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>
23 #include <linux/irq_work.h>
25 #include <asm/irq_regs.h>
27 #include "tick-internal.h"
30 * Per cpu nohz control structure
32 DEFINE_PER_CPU(struct tick_sched, tick_cpu_sched);
35 * The time, when the last jiffy update happened. Protected by xtime_lock.
37 static ktime_t last_jiffies_update;
39 struct tick_sched *tick_get_tick_sched(int cpu)
41 return &per_cpu(tick_cpu_sched, cpu);
45 * Must be called with interrupts disabled !
47 static void tick_do_update_jiffies64(ktime_t now)
49 unsigned long ticks = 0;
53 * Do a quick check without holding xtime_lock:
55 delta = ktime_sub(now, last_jiffies_update);
56 if (delta.tv64 < tick_period.tv64)
59 /* Reevalute with xtime_lock held */
60 write_seqlock(&xtime_lock);
62 delta = ktime_sub(now, last_jiffies_update);
63 if (delta.tv64 >= tick_period.tv64) {
65 delta = ktime_sub(delta, tick_period);
66 last_jiffies_update = ktime_add(last_jiffies_update,
69 /* Slow path for long timeouts */
70 if (unlikely(delta.tv64 >= tick_period.tv64)) {
71 s64 incr = ktime_to_ns(tick_period);
73 ticks = ktime_divns(delta, incr);
75 last_jiffies_update = ktime_add_ns(last_jiffies_update,
80 /* Keep the tick_next_period variable up to date */
81 tick_next_period = ktime_add(last_jiffies_update, tick_period);
83 write_sequnlock(&xtime_lock);
87 * Initialize and return retrieve the jiffies update.
89 static ktime_t tick_init_jiffy_update(void)
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);
103 * NOHZ - aka dynamic tick functionality
109 int tick_nohz_enabled __read_mostly = 1;
112 * Enable / Disable tickless mode
114 static int __init setup_tick_nohz(char *str)
116 if (!strcmp(str, "off"))
117 tick_nohz_enabled = 0;
118 else if (!strcmp(str, "on"))
119 tick_nohz_enabled = 1;
125 __setup("nohz=", setup_tick_nohz);
128 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
130 * Called from interrupt entry when the CPU was idle
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.
137 static void tick_nohz_update_jiffies(ktime_t now)
139 int cpu = smp_processor_id();
140 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
143 ts->idle_waketime = now;
145 local_irq_save(flags);
146 tick_do_update_jiffies64(now);
147 local_irq_restore(flags);
149 touch_softlockup_watchdog();
153 * Updates the per cpu time idle statistics counters
156 update_ts_time_stats(int cpu, struct tick_sched *ts, ktime_t now, u64 *last_update_time)
160 if (ts->idle_active) {
161 delta = ktime_sub(now, ts->idle_entrytime);
162 if (nr_iowait_cpu(cpu) > 0)
163 ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
165 ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
166 ts->idle_entrytime = now;
169 if (last_update_time)
170 *last_update_time = ktime_to_us(now);
174 static void tick_nohz_stop_idle(int cpu, ktime_t now)
176 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
178 update_ts_time_stats(cpu, ts, now, NULL);
181 sched_clock_idle_wakeup_event(0);
184 static ktime_t tick_nohz_start_idle(int cpu, struct tick_sched *ts)
186 ktime_t now = ktime_get();
188 ts->idle_entrytime = now;
190 sched_clock_idle_sleep_event();
195 * get_cpu_idle_time_us - get the total idle time of a cpu
196 * @cpu: CPU number to query
197 * @last_update_time: variable to store update time in. Do not update
200 * Return the cummulative idle time (since boot) for a given
201 * CPU, in microseconds.
203 * This time is measured via accounting rather than sampling,
204 * and is as accurate as ktime_get() is.
206 * This function returns -1 if NOHZ is not enabled.
