2 * linux/kernel/time/tick-sched.c
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
8 * No idle tick implementation for low and high resolution timers
10 * Started by: Thomas Gleixner and Ingo Molnar
12 * Distribute under GPLv2.
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/module.h>
24 #include <asm/irq_regs.h>
26 #include "tick-internal.h"
29 * Per cpu nohz control structure
31 static DEFINE_PER_CPU(struct tick_sched
, tick_cpu_sched
);
34 * The time, when the last jiffy update happened. Protected by xtime_lock.
36 static ktime_t last_jiffies_update
;
38 struct tick_sched
*tick_get_tick_sched(int cpu
)
40 return &per_cpu(tick_cpu_sched
, cpu
);
44 * Must be called with interrupts disabled !
46 static void tick_do_update_jiffies64(ktime_t now
)
48 unsigned long ticks
= 0;
52 * Do a quick check without holding xtime_lock:
54 delta
= ktime_sub(now
, last_jiffies_update
);
55 if (delta
.tv64
< tick_period
.tv64
)
58 /* Reevalute with xtime_lock held */
59 write_seqlock(&xtime_lock
);
61 delta
= ktime_sub(now
, last_jiffies_update
);
62 if (delta
.tv64
>= tick_period
.tv64
) {
64 delta
= ktime_sub(delta
, tick_period
);
65 last_jiffies_update
= ktime_add(last_jiffies_update
,
68 /* Slow path for long timeouts */
69 if (unlikely(delta
.tv64
>= tick_period
.tv64
)) {
70 s64 incr
= ktime_to_ns(tick_period
);
72 ticks
= ktime_divns(delta
, incr
);
74 last_jiffies_update
= ktime_add_ns(last_jiffies_update
,
79 /* Keep the tick_next_period variable up to date */
80 tick_next_period
= ktime_add(last_jiffies_update
, tick_period
);
82 write_sequnlock(&xtime_lock
);
86 * Initialize and return retrieve the jiffies update.
88 static ktime_t
tick_init_jiffy_update(void)
92 write_seqlock(&xtime_lock
);
93 /* Did we start the jiffies update yet ? */
94 if (last_jiffies_update
.tv64
== 0)
95 last_jiffies_update
= tick_next_period
;
96 period
= last_jiffies_update
;
97 write_sequnlock(&xtime_lock
);
102 * NOHZ - aka dynamic tick functionality
108 static int tick_nohz_enabled __read_mostly
= 1;
111 * Enable / Disable tickless mode
113 static int __init
setup_tick_nohz(char *str
)
115 if (!strcmp(str
, "off"))
116 tick_nohz_enabled
= 0;
117 else if (!strcmp(str
, "on"))
118 tick_nohz_enabled
= 1;
124 __setup("nohz=", setup_tick_nohz
);
127 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
129 * Called from interrupt entry when the CPU was idle
131 * In case the sched_tick was stopped on this CPU, we have to check if jiffies
132 * must be updated. Otherwise an interrupt handler could use a stale jiffy
133 * value. We do this unconditionally on any cpu, as we don't know whether the
134 * cpu, which has the update task assigned is in a long sleep.
136 static void tick_nohz_update_jiffies(ktime_t now
)
138 int cpu
= smp_processor_id();
139 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
142 ts
->idle_waketime
= now
;
144 local_irq_save(flags
);
145 tick_do_update_jiffies64(now
);
146 local_irq_restore(flags
);
148 touch_softlockup_watchdog();
152 * Updates the per cpu time idle statistics counters
155 update_ts_time_stats(int cpu
, struct tick_sched
*ts
, ktime_t now
, u64
*last_update_time
)
159 if (ts
->idle_active
) {
160 delta
= ktime_sub(now
, ts
->idle_entrytime
);
161 if (nr_iowait_cpu(cpu
) > 0)
162 ts
->iowait_sleeptime
= ktime_add(ts
->iowait_sleeptime
, delta
);
164 ts
->idle_sleeptime
= ktime_add(ts
->idle_sleeptime
, delta
);
165 ts
->idle_entrytime
= now
;
168 if (last_update_time
)
169 *last_update_time
= ktime_to_us(now
);
173 static void tick_nohz_stop_idle(int cpu
, ktime_t now
)
175 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
177 update_ts_time_stats(cpu
, ts
, now
, NULL
);
180 sched_clock_idle_wakeup_event(0);
183 static ktime_t
tick_nohz_start_idle(int cpu
, struct tick_sched
*ts
)
185 ktime_t now
= ktime_get();
187 ts
->idle_entrytime
= now
;
189 sched_clock_idle_sleep_event();
194 * get_cpu_idle_time_us - get the total idle time of a cpu
195 * @cpu: CPU number to query
196 * @last_update_time: variable to store update time in. Do not update
199 * Return the cummulative idle time (since boot) for a given
200 * CPU, in microseconds.
