1098 1099 1100 1101 1102 1103 1104
	ktime_t now = ktime_get();

	/*
	 * Emulate tick processing via per-CPU hrtimers:
	 */
	hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
	ts->sched_timer.function = tick_sched_timer;

1105
	/* Get the next period (per cpu) */
1106
	hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
1107

1108
	/* Offset the tick to avert jiffies_lock contention. */
M
Mike Galbraith 已提交
1109 1110 1111 1112 1113 1114 1115
	if (sched_skew_tick) {
		u64 offset = ktime_to_ns(tick_period) >> 1;
		do_div(offset, num_possible_cpus());
		offset *= smp_processor_id();
		hrtimer_add_expires_ns(&ts->sched_timer, offset);
	}

1116 1117
	hrtimer_forward(&ts->sched_timer, now, tick_period);
	hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED);
1118
	tick_nohz_activate(ts, NOHZ_MODE_HIGHRES);
1119
}
1120
#endif /* HIGH_RES_TIMERS */
1121

1122
#if defined CONFIG_NO_HZ_COMMON || defined CONFIG_HIGH_RES_TIMERS
1123 1124 1125 1126
void tick_cancel_sched_timer(int cpu)
{
	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);

1127
# ifdef CONFIG_HIGH_RES_TIMERS
1128 1129
	if (ts->sched_timer.base)
		hrtimer_cancel(&ts->sched_timer);
1130
# endif
1131

1132
	memset(ts, 0, sizeof(*ts));
1133
}
1134
#endif
1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151

/**
 * Async notification about clocksource changes
 */
void tick_clock_notify(void)
{
	int cpu;

	for_each_possible_cpu(cpu)
		set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
}

/*
 * Async notification about clock event changes
 */
void tick_oneshot_notify(void)
{
1152
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1153 1154 1155 1156 1157 1158 1159 1160 1161 1162

	set_bit(0, &ts->check_clocks);
}

/**
 * Check, if a change happened, which makes oneshot possible.
 *
 * Called cyclic from the hrtimer softirq (driven by the timer
 * softirq) allow_nohz signals, that we can switch into low-res nohz
 * mode, because high resolution timers are disabled (either compile
1163
 * or runtime). Called with interrupts disabled.
1164 1165 1166
 */
int tick_check_oneshot_change(int allow_nohz)
{
1167
	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
1168 1169 1170 1171 1172 1173 1174

	if (!test_and_clear_bit(0, &ts->check_clocks))
		return 0;

	if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
		return 0;

1175
	if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
1176 1177 1178 1179 1180 1181 1182 1183
		return 0;

	if (!allow_nohz)
		return 1;

	tick_nohz_switch_to_nohz();
	return 0;
}