208 u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
210 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
213 if (!tick_nohz_enabled)
217 if (last_update_time) {
218 update_ts_time_stats(cpu, ts, now, last_update_time);
219 idle = ts->idle_sleeptime;
221 if (ts->idle_active && !nr_iowait_cpu(cpu)) {
222 ktime_t delta = ktime_sub(now, ts->idle_entrytime);
224 idle = ktime_add(ts->idle_sleeptime, delta);
226 idle = ts->idle_sleeptime;
230 return ktime_to_us(idle);
233 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
236 * get_cpu_iowait_time_us - get the total iowait time of a cpu
237 * @cpu: CPU number to query
238 * @last_update_time: variable to store update time in. Do not update
241 * Return the cummulative iowait time (since boot) for a given
242 * CPU, in microseconds.
244 * This time is measured via accounting rather than sampling,
245 * and is as accurate as ktime_get() is.
247 * This function returns -1 if NOHZ is not enabled.
249 u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
251 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
254 if (!tick_nohz_enabled)
258 if (last_update_time) {
259 update_ts_time_stats(cpu, ts, now, last_update_time);
260 iowait = ts->iowait_sleeptime;
262 if (ts->idle_active && nr_iowait_cpu(cpu) > 0) {
263 ktime_t delta = ktime_sub(now, ts->idle_entrytime);
265 iowait = ktime_add(ts->iowait_sleeptime, delta);
267 iowait = ts->iowait_sleeptime;
271 return ktime_to_us(iowait);
273 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
275 static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
276 ktime_t now, int cpu)
278 unsigned long seq, last_jiffies, next_jiffies, delta_jiffies;
279 ktime_t last_update, expires, ret = { .tv64 = 0 };
280 unsigned long rcu_delta_jiffies;
281 struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
284 /* Read jiffies and the time when jiffies were updated last */
286 seq = read_seqbegin(&xtime_lock);
287 last_update = last_jiffies_update;
288 last_jiffies = jiffies;
289 time_delta = timekeeping_max_deferment();
290 } while (read_seqretry(&xtime_lock, seq));
292 if (rcu_needs_cpu(cpu, &rcu_delta_jiffies) ||
293 arch_needs_cpu(cpu) || irq_work_needs_cpu()) {
294 next_jiffies = last_jiffies + 1;
297 /* Get the next timer wheel timer */
298 next_jiffies = get_next_timer_interrupt(last_jiffies);
299 delta_jiffies = next_jiffies - last_jiffies;
300 if (rcu_delta_jiffies < delta_jiffies) {
301 next_jiffies = last_jiffies + rcu_delta_jiffies;
302 delta_jiffies = rcu_delta_jiffies;
306 * Do not stop the tick, if we are only one off
307 * or if the cpu is required for rcu
309 if (!ts->tick_stopped && delta_jiffies == 1)
312 /* Schedule the tick, if we are at least one jiffie off */
313 if ((long)delta_jiffies >= 1) {
316 * If this cpu is the one which updates jiffies, then
317 * give up the assignment and let it be taken by the
318 * cpu which runs the tick timer next, which might be
319 * this cpu as well. If we don't drop this here the
320 * jiffies might be stale and do_timer() never
321 * invoked. Keep track of the fact that it was the one
322 * which had the do_timer() duty last. If this cpu is
323 * the one which had the do_timer() duty last, we
324 * limit the sleep time to the timekeeping
325 * max_deferement value which we retrieved
326 * above. Otherwise we can sleep as long as we want.
328 if (cpu == tick_do_timer_cpu) {
329 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
330 ts->do_timer_last = 1;
331 } else if (tick_do_timer_cpu != TICK_DO_TIMER_NONE) {
332 time_delta = KTIME_MAX;
333 ts->do_timer_last = 0;
334 } else if (!ts->do_timer_last) {
335 time_delta = KTIME_MAX;
339 * calculate the expiry time for the next timer wheel
340 * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals
341 * that there is no timer pending or at least extremely
342 * far into the future (12 days for HZ=1000). In this
343 * case we set the expiry to the end of time.