202 * This time is measured via accounting rather than sampling,
203 * and is as accurate as ktime_get() is.
205 * This function returns -1 if NOHZ is not enabled.
207 u64
get_cpu_idle_time_us(int cpu
, u64
*last_update_time
)
209 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
212 if (!tick_nohz_enabled
)
216 if (last_update_time
) {
217 update_ts_time_stats(cpu
, ts
, now
, last_update_time
);
218 idle
= ts
->idle_sleeptime
;
220 if (ts
->idle_active
&& !nr_iowait_cpu(cpu
)) {
221 ktime_t delta
= ktime_sub(now
, ts
->idle_entrytime
);
223 idle
= ktime_add(ts
->idle_sleeptime
, delta
);
225 idle
= ts
->idle_sleeptime
;
229 return ktime_to_us(idle
);
232 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us
);
235 * get_cpu_iowait_time_us - get the total iowait time of a cpu
236 * @cpu: CPU number to query
237 * @last_update_time: variable to store update time in. Do not update
240 * Return the cummulative iowait time (since boot) for a given
241 * CPU, in microseconds.
243 * This time is measured via accounting rather than sampling,
244 * and is as accurate as ktime_get() is.
246 * This function returns -1 if NOHZ is not enabled.
248 u64
get_cpu_iowait_time_us(int cpu
, u64
*last_update_time
)
250 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
253 if (!tick_nohz_enabled
)
257 if (last_update_time
) {
258 update_ts_time_stats(cpu
, ts
, now
, last_update_time
);
259 iowait
= ts
->iowait_sleeptime
;
261 if (ts
->idle_active
&& nr_iowait_cpu(cpu
) > 0) {
262 ktime_t delta
= ktime_sub(now
, ts
->idle_entrytime
);
264 iowait
= ktime_add(ts
->iowait_sleeptime
, delta
);
266 iowait
= ts
->iowait_sleeptime
;
270 return ktime_to_us(iowait
);
272 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us
);
274 static void tick_nohz_stop_sched_tick(struct tick_sched
*ts
)
276 unsigned long seq
, last_jiffies
, next_jiffies
, delta_jiffies
;
277 unsigned long rcu_delta_jiffies
;
278 ktime_t last_update
, expires
, now
;
279 struct clock_event_device
*dev
= __get_cpu_var(tick_cpu_device
).evtdev
;
283 cpu
= smp_processor_id();
284 ts
= &per_cpu(tick_cpu_sched
, cpu
);
286 now
= tick_nohz_start_idle(cpu
, ts
);
289 * If this cpu is offline and it is the one which updates
290 * jiffies, then give up the assignment and let it be taken by
291 * the cpu which runs the tick timer next. If we don't drop
292 * this here the jiffies might be stale and do_timer() never
295 if (unlikely(!cpu_online(cpu
))) {
296 if (cpu
== tick_do_timer_cpu
)
297 tick_do_timer_cpu
= TICK_DO_TIMER_NONE
;
300 if (unlikely(ts
->nohz_mode
== NOHZ_MODE_INACTIVE
))
306 if (unlikely(local_softirq_pending() && cpu_online(cpu
))) {
307 static int ratelimit
;
309 if (ratelimit
< 10) {
310 printk(KERN_ERR
"NOHZ: local_softirq_pending %02x\n",
311 (unsigned int) local_softirq_pending());
318 /* Read jiffies and the time when jiffies were updated last */
320 seq
= read_seqbegin(&xtime_lock
);
321 last_update
= last_jiffies_update
;
322 last_jiffies
= jiffies
;
323 time_delta
= timekeeping_max_deferment();
324 } while (read_seqretry(&xtime_lock
, seq
));
326 if (rcu_needs_cpu(cpu
, &rcu_delta_jiffies
) || printk_needs_cpu(cpu
) ||
327 arch_needs_cpu(cpu
)) {
328 next_jiffies
= last_jiffies
+ 1;
331 /* Get the next timer wheel timer */
332 next_jiffies
= get_next_timer_interrupt(last_jiffies
);
333 delta_jiffies
= next_jiffies
- last_jiffies
;
334 if (rcu_delta_jiffies
< delta_jiffies
) {
335 next_jiffies
= last_jiffies
