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 jiffies_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 jiffies_lock:
55 delta
= ktime_sub(now
, last_jiffies_update
);
56 if (delta
.tv64
< tick_period
.tv64
)
59 /* Reevalute with jiffies_lock held */
60 write_seqlock(&jiffies_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(&jiffies_lock
);
87 * Initialize and return retrieve the jiffies update.
89 static ktime_t
tick_init_jiffy_update(void)
93 write_seqlock(&jiffies_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(&jiffies_lock
);
103 static void tick_sched_do_timer(ktime_t now
)
105 int cpu
= smp_processor_id();
109 * Check if the do_timer duty was dropped. We don't care about
110 * concurrency: This happens only when the cpu in charge went
111 * into a long sleep. If two cpus happen to assign themself to
112 * this duty, then the jiffies update is still serialized by
115 if (unlikely(tick_do_timer_cpu
== TICK_DO_TIMER_NONE
))
116 tick_do_timer_cpu
= cpu
;
119 /* Check, if the jiffies need an update */
120 if (tick_do_timer_cpu
== cpu
)
121 tick_do_update_jiffies64(now
);
124 static void tick_sched_handle(struct tick_sched
*ts
, struct pt_regs
*regs
)
128 * When we are idle and the tick is stopped, we have to touch
129 * the watchdog as we might not schedule for a really long
130 * time. This happens on complete idle SMP systems while
131 * waiting on the login prompt. We also increment the "start of
132 * idle" jiffy stamp so the idle accounting adjustment we do
133 * when we go busy again does not account too much ticks.
135 if (ts
->tick_stopped
) {
136 touch_softlockup_watchdog();
137 if (is_idle_task(current
))
141 update_process_times(user_mode(regs
));
142 profile_tick(CPU_PROFILING
);
146 * NOHZ - aka dynamic tick functionality
152 int tick_nohz_enabled __read_mostly
= 1;
155 * Enable / Disable tickless mode
157 static int __init
setup_tick_nohz(char *str
)
159 if (!strcmp(str
, "off"))
160 tick_nohz_enabled
= 0;
161 else if (!strcmp(str
, "on"))
162 tick_nohz_enabled
= 1;
168 __setup("nohz=", setup_tick_nohz
);
171 * tick_nohz_update_jiffies - update jiffies when idle was interrupted
173 * Called from interrupt entry when the CPU was idle
175 * In case the sched_tick was stopped on this CPU, we have to check if jiffies
176 * must be updated. Otherwise an interrupt handler could use a stale jiffy
177 * value. We do this unconditionally on any cpu, as we don't know whether the
178 * cpu, which has the update task assigned is in a long sleep.
180 static void tick_nohz_update_jiffies(ktime_t now
)
182 int cpu
= smp_processor_id();
183 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
186 ts
->idle_waketime
= now
;
188 local_irq_save(flags
);
189 tick_do_update_jiffies64(now
);
190 local_irq_restore(flags
);
192 touch_softlockup_watchdog();
196 * Updates the per cpu time idle statistics counters
199 update_ts_time_stats(int cpu
, struct tick_sched
*ts
, ktime_t now
, u64
*last_update_time
)
203 if (ts
->idle_active
) {
204 delta
= ktime_sub(now
, ts
->idle_entrytime
);
205 if (nr_iowait_cpu(cpu
) > 0)
206 ts
->iowait_sleeptime
= ktime_add(ts
->iowait_sleeptime
, delta
);
208 ts
->idle_sleeptime
= ktime_add(ts
->idle_sleeptime
, delta
);
209 ts
->idle_entrytime
= now
;
212 if (last_update_time
)
213 *last_update_time
= ktime_to_us(now
);
217 static void tick_nohz_stop_idle(int cpu
, ktime_t now
)
219 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
221 update_ts_time_stats(cpu
, ts
, now
, NULL
);
224 sched_clock_idle_wakeup_event(0);
227 static ktime_t
tick_nohz_start_idle(int cpu
, struct tick_sched
*ts
)
229 ktime_t now
= ktime_get();
231 ts
->idle_entrytime
= now
;
233 sched_clock_idle_sleep_event();
238 * get_cpu_idle_time_us - get the total idle time of a cpu
239 * @cpu: CPU number to query
240 * @last_update_time: variable to store update time in. Do not update
243 * Return the cummulative idle time (since boot) for a given
244 * CPU, in microseconds.