345 if (likely(delta_jiffies < NEXT_TIMER_MAX_DELTA)) {
347 * Calculate the time delta for the next timer event.
348 * If the time delta exceeds the maximum time delta
349 * permitted by the current clocksource then adjust
350 * the time delta accordingly to ensure the
351 * clocksource does not wrap.
353 time_delta = min_t(u64, time_delta,
354 tick_period.tv64 * delta_jiffies);
357 if (time_delta < KTIME_MAX)
358 expires = ktime_add_ns(last_update, time_delta);
360 expires.tv64 = KTIME_MAX;
362 /* Skip reprogram of event if its not changed */
363 if (ts->tick_stopped && ktime_equal(expires, dev->next_event))
369 * nohz_stop_sched_tick can be called several times before
370 * the nohz_restart_sched_tick is called. This happens when
371 * interrupts arrive which do not cause a reschedule. In the
372 * first call we save the current tick time, so we can restart
373 * the scheduler tick in nohz_restart_sched_tick.
375 if (!ts->tick_stopped) {
376 nohz_balance_enter_idle(cpu);
377 calc_load_enter_idle();
379 ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
380 ts->tick_stopped = 1;
384 * If the expiration time == KTIME_MAX, then
385 * in this case we simply stop the tick timer.
387 if (unlikely(expires.tv64 == KTIME_MAX)) {
388 if (ts->nohz_mode == NOHZ_MODE_HIGHRES)
389 hrtimer_cancel(&ts->sched_timer);
393 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
394 hrtimer_start(&ts->sched_timer, expires,
395 HRTIMER_MODE_ABS_PINNED);
396 /* Check, if the timer was already in the past */
397 if (hrtimer_active(&ts->sched_timer))
399 } else if (!tick_program_event(expires, 0))
402 * We are past the event already. So we crossed a
403 * jiffie boundary. Update jiffies and raise the
406 tick_do_update_jiffies64(ktime_get());
408 raise_softirq_irqoff(TIMER_SOFTIRQ);
410 ts->next_jiffies = next_jiffies;
411 ts->last_jiffies = last_jiffies;
412 ts->sleep_length = ktime_sub(dev->next_event, now);
417 static bool can_stop_idle_tick(int cpu, struct tick_sched *ts)
420 * If this cpu is offline and it is the one which updates
421 * jiffies, then give up the assignment and let it be taken by
422 * the cpu which runs the tick timer next. If we don't drop
423 * this here the jiffies might be stale and do_timer() never
426 if (unlikely(!cpu_online(cpu))) {
427 if (cpu == tick_do_timer_cpu)
428 tick_do_timer_cpu = TICK_DO_TIMER_NONE;
431 if (unlikely(ts->nohz_mode == NOHZ_MODE_INACTIVE))
437 if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
438 static int ratelimit;
440 if (ratelimit < 10 &&
441 (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
442 printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
443 (unsigned int) local_softirq_pending());
452 static void __tick_nohz_idle_enter(struct tick_sched *ts)
454 ktime_t now, expires;
455 int cpu = smp_processor_id();
457 now = tick_nohz_start_idle(cpu, ts);
459 if (can_stop_idle_tick(cpu, ts)) {
460 int was_stopped = ts->tick_stopped;
464 expires = tick_nohz_stop_sched_tick(ts, now, cpu);
465 if (expires.tv64 > 0LL) {
467 ts->idle_expires = expires;
470 if (!was_stopped && ts->tick_stopped)
471 ts->idle_jiffies = ts->last_jiffies;
476 * tick_nohz_idle_enter - stop the idle tick from the idle task
478 * When the next event is more than a tick into the future, stop the idle tick
479 * Called when we start the idle loop.