+ rcu_delta_jiffies
;
336 delta_jiffies
= rcu_delta_jiffies
;
340 * Do not stop the tick, if we are only one off
341 * or if the cpu is required for rcu
343 if (!ts
->tick_stopped
&& delta_jiffies
== 1)
346 /* Schedule the tick, if we are at least one jiffie off */
347 if ((long)delta_jiffies
>= 1) {
350 * If this cpu is the one which updates jiffies, then
351 * give up the assignment and let it be taken by the
352 * cpu which runs the tick timer next, which might be
353 * this cpu as well. If we don't drop this here the
354 * jiffies might be stale and do_timer() never
355 * invoked. Keep track of the fact that it was the one
356 * which had the do_timer() duty last. If this cpu is
357 * the one which had the do_timer() duty last, we
358 * limit the sleep time to the timekeeping
359 * max_deferement value which we retrieved
360 * above. Otherwise we can sleep as long as we want.
362 if (cpu
== tick_do_timer_cpu
) {
363 tick_do_timer_cpu
= TICK_DO_TIMER_NONE
;
364 ts
->do_timer_last
= 1;
365 } else if (tick_do_timer_cpu
!= TICK_DO_TIMER_NONE
) {
366 time_delta
= KTIME_MAX
;
367 ts
->do_timer_last
= 0;
368 } else if (!ts
->do_timer_last
) {
369 time_delta
= KTIME_MAX
;
373 * calculate the expiry time for the next timer wheel
374 * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals
375 * that there is no timer pending or at least extremely
376 * far into the future (12 days for HZ=1000). In this
377 * case we set the expiry to the end of time.
379 if (likely(delta_jiffies
< NEXT_TIMER_MAX_DELTA
)) {
381 * Calculate the time delta for the next timer event.
382 * If the time delta exceeds the maximum time delta
383 * permitted by the current clocksource then adjust
384 * the time delta accordingly to ensure the
385 * clocksource does not wrap.
387 time_delta
= min_t(u64
, time_delta
,
388 tick_period
.tv64
* delta_jiffies
);
391 if (time_delta
< KTIME_MAX
)
392 expires
= ktime_add_ns(last_update
, time_delta
);
394 expires
.tv64
= KTIME_MAX
;
396 /* Skip reprogram of event if its not changed */
397 if (ts
->tick_stopped
&& ktime_equal(expires
, dev
->next_event
))
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.
407 if (!ts
->tick_stopped
) {
408 select_nohz_load_balancer(1);
410 ts
->idle_tick
= hrtimer_get_expires(&ts
->sched_timer
);
411 ts
->tick_stopped
= 1;
412 ts
->idle_jiffies
= last_jiffies
;
418 ts
->idle_expires
= expires
;
421 * If the expiration time == KTIME_MAX, then
422 * in this case we simply stop the tick timer.
424 if (unlikely(expires
.tv64
== KTIME_MAX
)) {
425 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
)
426 hrtimer_cancel(&ts
->sched_timer
);
430 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
) {
431 hrtimer_start(&ts
->sched_timer
, expires
,
432 HRTIMER_MODE_ABS_PINNED
);
433 /* Check, if the timer was already in the past */
434 if (hrtimer_active(&ts
->sched_timer
))
436 } else if (!tick_program_event(expires
, 0))
439 * We are past the event already. So we crossed a
440 * jiffie boundary. Update jiffies and raise the
443 tick_do_update_jiffies64(ktime_get());
445 raise_softirq_irqoff(TIMER_SOFTIRQ
);
447 ts
->next_jiffies
= next_jiffies
;
448 ts
->last_jiffies
= last_jiffies
;
449 ts
->sleep_length
= ktime_sub(dev
->next_event
, now
);
453 * tick_nohz_idle_enter - stop the idle tick from the idle task
455 * When the next event is more than a tick into the future, stop the idle tick
456 * Called when we start the idle loop.
