• Home
  • History
  • Annotate
  • Line#
  • Navigate
  • Raw
  • Download
  • only in /netgear-R7000-V1.0.7.12_1.2.5/components/opensource/linux/linux-2.6.36/kernel/time/
1/*
2 *  linux/kernel/time/tick-sched.c
3 *
4 *  Copyright(C) 2005-2006, Thomas Gleixner <tglx@linutronix.de>
5 *  Copyright(C) 2005-2007, Red Hat, Inc., Ingo Molnar
6 *  Copyright(C) 2006-2007  Timesys Corp., Thomas Gleixner
7 *
8 *  No idle tick implementation for low and high resolution timers
9 *
10 *  Started by: Thomas Gleixner and Ingo Molnar
11 *
12 *  Distribute under GPLv2.
13 */
14#include <linux/cpu.h>
15#include <linux/err.h>
16#include <linux/hrtimer.h>
17#include <linux/interrupt.h>
18#include <linux/kernel_stat.h>
19#include <linux/percpu.h>
20#include <linux/profile.h>
21#include <linux/sched.h>
22#include <linux/tick.h>
23#include <linux/module.h>
24
25#include <asm/irq_regs.h>
26
27#include "tick-internal.h"
28
29/*
30 * Per cpu nohz control structure
31 */
32static DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched);
33
34/*
35 * The time, when the last jiffy update happened. Protected by xtime_lock.
36 */
37static ktime_t last_jiffies_update;
38
39struct tick_sched *tick_get_tick_sched(int cpu)
40{
41	return &per_cpu(tick_cpu_sched, cpu);
42}
43
44/*
45 * Must be called with interrupts disabled !
46 */
47static void tick_do_update_jiffies64(ktime_t now)
48{
49	unsigned long ticks = 0;
50	ktime_t delta;
51
52	/*
53	 * Do a quick check without holding xtime_lock:
54	 */
55	delta = ktime_sub(now, last_jiffies_update);
56	if (delta.tv64 < tick_period.tv64)
57		return;
58
59	/* Reevalute with xtime_lock held */
60	write_seqlock(&xtime_lock);
61
62	delta = ktime_sub(now, last_jiffies_update);
63	if (delta.tv64 >= tick_period.tv64) {
64
65		delta = ktime_sub(delta, tick_period);
66		last_jiffies_update = ktime_add(last_jiffies_update,
67						tick_period);
68
69		/* Slow path for long timeouts */
70		if (unlikely(delta.tv64 >= tick_period.tv64)) {
71			s64 incr = ktime_to_ns(tick_period);
72
73			ticks = ktime_divns(delta, incr);
74
75			last_jiffies_update = ktime_add_ns(last_jiffies_update,
76							   incr * ticks);
77		}
78		do_timer(++ticks);
79
80		/* Keep the tick_next_period variable up to date */
81		tick_next_period = ktime_add(last_jiffies_update, tick_period);
82	}
83	write_sequnlock(&xtime_lock);
84}
85
86/*
87 * Initialize and return retrieve the jiffies update.
88 */
89static ktime_t tick_init_jiffy_update(void)
90{
91	ktime_t period;
92
93	write_seqlock(&xtime_lock);
94	/* Did we start the jiffies update yet ? */
95	if (last_jiffies_update.tv64 == 0)
96		last_jiffies_update = tick_next_period;
97	period = last_jiffies_update;
98	write_sequnlock(&xtime_lock);
99	return period;
100}
101
102/*
103 * NOHZ - aka dynamic tick functionality
104 */
105#ifdef CONFIG_NO_HZ
106/*
107 * NO HZ enabled ?
108 */
109static int tick_nohz_enabled __read_mostly  = 1;
110
111/*
112 * Enable / Disable tickless mode
113 */
114static int __init setup_tick_nohz(char *str)
115{
116	if (!strcmp(str, "off"))
117		tick_nohz_enabled = 0;
118	else if (!strcmp(str, "on"))
119		tick_nohz_enabled = 1;
120	else
121		return 0;
122	return 1;
123}
124
125__setup("nohz=", setup_tick_nohz);
126
127/**
128 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
129 *
130 * Called from interrupt entry when the CPU was idle
131 *
132 * In case the sched_tick was stopped on this CPU, we have to check if jiffies
133 * must be updated. Otherwise an interrupt handler could use a stale jiffy
134 * value. We do this unconditionally on any cpu, as we don't know whether the
135 * cpu, which has the update task assigned is in a long sleep.