481 * The arch is responsible of calling:
483 * - rcu_idle_enter() after its last use of RCU before the CPU is put
485 * - rcu_idle_exit() before the first use of RCU after the CPU is woken up.
487 void tick_nohz_idle_enter(void)
489 struct tick_sched *ts;
491 WARN_ON_ONCE(irqs_disabled());
494 * Update the idle state in the scheduler domain hierarchy
495 * when tick_nohz_stop_sched_tick() is called from the idle loop.
496 * State will be updated to busy during the first busy tick after
499 set_cpu_sd_state_idle();
503 ts = &__get_cpu_var(tick_cpu_sched);
505 * set ts->inidle unconditionally. even if the system did not
506 * switch to nohz mode the cpu frequency governers rely on the
507 * update of the idle time accounting in tick_nohz_start_idle().
510 __tick_nohz_idle_enter(ts);
516 * tick_nohz_irq_exit - update next tick event from interrupt exit
518 * When an interrupt fires while we are idle and it doesn't cause
519 * a reschedule, it may still add, modify or delete a timer, enqueue
520 * an RCU callback, etc...
521 * So we need to re-calculate and reprogram the next tick event.
523 void tick_nohz_irq_exit(void)
525 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
530 __tick_nohz_idle_enter(ts);
534 * tick_nohz_get_sleep_length - return the length of the current sleep
536 * Called from power state control code with interrupts disabled
538 ktime_t tick_nohz_get_sleep_length(void)
540 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
542 return ts->sleep_length;
545 static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
547 hrtimer_cancel(&ts->sched_timer);
548 hrtimer_set_expires(&ts->sched_timer, ts->last_tick);
551 /* Forward the time to expire in the future */
552 hrtimer_forward(&ts->sched_timer, now, tick_period);
554 if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
555 hrtimer_start_expires(&ts->sched_timer,
556 HRTIMER_MODE_ABS_PINNED);
557 /* Check, if the timer was already in the past */
558 if (hrtimer_active(&ts->sched_timer))
561 if (!tick_program_event(
562 hrtimer_get_expires(&ts->sched_timer), 0))
565 /* Reread time and update jiffies */
567 tick_do_update_jiffies64(now);
571 static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now)
573 /* Update jiffies first */
574 tick_do_update_jiffies64(now);
575 update_cpu_load_nohz();
577 calc_load_exit_idle();
578 touch_softlockup_watchdog();
580 * Cancel the scheduled timer and restore the tick
582 ts->tick_stopped = 0;
583 ts->idle_exittime = now;
585 tick_nohz_restart(ts, now);
588 static void tick_nohz_account_idle_ticks(struct tick_sched *ts)
590 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
593 * We stopped the tick in idle. Update process times would miss the
594 * time we slept as update_process_times does only a 1 tick
595 * accounting. Enforce that this is accounted to idle !
597 ticks = jiffies - ts->idle_jiffies;
599 * We might be one off. Do not randomly account a huge number of ticks!
601 if (ticks && ticks < LONG_MAX)
602 account_idle_ticks(ticks);
607 * tick_nohz_idle_exit - restart the idle tick from the idle task
609 * Restart the idle tick when the CPU is woken up from idle
610 * This also exit the RCU extended quiescent state. The CPU
611 * can use RCU again after this function is called.