458 * The arch is responsible of calling:
460 * - rcu_idle_enter() after its last use of RCU before the CPU is put
462 * - rcu_idle_exit() before the first use of RCU after the CPU is woken up.
464 void tick_nohz_idle_enter(void)
466 struct tick_sched
*ts
;
468 WARN_ON_ONCE(irqs_disabled());
471 * Update the idle state in the scheduler domain hierarchy
472 * when tick_nohz_stop_sched_tick() is called from the idle loop.
473 * State will be updated to busy during the first busy tick after
476 set_cpu_sd_state_idle();
480 ts
= &__get_cpu_var(tick_cpu_sched
);
482 * set ts->inidle unconditionally. even if the system did not
483 * switch to nohz mode the cpu frequency governers rely on the
484 * update of the idle time accounting in tick_nohz_start_idle().
487 tick_nohz_stop_sched_tick(ts
);
493 * tick_nohz_irq_exit - update next tick event from interrupt exit
495 * When an interrupt fires while we are idle and it doesn't cause
496 * a reschedule, it may still add, modify or delete a timer, enqueue
497 * an RCU callback, etc...
498 * So we need to re-calculate and reprogram the next tick event.
500 void tick_nohz_irq_exit(void)
502 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
507 tick_nohz_stop_sched_tick(ts
);
511 * tick_nohz_get_sleep_length - return the length of the current sleep
513 * Called from power state control code with interrupts disabled
515 ktime_t
tick_nohz_get_sleep_length(void)
517 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
519 return ts
->sleep_length
;
522 static void tick_nohz_restart(struct tick_sched
*ts
, ktime_t now
)
524 hrtimer_cancel(&ts
->sched_timer
);
525 hrtimer_set_expires(&ts
->sched_timer
, ts
->idle_tick
);
528 /* Forward the time to expire in the future */
529 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
531 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
) {
532 hrtimer_start_expires(&ts
->sched_timer
,
533 HRTIMER_MODE_ABS_PINNED
);
534 /* Check, if the timer was already in the past */
535 if (hrtimer_active(&ts
->sched_timer
))
538 if (!tick_program_event(
539 hrtimer_get_expires(&ts
->sched_timer
), 0))
542 /* Reread time and update jiffies */
544 tick_do_update_jiffies64(now
);
549 * tick_nohz_idle_exit - restart the idle tick from the idle task
551 * Restart the idle tick when the CPU is woken up from idle
552 * This also exit the RCU extended quiescent state. The CPU
553 * can use RCU again after this function is called.
555 void tick_nohz_idle_exit(void)
557 int cpu
= smp_processor_id();
558 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
559 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
566 WARN_ON_ONCE(!ts
->inidle
);
570 if (ts
->idle_active
|| ts
->tick_stopped
)
574 tick_nohz_stop_idle(cpu
, now
);
576 if (!ts
->tick_stopped
) {
581 /* Update jiffies first */
582 select_nohz_load_balancer(0);
583 tick_do_update_jiffies64(now
);
584 update_cpu_load_nohz();
586 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
588 * We stopped the tick in idle. Update process times would miss the
589 * time we slept as update_process_times does only a 1 tick
590 * accounting. Enforce that this is accounted to idle !
592 ticks
= jiffies
- ts
->idle_jiffies
;
594 * We might be one off. Do not randomly account a huge number of ticks!