136 */
137static void tick_nohz_update_jiffies(ktime_t now)
138{
139	int cpu = smp_processor_id();
140	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
141	unsigned long flags;
142
143	cpumask_clear_cpu(cpu, nohz_cpu_mask);
144	ts->idle_waketime = now;
145
146	local_irq_save(flags);
147	tick_do_update_jiffies64(now);
148	local_irq_restore(flags);
149
150	touch_softlockup_watchdog();
151}
152
153/*
154 * Updates the per cpu time idle statistics counters
155 */
156static void
157update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time)
158{
159	ktime_t delta;
160
161	if (ts->idle_active) {
162		delta = ktime_sub(now, ts->idle_entrytime);
163		ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
164		if (nr_iowait_cpu(cpu) > 0)
165			ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
166		ts->idle_entrytime = now;
167	}
168
169	if (last_update_time)
170		*last_update_time = ktime_to_us(now);
171
172}
173
174static void tick_nohz_stop_idle(int cpu, ktime_t now)
175{
176	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
177
178	update_ts_time_stats(cpu, ts, now, NULL);
179	ts->idle_active = 0;
180
181	sched_clock_idle_wakeup_event(0);
182}
183
184static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts)
185{
186	ktime_t now;
187
188	now = ktime_get();
189
190	update_ts_time_stats(cpu, ts, now, NULL);
191
192	ts->idle_entrytime = now;
193	ts->idle_active = 1;
194	sched_clock_idle_sleep_event();
195	return now;
196}
197
198/**
199 * get_cpu_idle_time_us - get the total idle time of a cpu
200 * @cpu: CPU number to query
201 * @last_update_time: variable to store update time in
202 *
203 * Return the cummulative idle time (since boot) for a given
204 * CPU, in microseconds. The idle time returned includes
205 * the iowait time (unlike what "top" and co report).
206 *
207 * This time is measured via accounting rather than sampling,
208 * and is as accurate as ktime_get() is.
209 *
210 * This function returns -1 if NOHZ is not enabled.
211 */
212u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
213{
214	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
215
216	if (!tick_nohz_enabled)
217		return -1;
218
219	update_ts_time_stats(cpu, ts, ktime_get(), last_update_time);
220
221	return ktime_to_us(ts->idle_sleeptime);
222}
223EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
224
225/*
226 * get_cpu_iowait_time_us - get the total iowait time of a cpu
227 * @cpu: CPU number to query
228 * @last_update_time: variable to store update time in
229 *
230 * Return the cummulative iowait time (since boot) for a given
231 * CPU, in microseconds.
232 *
233 * This time is measured via accounting rather than sampling,
234 * and is as accurate as ktime_get() is.
235 *
236 * This function returns -1 if NOHZ is not enabled.
237 */
238u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
239{
240	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
241
242	if (!tick_nohz_enabled)
243		return -1;
244
245	update_ts_time_stats(cpu, ts, ktime_get(), last_update_time);
246
247	return ktime_to_us(ts->iowait_sleeptime);
248}
249EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
250
251/**
252 * tick_nohz_stop_sched_tick - stop the idle tick from the idle task
253 *
254 * When the next event is more than a tick into the future, stop the idle tick
255 * Called either from the idle loop or from irq_exit() when an idle period was
256 * just interrupted by an interrupt which did not cause a reschedule.
257 */
258void tick_nohz_stop_sched_tick(int inidle)
259{
260	unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags;
261	struct tick_sched *ts;
262	ktime_t last_update, expires, now;
263	struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
264	u64 time_delta;
265	int cpu;
266
267	local_irq_save(flags);
268
269	cpu = smp_processor_id();
270	ts = &per_cpu(tick_cpu_sched, cpu);
271
272	/*
273	 * Call to tick_nohz_start_idle stops the last_update_time from being
274	 * updated. Thus, it must not be called in the event we are called from
275	 * irq_exit() with the prior state different than idle.
276	 */
277	if (!inidle && !ts->inidle)
278		goto end;
279
280	/*
281	 * Set ts->inidle unconditionally. Even if the system did not
282	 * switch to NOHZ mode the cpu frequency governers rely on the
283	 * update of the idle time accounting in tick_nohz_start_idle().