613 void tick_nohz_idle_exit(void)
615 int cpu = smp_processor_id();
616 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
621 WARN_ON_ONCE(!ts->inidle);
625 if (ts->idle_active || ts->tick_stopped)
629 tick_nohz_stop_idle(cpu, now);
631 if (ts->tick_stopped) {
632 tick_nohz_restart_sched_tick(ts, now);
633 tick_nohz_account_idle_ticks(ts);
639 static int tick_nohz_reprogram(struct tick_sched *ts, ktime_t now)
641 hrtimer_forward(&ts->sched_timer, now, tick_period);
642 return tick_program_event(hrtimer_get_expires(&ts->sched_timer), 0);
646 * The nohz low res interrupt handler
648 static void tick_nohz_handler(struct clock_event_device *dev)
650 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
651 struct pt_regs *regs = get_irq_regs();
652 int cpu = smp_processor_id();
653 ktime_t now = ktime_get();
655 dev->next_event.tv64 = KTIME_MAX;
658 * Check if the do_timer duty was dropped. We don't care about
659 * concurrency: This happens only when the cpu in charge went
660 * into a long sleep. If two cpus happen to assign themself to
661 * this duty, then the jiffies update is still serialized by
664 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
665 tick_do_timer_cpu = cpu;
667 /* Check, if the jiffies need an update */
668 if (tick_do_timer_cpu == cpu)
669 tick_do_update_jiffies64(now);
672 * When we are idle and the tick is stopped, we have to touch
673 * the watchdog as we might not schedule for a really long
674 * time. This happens on complete idle SMP systems while
675 * waiting on the login prompt. We also increment the "start
676 * of idle" jiffy stamp so the idle accounting adjustment we
677 * do when we go busy again does not account too much ticks.
679 if (ts->tick_stopped) {
680 touch_softlockup_watchdog();
684 update_process_times(user_mode(regs));
685 profile_tick(CPU_PROFILING);
687 while (tick_nohz_reprogram(ts, now)) {
689 tick_do_update_jiffies64(now);
694 * tick_nohz_switch_to_nohz - switch to nohz mode
696 static void tick_nohz_switch_to_nohz(void)
698 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
701 if (!tick_nohz_enabled)
705 if (tick_switch_to_oneshot(tick_nohz_handler)) {
710 ts->nohz_mode = NOHZ_MODE_LOWRES;
713 * Recycle the hrtimer in ts, so we can share the
714 * hrtimer_forward with the highres code.
716 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
717 /* Get the next period */
718 next = tick_init_jiffy_update();
721 hrtimer_set_expires(&ts->sched_timer, next);
722 if (!tick_program_event(next, 0))
724 next = ktime_add(next, tick_period);
730 * When NOHZ is enabled and the tick is stopped, we need to kick the
731 * tick timer from irq_enter() so that the jiffies update is kept
732 * alive during long running softirqs. That's ugly as hell, but
733 * correctness is key even if we need to fix the offending softirq in
736 * Note, this is different to tick_nohz_restart. We just kick the
737 * timer and do not touch the other magic bits which need to be done
740 static void tick_nohz_kick_tick(int cpu, ktime_t now)
743 /* Switch back to 2.6.27 behaviour */
745 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
749 * Do not touch the tick device, when the next expiry is either
750 * already reached or less/equal than the tick period.
752 delta = ktime_sub(hrtimer_get_expires(&ts->sched_timer), now);
753 if (delta.tv64 <= tick_period.tv64)
756 tick_nohz_restart(ts, now);
760 static inline void tick_check_nohz(int cpu)
762 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
765 if (!ts->idle_active && !ts->tick_stopped)
769 tick_nohz_stop_idle(cpu, now);
770 if (ts->tick_stopped) {
771 tick_nohz_update_jiffies(now);
772 tick_nohz_kick_tick(cpu, now);
778 static inline void tick_nohz_switch_to_nohz(void) { }
779 static inline void tick_check_nohz(int cpu) { }
784 * Called from irq_enter to notify about the possible interruption of idle()
786 void tick_check_idle(int cpu)
788 tick_check_oneshot_broadcast(cpu);
789 tick_check_nohz(cpu);
793 * High resolution timer specific code
795 #ifdef CONFIG_HIGH_RES_TIMERS
797 * We rearm the timer until we get disabled by the idle code.
798 * Called with interrupts disabled and timer->base->cpu_base->lock held.