246 * This time is measured via accounting rather than sampling,
247 * and is as accurate as ktime_get() is.
249 * This function returns -1 if NOHZ is not enabled.
251 u64
get_cpu_idle_time_us(int cpu
, u64
*last_update_time
)
253 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
256 if (!tick_nohz_enabled
)
260 if (last_update_time
) {
261 update_ts_time_stats(cpu
, ts
, now
, last_update_time
);
262 idle
= ts
->idle_sleeptime
;
264 if (ts
->idle_active
&& !nr_iowait_cpu(cpu
)) {
265 ktime_t delta
= ktime_sub(now
, ts
->idle_entrytime
);
267 idle
= ktime_add(ts
->idle_sleeptime
, delta
);
269 idle
= ts
->idle_sleeptime
;
273 return ktime_to_us(idle
);
276 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us
);
279 * get_cpu_iowait_time_us - get the total iowait time of a cpu
280 * @cpu: CPU number to query
281 * @last_update_time: variable to store update time in. Do not update
284 * Return the cummulative iowait time (since boot) for a given
285 * CPU, in microseconds.
287 * This time is measured via accounting rather than sampling,
288 * and is as accurate as ktime_get() is.
290 * This function returns -1 if NOHZ is not enabled.
292 u64
get_cpu_iowait_time_us(int cpu
, u64
*last_update_time
)
294 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
297 if (!tick_nohz_enabled
)
301 if (last_update_time
) {
302 update_ts_time_stats(cpu
, ts
, now
, last_update_time
);
303 iowait
= ts
->iowait_sleeptime
;
305 if (ts
->idle_active
&& nr_iowait_cpu(cpu
) > 0) {
306 ktime_t delta
= ktime_sub(now
, ts
->idle_entrytime
);
308 iowait
= ktime_add(ts
->iowait_sleeptime
, delta
);
310 iowait
= ts
->iowait_sleeptime
;
314 return ktime_to_us(iowait
);
316 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us
);
318 static ktime_t
tick_nohz_stop_sched_tick(struct tick_sched
*ts
,
319 ktime_t now
, int cpu
)
321 unsigned long seq
, last_jiffies
, next_jiffies
, delta_jiffies
;
322 ktime_t last_update
, expires
, ret
= { .tv64
= 0 };
323 unsigned long rcu_delta_jiffies
;
324 struct clock_event_device
*dev
= __get_cpu_var(tick_cpu_device
).evtdev
;
327 /* Read jiffies and the time when jiffies were updated last */
329 seq
= read_seqbegin(&jiffies_lock
);
330 last_update
= last_jiffies_update
;
331 last_jiffies
= jiffies
;
332 time_delta
= timekeeping_max_deferment();
333 } while (read_seqretry(&jiffies_lock
, seq
));
335 if (rcu_needs_cpu(cpu
, &rcu_delta_jiffies
) ||
336 arch_needs_cpu(cpu
) || irq_work_needs_cpu()) {
337 next_jiffies
= last_jiffies
+ 1;
340 /* Get the next timer wheel timer */
341 next_jiffies
= get_next_timer_interrupt(last_jiffies
);
342 delta_jiffies
= next_jiffies
- last_jiffies
;
343 if (rcu_delta_jiffies
< delta_jiffies
) {
344 next_jiffies
= last_jiffies
+ rcu_delta_jiffies
;
345 delta_jiffies
= rcu_delta_jiffies
;
349 * Do not stop the tick, if we are only one off
350 * or if the cpu is required for rcu
352 if (!ts
->tick_stopped
&& delta_jiffies
== 1)
355 /* Schedule the tick, if we are at least one jiffie off */
356 if ((long)delta_jiffies
>= 1) {
359 * If this cpu is the one which updates jiffies, then
360 * give up the assignment and let it be taken by the
361 * cpu which runs the tick timer next, which might be
362 * this cpu as well. If we don't drop this here the
363 * jiffies might be stale and do_timer() never
364 * invoked. Keep track of the fact that it was the one
365 * which had the do_timer() duty last. If this cpu is
366 * the one which had the do_timer() duty last, we
367 * limit the sleep time to the timekeeping
368 * max_deferement value which we retrieved
369 * above. Otherwise we can sleep as long as we want.