284	 */
285	ts->inidle = 1;
286
287	now = tick_nohz_start_idle(cpu, ts);
288
289	/*
290	 * If this cpu is offline and it is the one which updates
291	 * jiffies, then give up the assignment and let it be taken by
292	 * the cpu which runs the tick timer next. If we don't drop
293	 * this here the jiffies might be stale and do_timer() never
294	 * invoked.
295	 */
296	if (unlikely(!cpu_online(cpu))) {
297		if (cpu == tick_do_timer_cpu)
298			tick_do_timer_cpu = TICK_DO_TIMER_NONE;
299	}
300
301	if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE))
302		goto end;
303
304	if (need_resched())
305		goto end;
306
307	if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
308		static int ratelimit;
309
310		if (ratelimit < 10) {
311			printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
312			       (unsigned int) local_softirq_pending());
313			ratelimit++;
314		}
315		goto end;
316	}
317
318	ts->idle_calls++;
319	/* Read jiffies and the time when jiffies were updated last */
320	do {
321		seq = read_seqbegin(&xtime_lock);
322		last_update = last_jiffies_update;
323		last_jiffies = jiffies;
324		time_delta = timekeeping_max_deferment();
325	} while (read_seqretry(&xtime_lock, seq));
326
327	if (rcu_needs_cpu(cpu) || printk_needs_cpu(cpu) ||
328	    arch_needs_cpu(cpu)) {
329		next_jiffies = last_jiffies + 1;
330		delta_jiffies = 1;
331	} else {
332		/* Get the next timer wheel timer */
333		next_jiffies = get_next_timer_interrupt(last_jiffies);
334		delta_jiffies = next_jiffies - last_jiffies;
335	}
336	/*
337	 * Do not stop the tick, if we are only one off
338	 * or if the cpu is required for rcu
339	 */
340	if (!ts->tick_stopped && delta_jiffies == 1)
341		goto out;
342
343	/* Schedule the tick, if we are at least one jiffie off */
344	if ((long)delta_jiffies >= 1) {
345
346		/*
347		 * If this cpu is the one which updates jiffies, then
348		 * give up the assignment and let it be taken by the
349		 * cpu which runs the tick timer next, which might be
350		 * this cpu as well. If we don't drop this here the
351		 * jiffies might be stale and do_timer() never
352		 * invoked. Keep track of the fact that it was the one
353		 * which had the do_timer() duty last. If this cpu is
354		 * the one which had the do_timer() duty last, we
355		 * limit the sleep time to the timekeeping
356		 * max_deferement value which we retrieved
357		 * above. Otherwise we can sleep as long as we want.
358		 */
359		if (cpu == tick_do_timer_cpu) {
360			tick_do_timer_cpu = TICK_DO_TIMER_NONE;
361			ts->do_timer_last = 1;
362		} else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) {
363			time_delta = KTIME_MAX;
364			ts->do_timer_last = 0;
365		} else if (!ts->do_timer_last) {
366			time_delta = KTIME_MAX;
367		}
368
369		/*
370		 * calculate the expiry time for the next timer wheel
371		 * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals
372		 * that there is no timer pending or at least extremely
373		 * far into the future (12 days for HZ=1000). In this
374		 * case we set the expiry to the end of time.
375		 */
376		if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) {
377			/*
378			 * Calculate the time delta for the next timer event.
379			 * If the time delta exceeds the maximum time delta
380			 * permitted by the current clocksource then adjust
381			 * the time delta accordingly to ensure the
382			 * clocksource does not wrap.
383			 */
384			time_delta = min_t(u64, time_delta,
385					   tick_period.tv64 * delta_jiffies);
386		}
387
388		if (time_delta < KTIME_MAX)
389			expires = ktime_add_ns(last_update, time_delta);
390		else
391			expires.tv64 = KTIME_MAX;
392
393		if (delta_jiffies > 1)
394			cpumask_set_cpu(cpu, nohz_cpu_mask);
395
396		/* Skip reprogram of event if its not changed */
397		if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
398			goto out;
399
400		/*
401		 * nohz_stop_sched_tick can be called several times before
402		 * the nohz_restart_sched_tick is called. This happens when
403		 * interrupts arrive which do not cause a reschedule. In the
404		 * first call we save the current tick time, so we can restart
405		 * the scheduler tick in nohz_restart_sched_tick.