596 if (ticks
&& ticks
< LONG_MAX
)
597 account_idle_ticks(ticks
);
600 touch_softlockup_watchdog();
602 * Cancel the scheduled timer and restore the tick
604 ts
->tick_stopped
= 0;
605 ts
->idle_exittime
= now
;
607 tick_nohz_restart(ts
, now
);
612 static int tick_nohz_reprogram(struct tick_sched
*ts
, ktime_t now
)
614 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
615 return tick_program_event(hrtimer_get_expires(&ts
->sched_timer
), 0);
619 * The nohz low res interrupt handler
621 static void tick_nohz_handler(struct clock_event_device
*dev
)
623 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
624 struct pt_regs
*regs
= get_irq_regs();
625 int cpu
= smp_processor_id();
626 ktime_t now
= ktime_get();
628 dev
->next_event
.tv64
= KTIME_MAX
;
631 * Check if the do_timer duty was dropped. We don't care about
632 * concurrency: This happens only when the cpu in charge went
633 * into a long sleep. If two cpus happen to assign themself to
634 * this duty, then the jiffies update is still serialized by
637 if (unlikely(tick_do_timer_cpu
== TICK_DO_TIMER_NONE
))
638 tick_do_timer_cpu
= cpu
;
640 /* Check, if the jiffies need an update */
641 if (tick_do_timer_cpu
== cpu
)
642 tick_do_update_jiffies64(now
);
645 * When we are idle and the tick is stopped, we have to touch
646 * the watchdog as we might not schedule for a really long
647 * time. This happens on complete idle SMP systems while
648 * waiting on the login prompt. We also increment the "start
649 * of idle" jiffy stamp so the idle accounting adjustment we
650 * do when we go busy again does not account too much ticks.
652 if (ts
->tick_stopped
) {
653 touch_softlockup_watchdog();
657 update_process_times(user_mode(regs
));
658 profile_tick(CPU_PROFILING
);
660 while (tick_nohz_reprogram(ts
, now
)) {
662 tick_do_update_jiffies64(now
);
667 * tick_nohz_switch_to_nohz - switch to nohz mode
669 static void tick_nohz_switch_to_nohz(void)
671 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
674 if (!tick_nohz_enabled
)
678 if (tick_switch_to_oneshot(tick_nohz_handler
)) {
683 ts
->nohz_mode
= NOHZ_MODE_LOWRES
;
686 * Recycle the hrtimer in ts, so we can share the
687 * hrtimer_forward with the highres code.
689 hrtimer_init(&ts
->sched_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
690 /* Get the next period */
691 next
= tick_init_jiffy_update();
694 hrtimer_set_expires(&ts
->sched_timer
, next
);
695 if (!tick_program_event(next
, 0))
697 next
= ktime_add(next
, tick_period
);
703 * When NOHZ is enabled and the tick is stopped, we need to kick the
704 * tick timer from irq_enter() so that the jiffies update is kept
705 * alive during long running softirqs. That's ugly as hell, but
706 * correctness is key even if we need to fix the offending softirq in
709 * Note, this is different to tick_nohz_restart. We just kick the
710 * timer and do not touch the other magic bits which need to be done
713 static void tick_nohz_kick_tick(int cpu
, ktime_t now
)
716 /* Switch back to 2.6.27 behaviour */
718 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
722 * Do not touch the tick device, when the next expiry is either
723 * already reached or less/equal than the tick period.
725 delta
= ktime_sub(hrtimer_get_expires(&ts
->sched_timer
), now
);
726 if (delta
.tv64
<= tick_period
.tv64
)
729 tick_nohz_restart(ts
, now
);
733 static inline void tick_check_nohz(int cpu
)
735 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
738 if (!ts
->idle_active
&& !ts
->tick_stopped
)
742 tick_nohz_stop_idle(cpu
, now
);
743 if (ts
->tick_stopped
) {
744 tick_nohz_update_jiffies(now
);
745 tick_nohz_kick_tick(cpu
, now
);
751 static inline void tick_nohz_switch_to_nohz(void) { }
752 static inline void tick_check_nohz(int cpu
) { }
757 * Called from irq_enter to notify about the possible interruption of idle()
759 void tick_check_idle(int cpu
)
761 tick_check_oneshot_broadcast(cpu
);
762 tick_check_nohz(cpu
);
766 * High resolution timer specific code
768 #ifdef CONFIG_HIGH_RES_TIMERS
770 * We rearm the timer until we get disabled by the idle code.
771 * Called with interrupts disabled and timer->base->cpu_base->lock held.