371 if (cpu
== tick_do_timer_cpu
) {
372 tick_do_timer_cpu
= TICK_DO_TIMER_NONE
;
373 ts
->do_timer_last
= 1;
374 } else if (tick_do_timer_cpu
!= TICK_DO_TIMER_NONE
) {
375 time_delta
= KTIME_MAX
;
376 ts
->do_timer_last
= 0;
377 } else if (!ts
->do_timer_last
) {
378 time_delta
= KTIME_MAX
;
382 * calculate the expiry time for the next timer wheel
383 * timer. delta_jiffies >= NEXT_TIMER_MAX_DELTA signals
384 * that there is no timer pending or at least extremely
385 * far into the future (12 days for HZ=1000). In this
386 * case we set the expiry to the end of time.
388 if (likely(delta_jiffies
< NEXT_TIMER_MAX_DELTA
)) {
390 * Calculate the time delta for the next timer event.
391 * If the time delta exceeds the maximum time delta
392 * permitted by the current clocksource then adjust
393 * the time delta accordingly to ensure the
394 * clocksource does not wrap.
396 time_delta
= min_t(u64
, time_delta
,
397 tick_period
.tv64
* delta_jiffies
);
400 if (time_delta
< KTIME_MAX
)
401 expires
= ktime_add_ns(last_update
, time_delta
);
403 expires
.tv64
= KTIME_MAX
;
405 /* Skip reprogram of event if its not changed */
406 if (ts
->tick_stopped
&& ktime_equal(expires
, dev
->next_event
))
412 * nohz_stop_sched_tick can be called several times before
413 * the nohz_restart_sched_tick is called. This happens when
414 * interrupts arrive which do not cause a reschedule. In the
415 * first call we save the current tick time, so we can restart
416 * the scheduler tick in nohz_restart_sched_tick.
418 if (!ts
->tick_stopped
) {
419 nohz_balance_enter_idle(cpu
);
420 calc_load_enter_idle();
422 ts
->last_tick
= hrtimer_get_expires(&ts
->sched_timer
);
423 ts
->tick_stopped
= 1;
427 * If the expiration time == KTIME_MAX, then
428 * in this case we simply stop the tick timer.
430 if (unlikely(expires
.tv64
== KTIME_MAX
)) {
431 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
)
432 hrtimer_cancel(&ts
->sched_timer
);
436 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
) {
437 hrtimer_start(&ts
->sched_timer
, expires
,
438 HRTIMER_MODE_ABS_PINNED
);
439 /* Check, if the timer was already in the past */
440 if (hrtimer_active(&ts
->sched_timer
))
442 } else if (!tick_program_event(expires
, 0))
445 * We are past the event already. So we crossed a
446 * jiffie boundary. Update jiffies and raise the
449 tick_do_update_jiffies64(ktime_get());
451 raise_softirq_irqoff(TIMER_SOFTIRQ
);
453 ts
->next_jiffies
= next_jiffies
;
454 ts
->last_jiffies
= last_jiffies
;
455 ts
->sleep_length
= ktime_sub(dev
->next_event
, now
);
460 static bool can_stop_idle_tick(int cpu
, struct tick_sched
*ts
)
463 * If this cpu is offline and it is the one which updates
464 * jiffies, then give up the assignment and let it be taken by
465 * the cpu which runs the tick timer next. If we don't drop
466 * this here the jiffies might be stale and do_timer() never
469 if (unlikely(!cpu_online(cpu
))) {
470 if (cpu
== tick_do_timer_cpu
)
471 tick_do_timer_cpu
= TICK_DO_TIMER_NONE
;
475 if (unlikely(ts
->nohz_mode
== NOHZ_MODE_INACTIVE
))
481 if (unlikely(local_softirq_pending() && cpu_online(cpu
))) {
482 static int ratelimit
;
484 if (ratelimit
< 10 &&
485 (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK
)) {
486 pr_warn("NOHZ: local_softirq_pending %02x\n",
487 (unsigned int) local_softirq_pending());
496 static void __tick_nohz_idle_enter(struct tick_sched
*ts
)
498 ktime_t now
, expires
;
499 int cpu
= smp_processor_id();
501 now
= tick_nohz_start_idle(cpu
, ts
);
503 if (can_stop_idle_tick(cpu
, ts
)) {
504 int was_stopped
= ts
->tick_stopped
;
508 expires
= tick_nohz_stop_sched_tick(ts
, now
, cpu
);
509 if (expires
.tv64
> 0LL) {
511 ts
->idle_expires
= expires
;
514 if (!was_stopped
&& ts
->tick_stopped
)
515 ts
->idle_jiffies
= ts
->last_jiffies
;
520 * tick_nohz_idle_enter - stop the idle tick from the idle task
522 * When the next event is more than a tick into the future, stop the idle tick
523 * Called when we start the idle loop.