406		 */
407		if (!ts->tick_stopped) {
408			select_nohz_load_balancer(1);
409
410			ts->idle_tick = hrtimer_get_expires(&ts->sched_timer);
411			ts->tick_stopped = 1;
412			ts->idle_jiffies = last_jiffies;
413			rcu_enter_nohz();
414		}
415
416		ts->idle_sleeps++;
417
418		/* Mark expires */
419		ts->idle_expires = expires;
420
421		/*
422		 * If the expiration time == KTIME_MAX, then
423		 * in this case we simply stop the tick timer.
424		 */
425		 if (unlikely(expires.tv64 == KTIME_MAX)) {
426			if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
427				hrtimer_cancel(&ts->sched_timer);
428			goto out;
429		}
430
431		if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
432			hrtimer_start(&ts->sched_timer, expires,
433				      HRTIMER_MODE_ABS_PINNED);
434			/* Check, if the timer was already in the past */
435			if (hrtimer_active(&ts->sched_timer))
436				goto out;
437		} else if (!tick_program_event(expires, 0))
438				goto out;
439		/*
440		 * We are past the event already. So we crossed a
441		 * jiffie boundary. Update jiffies and raise the
442		 * softirq.
443		 */
444		tick_do_update_jiffies64(ktime_get());
445		cpumask_clear_cpu(cpu, nohz_cpu_mask);
446	}
447	raise_softirq_irqoff(TIMER_SOFTIRQ);
448out:
449	ts->next_jiffies = next_jiffies;
450	ts->last_jiffies = last_jiffies;
451	ts->sleep_length = ktime_sub(dev->next_event, now);
452end:
453	local_irq_restore(flags);
454}
455
456/**
457 * tick_nohz_get_sleep_length - return the length of the current sleep
458 *
459 * Called from power state control code with interrupts disabled
460 */
461ktime_t tick_nohz_get_sleep_length(void)
462{
463	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
464
465	return ts->sleep_length;
466}
467
468static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
469{
470	hrtimer_cancel(&ts->sched_timer);
471	hrtimer_set_expires(&ts->sched_timer, ts->idle_tick);
472
473	while (1) {
474		/* Forward the time to expire in the future */
475		hrtimer_forward(&ts->sched_timer, now, tick_period);
476
477		if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
478			hrtimer_start_expires(&ts->sched_timer,
479					      HRTIMER_MODE_ABS_PINNED);
480			/* Check, if the timer was already in the past */
481			if (hrtimer_active(&ts->sched_timer))
482				break;
483		} else {
484			if (!tick_program_event(
485				hrtimer_get_expires(&ts->sched_timer), 0))
486				break;
487		}
488		/* Update jiffies and reread time */
489		tick_do_update_jiffies64(now);
490		now = ktime_get();
491	}
492}
493
494/**
495 * tick_nohz_restart_sched_tick - restart the idle tick from the idle task
496 *
497 * Restart the idle tick when the CPU is woken up from idle
498 */
499void tick_nohz_restart_sched_tick(void)
500{
501	int cpu = smp_processor_id();
502	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
503#ifndef CONFIG_VIRT_CPU_ACCOUNTING
504	unsigned long ticks;
505#endif
506	ktime_t now;
507
508	local_irq_disable();
509	if (ts->idle_active || (ts->inidle && ts->tick_stopped))
510		now = ktime_get();
511
512	if (ts->idle_active)
513		tick_nohz_stop_idle(cpu, now);
514
515	if (!ts->inidle || !ts->tick_stopped) {
516		ts->inidle = 0;
517		local_irq_enable();
518		return;
519	}
520
521	ts->inidle = 0;
522
523	rcu_exit_nohz();
524
525	/* Update jiffies first */
526	select_nohz_load_balancer(0);
527	tick_do_update_jiffies64(now);
528	cpumask_clear_cpu(cpu, nohz_cpu_mask);
529
530#ifndef CONFIG_VIRT_CPU_ACCOUNTING
531	/*
532	 * We stopped the tick in idle. Update process times would miss the
533	 * time we slept as update_process_times does only a 1 tick
534	 * accounting. Enforce that this is accounted to idle !
535	 */
536	ticks = jiffies - ts->idle_jiffies;
537	/*
538	 * We might be one off. Do not randomly account a huge number of ticks!