773 static enum hrtimer_restart
tick_sched_timer(struct hrtimer
*timer
)
775 struct tick_sched
*ts
=
776 container_of(timer
, struct tick_sched
, sched_timer
);
777 struct pt_regs
*regs
= get_irq_regs();
778 ktime_t now
= ktime_get();
779 int cpu
= smp_processor_id();
783 * Check if the do_timer duty was dropped. We don't care about
784 * concurrency: This happens only when the cpu in charge went
785 * into a long sleep. If two cpus happen to assign themself to
786 * this duty, then the jiffies update is still serialized by
789 if (unlikely(tick_do_timer_cpu
== TICK_DO_TIMER_NONE
))
790 tick_do_timer_cpu
= cpu
;
793 /* Check, if the jiffies need an update */
794 if (tick_do_timer_cpu
== cpu
)
795 tick_do_update_jiffies64(now
);
798 * Do not call, when we are not in irq context and have
799 * no valid regs pointer
803 * When we are idle and the tick is stopped, we have to touch
804 * the watchdog as we might not schedule for a really long
805 * time. This happens on complete idle SMP systems while
806 * waiting on the login prompt. We also increment the "start of
807 * idle" jiffy stamp so the idle accounting adjustment we do
808 * when we go busy again does not account too much ticks.
810 if (ts
->tick_stopped
) {
811 touch_softlockup_watchdog();
814 update_process_times(user_mode(regs
));
815 profile_tick(CPU_PROFILING
);
818 hrtimer_forward(timer
, now
, tick_period
);
820 return HRTIMER_RESTART
;
823 static int sched_skew_tick
;
825 static int __init
skew_tick(char *str
)
827 get_option(&str
, &sched_skew_tick
);
831 early_param("skew_tick", skew_tick
);
834 * tick_setup_sched_timer - setup the tick emulation timer
836 void tick_setup_sched_timer(void)
838 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
839 ktime_t now
= ktime_get();
842 * Emulate tick processing via per-CPU hrtimers:
844 hrtimer_init(&ts
->sched_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
845 ts
->sched_timer
.function
= tick_sched_timer
;
847 /* Get the next period (per cpu) */
848 hrtimer_set_expires(&ts
->sched_timer
, tick_init_jiffy_update());
850 /* Offset the tick to avert xtime_lock contention. */
851 if (sched_skew_tick
) {
852 u64 offset
= ktime_to_ns(tick_period
) >> 1;
853 do_div(offset
, num_possible_cpus());
854 offset
*= smp_processor_id();
855 hrtimer_add_expires_ns(&ts
->sched_timer
, offset
);
859 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
860 hrtimer_start_expires(&ts
->sched_timer
,
861 HRTIMER_MODE_ABS_PINNED
);
862 /* Check, if the timer was already in the past */
863 if (hrtimer_active(&ts
->sched_timer
))
869 if (tick_nohz_enabled
)
870 ts
->nohz_mode
= NOHZ_MODE_HIGHRES
;
873 #endif /* HIGH_RES_TIMERS */
875 #if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
876 void tick_cancel_sched_timer(int cpu
)
878 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
880 # ifdef CONFIG_HIGH_RES_TIMERS
881 if (ts
->sched_timer
.base
)
882 hrtimer_cancel(&ts
->sched_timer
);
885 ts
->nohz_mode
= NOHZ_MODE_INACTIVE
;
890 * Async notification about clocksource changes
892 void tick_clock_notify(void)
896 for_each_possible_cpu(cpu
)
897 set_bit(0, &per_cpu(tick_cpu_sched
, cpu
).check_clocks
);
901 * Async notification about clock event changes
903 void tick_oneshot_notify(void)
905 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
907 set_bit(0, &ts
->check_clocks
);
911 * Check, if a change happened, which makes oneshot possible.
913 * Called cyclic from the hrtimer softirq (driven by the timer
914 * softirq) allow_nohz signals, that we can switch into low-res nohz
915 * mode, because high resolution timers are disabled (either compile
918 int tick_check_oneshot_change(int allow_nohz
)
920 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
922 if (!test_and_clear_bit(0, &ts
->check_clocks
))
925 if (ts
->nohz_mode
!= NOHZ_MODE_INACTIVE
)
928 if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
934 tick_nohz_switch_to_nohz();