800 static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
802 struct tick_sched *ts =
803 container_of(timer, struct tick_sched, sched_timer);
804 struct pt_regs *regs = get_irq_regs();
805 ktime_t now = ktime_get();
806 int cpu = smp_processor_id();
810 * Check if the do_timer duty was dropped. We don't care about
811 * concurrency: This happens only when the cpu in charge went
812 * into a long sleep. If two cpus happen to assign themself to
813 * this duty, then the jiffies update is still serialized by
816 if (unlikely(tick_do_timer_cpu == TICK_DO_TIMER_NONE))
817 tick_do_timer_cpu = cpu;
820 /* Check, if the jiffies need an update */
821 if (tick_do_timer_cpu == cpu)
822 tick_do_update_jiffies64(now);
825 * Do not call, when we are not in irq context and have
826 * no valid regs pointer
830 * When we are idle and the tick is stopped, we have to touch
831 * the watchdog as we might not schedule for a really long
832 * time. This happens on complete idle SMP systems while
833 * waiting on the login prompt. We also increment the "start of
834 * idle" jiffy stamp so the idle accounting adjustment we do
835 * when we go busy again does not account too much ticks.
837 if (ts->tick_stopped) {
838 touch_softlockup_watchdog();
839 if (is_idle_task(current))
842 update_process_times(user_mode(regs));
843 profile_tick(CPU_PROFILING);
846 hrtimer_forward(timer, now, tick_period);
848 return HRTIMER_RESTART;
851 static int sched_skew_tick;
853 static int __init skew_tick(char *str)
855 get_option(&str, &sched_skew_tick);
859 early_param("skew_tick", skew_tick);
862 * tick_setup_sched_timer - setup the tick emulation timer
864 void tick_setup_sched_timer(void)
866 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
867 ktime_t now = ktime_get();
870 * Emulate tick processing via per-CPU hrtimers:
872 hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
873 ts->sched_timer.function = tick_sched_timer;
875 /* Get the next period (per cpu) */
876 hrtimer_set_expires(&ts->sched_timer, tick_init_jiffy_update());
878 /* Offset the tick to avert xtime_lock contention. */
879 if (sched_skew_tick) {
880 u64 offset = ktime_to_ns(tick_period) >> 1;
881 do_div(offset, num_possible_cpus());
882 offset *= smp_processor_id();
883 hrtimer_add_expires_ns(&ts->sched_timer, offset);
887 hrtimer_forward(&ts->sched_timer, now, tick_period);
888 hrtimer_start_expires(&ts->sched_timer,
889 HRTIMER_MODE_ABS_PINNED);
890 /* Check, if the timer was already in the past */
891 if (hrtimer_active(&ts->sched_timer))
897 if (tick_nohz_enabled)
898 ts->nohz_mode = NOHZ_MODE_HIGHRES;
901 #endif /* HIGH_RES_TIMERS */
903 #if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
904 void tick_cancel_sched_timer(int cpu)
906 struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
908 # ifdef CONFIG_HIGH_RES_TIMERS
909 if (ts->sched_timer.base)
910 hrtimer_cancel(&ts->sched_timer);
913 ts->nohz_mode = NOHZ_MODE_INACTIVE;
918 * Async notification about clocksource changes
920 void tick_clock_notify(void)
924 for_each_possible_cpu(cpu)
925 set_bit(0, &per_cpu(tick_cpu_sched, cpu).check_clocks);
929 * Async notification about clock event changes
931 void tick_oneshot_notify(void)
933 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
935 set_bit(0, &ts->check_clocks);
939 * Check, if a change happened, which makes oneshot possible.
941 * Called cyclic from the hrtimer softirq (driven by the timer
942 * softirq) allow_nohz signals, that we can switch into low-res nohz
943 * mode, because high resolution timers are disabled (either compile
946 int tick_check_oneshot_change(int allow_nohz)
948 struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
950 if (!test_and_clear_bit(0, &ts->check_clocks))
953 if (ts->nohz_mode != NOHZ_MODE_INACTIVE)
956 if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
962 tick_nohz_switch_to_nohz();