539	 */
540	if (ticks && ticks < LONG_MAX)
541		account_idle_ticks(ticks);
542#endif
543
544	touch_softlockup_watchdog();
545	/*
546	 * Cancel the scheduled timer and restore the tick
547	 */
548	ts->tick_stopped  = 0;
549	ts->idle_exittime = now;
550
551	tick_nohz_restart(ts, now);
552
553	local_irq_enable();
554}
555
556static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now)
557{
558	hrtimer_forward(&ts->sched_timer, now, tick_period);
559	return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0);
560}
561
562/*
563 * The nohz low res interrupt handler
564 */
565static void tick_nohz_handler(struct clock_event_device *dev)
566{
567	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
568	struct pt_regs *regs = get_irq_regs();
569	int cpu = smp_processor_id();
570	ktime_t now = ktime_get();
571
572	dev->next_event.tv64 = KTIME_MAX;
573
574	/*
575	 * Check if the do_timer duty was dropped. We don't care about
576	 * concurrency: This happens only when the cpu in charge went
577	 * into a long sleep. If two cpus happen to assign themself to
578	 * this duty, then the jiffies update is still serialized by
579	 * xtime_lock.
580	 */
581	if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
582		tick_do_timer_cpu = cpu;
583
584	/* Check, if the jiffies need an update */
585	if (tick_do_timer_cpu == cpu)
586		tick_do_update_jiffies64(now);
587
588	/*
589	 * When we are idle and the tick is stopped, we have to touch
590	 * the watchdog as we might not schedule for a really long
591	 * time. This happens on complete idle SMP systems while
592	 * waiting on the login prompt. We also increment the "start
593	 * of idle" jiffy stamp so the idle accounting adjustment we
594	 * do when we go busy again does not account too much ticks.
595	 */
596	if (ts->tick_stopped) {
597		touch_softlockup_watchdog();
598		ts->idle_jiffies++;
599	}
600
601	update_process_times(user_mode(regs));
602	profile_tick(CPU_PROFILING);
603
604	while (tick_nohz_reprogram(ts, now)) {
605		now = ktime_get();
606		tick_do_update_jiffies64(now);
607	}
608}
609
610/**
611 * tick_nohz_switch_to_nohz - switch to nohz mode
612 */
613static void tick_nohz_switch_to_nohz(void)
614{
615	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
616	ktime_t next;
617
618	if (!tick_nohz_enabled)
619		return;
620
621	local_irq_disable();
622	if (tick_switch_to_oneshot(tick_nohz_handler)) {
623		local_irq_enable();
624		return;
625	}
626
627	ts->nohz_mode = NOHZ_MODE_LOWRES;
628
629	/*
630	 * Recycle the hrtimer in ts, so we can share the
631	 * hrtimer_forward with the highres code.
632	 */
633	hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
634	/* Get the next period */
635	next = tick_init_jiffy_update();
636
637	for (;;) {
638		hrtimer_set_expires(&ts->sched_timer, next);
639		if (!tick_program_event(next, 0))
640			break;
641		next = ktime_add(next, tick_period);
642	}
643	local_irq_enable();
644
645	printk(KERN_INFO "Switched to NOHz mode on CPU #%d\n",
646	       smp_processor_id());
647}
648
649/*
650 * When NOHZ is enabled and the tick is stopped, we need to kick the
651 * tick timer from irq_enter() so that the jiffies update is kept
652 * alive during long running softirqs. That's ugly as hell, but
653 * correctness is key even if we need to fix the offending softirq in
654 * the first place.
655 *
656 * Note, this is different to tick_nohz_restart. We just kick the
657 * timer and do not touch the other magic bits which need to be done
658 * when idle is left.
659 */
660static void tick_nohz_kick_tick(int cpu, ktime_t now)
661{
662}
663
664static inline void tick_check_nohz(int cpu)
665{
666	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
667	ktime_t now;
668
669	if (!ts->idle_active && !ts->tick_stopped)
670		return;
671	now = ktime_get();
672	if (ts->idle_active)
673		tick_nohz_stop_idle(cpu, now);
674	if (ts->tick_stopped) {
675		tick_nohz_update_jiffies(now);
676		tick_nohz_kick_tick(cpu, now);
677	}
678}
679
680#else
681
682static inline void tick_nohz_switch_to_nohz(void) { }
683static inline void tick_check_nohz(int cpu) { }
684
685#endif /* NO_HZ */
686
687/*
688 * Called from irq_enter to notify about the possible interruption of idle()
689 */
690void tick_check_idle(int cpu)
691{
692	tick_check_oneshot_broadcast(cpu);
693	tick_check_nohz(cpu);
694}
695
696/*
697 * High resolution timer specific code
698 */
699#ifdef CONFIG_HIGH_RES_TIMERS
700/*
701 * We rearm the timer until we get disabled by the idle code.
702 * Called with interrupts disabled and timer->base->cpu_base->lock held.
703 */
704static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
705{
706	struct tick_sched *ts =
707		container_of(timer, struct tick_sched, sched_timer);
708	struct pt_regs *regs = get_irq_regs();
709	ktime_t now = ktime_get();
710	int cpu = smp_processor_id();
711
712#ifdef CONFIG_NO_HZ
713	/*
714	 * Check if the do_timer duty was dropped. We don't care about
715	 * concurrency: This happens only when the cpu in charge went
716	 * into a long sleep. If two cpus happen to assign themself to
717	 * this duty, then the jiffies update is still serialized by
718	 * xtime_lock.
719	 */
720	if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
721		tick_do_timer_cpu = cpu;
722#endif
723
724	/* Check, if the jiffies need an update */
725	if (tick_do_timer_cpu == cpu)
726		tick_do_update_jiffies64(now);
727
728	/*
729	 * Do not call, when we are not in irq context and have
730	 * no valid regs pointer
731	 */
732	if (regs) {
733		/*
734		 * When we are idle and the tick is stopped, we have to touch
735		 * the watchdog as we might not schedule for a really long
736		 * time. This happens on complete idle SMP systems while
737		 * waiting on the login prompt. We also increment the "start of
738		 * idle" jiffy stamp so the idle accounting adjustment we do
739		 * when we go busy again does not account too much ticks.
740		 */
741		if (ts->tick_stopped) {
742			touch_softlockup_watchdog();
743			ts->idle_jiffies++;
744		}
745		update_process_times(user_mode(regs));
746		profile_tick(CPU_PROFILING);
747	}
748
749	hrtimer_forward(timer, now, tick_period);
750
751	return HRTIMER_RESTART;
752}
753
754/**
755 * tick_setup_sched_timer - setup the tick emulation timer
756 */
757void tick_setup_sched_timer(void)
758{
759	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
760	ktime_t now = ktime_get();
761
762	/*
763	 * Emulate tick processing via per-CPU hrtimers:
764	 */
765	hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
766	ts->sched_timer.function = tick_sched_timer;
767
768	/* Get the next period (per cpu) */
769	hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
770
771	for (;;) {
772		hrtimer_forward(&ts->sched_timer, now, tick_period);
773		hrtimer_start_expires(&ts->sched_timer,
774				      HRTIMER_MODE_ABS_PINNED);
775		/* Check, if the timer was already in the past */
776		if (hrtimer_active(&ts->sched_timer))
777			break;
778		now = ktime_get();
779	}
780
781#ifdef CONFIG_NO_HZ
782	if (tick_nohz_enabled)
783		ts->nohz_mode = NOHZ_MODE_HIGHRES;
784#endif
785}
786#endif /* HIGH_RES_TIMERS */
787
788#if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
789void tick_cancel_sched_timer(int cpu)
790{
791	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
792
793# ifdef CONFIG_HIGH_RES_TIMERS
794	if (ts->sched_timer.base)
795		hrtimer_cancel(&ts->sched_timer);
796# endif
797
798	ts->nohz_mode = NOHZ_MODE_INACTIVE;
799}
800#endif
801
802/**
803 * Async notification about clocksource changes
804 */
805void tick_clock_notify(void)
806{
807	int cpu;
808
809	for_each_possible_cpu(cpu)
810		set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
811}
812
813/*
814 * Async notification about clock event changes
815 */
816void tick_oneshot_notify(void)
817{
818	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
819
820	set_bit(0, &ts->check_clocks);
821}
822
823/**
824 * Check, if a change happened, which makes oneshot possible.
825 *
826 * Called cyclic from the hrtimer softirq (driven by the timer
827 * softirq) allow_nohz signals, that we can switch into low-res nohz
828 * mode, because high resolution timers are disabled (either compile
829 * or runtime).
830 */
831int tick_check_oneshot_change(int allow_nohz)
832{
833	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
834
835	if (!test_and_clear_bit(0, &ts->check_clocks))
836		return 0;
837
838	if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
839		return 0;
840
841	if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
842		return 0;
843
844	if (!allow_nohz)
845		return 1;
846
847	tick_nohz_switch_to_nohz();
848	return 0;
849}
850