525 * The arch is responsible of calling:
527 * - rcu_idle_enter() after its last use of RCU before the CPU is put
529 * - rcu_idle_exit() before the first use of RCU after the CPU is woken up.
531 void tick_nohz_idle_enter(void)
533 struct tick_sched
*ts
;
535 WARN_ON_ONCE(irqs_disabled());
538 * Update the idle state in the scheduler domain hierarchy
539 * when tick_nohz_stop_sched_tick() is called from the idle loop.
540 * State will be updated to busy during the first busy tick after
543 set_cpu_sd_state_idle();
547 ts
= &__get_cpu_var(tick_cpu_sched
);
549 * set ts->inidle unconditionally. even if the system did not
550 * switch to nohz mode the cpu frequency governers rely on the
551 * update of the idle time accounting in tick_nohz_start_idle().
554 __tick_nohz_idle_enter(ts
);
558 EXPORT_SYMBOL_GPL(tick_nohz_idle_enter
);
561 * tick_nohz_irq_exit - update next tick event from interrupt exit
563 * When an interrupt fires while we are idle and it doesn't cause
564 * a reschedule, it may still add, modify or delete a timer, enqueue
565 * an RCU callback, etc...
566 * So we need to re-calculate and reprogram the next tick event.
568 void tick_nohz_irq_exit(void)
570 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
575 /* Cancel the timer because CPU already waken up from the C-states*/
576 menu_hrtimer_cancel();
577 __tick_nohz_idle_enter(ts
);
581 * tick_nohz_get_sleep_length - return the length of the current sleep
583 * Called from power state control code with interrupts disabled
585 ktime_t
tick_nohz_get_sleep_length(void)
587 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
589 return ts
->sleep_length
;
592 static void tick_nohz_restart(struct tick_sched
*ts
, ktime_t now
)
594 hrtimer_cancel(&ts
->sched_timer
);
595 hrtimer_set_expires(&ts
->sched_timer
, ts
->last_tick
);
598 /* Forward the time to expire in the future */
599 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
601 if (ts
->nohz_mode
== NOHZ_MODE_HIGHRES
) {
602 hrtimer_start_expires(&ts
->sched_timer
,
603 HRTIMER_MODE_ABS_PINNED
);
604 /* Check, if the timer was already in the past */
605 if (hrtimer_active(&ts
->sched_timer
))
608 if (!tick_program_event(
609 hrtimer_get_expires(&ts
->sched_timer
), 0))
612 /* Reread time and update jiffies */
614 tick_do_update_jiffies64(now
);
618 static void tick_nohz_restart_sched_tick(struct tick_sched
*ts
, ktime_t now
)
620 /* Update jiffies first */
621 tick_do_update_jiffies64(now
);
622 update_cpu_load_nohz();
624 calc_load_exit_idle();
625 touch_softlockup_watchdog();
627 * Cancel the scheduled timer and restore the tick
629 ts
->tick_stopped
= 0;
630 ts
->idle_exittime
= now
;
632 tick_nohz_restart(ts
, now
);
635 static void tick_nohz_account_idle_ticks(struct tick_sched
*ts
)
637 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
640 if (vtime_accounting_enabled())
643 * We stopped the tick in idle. Update process times would miss the
644 * time we slept as update_process_times does only a 1 tick
645 * accounting. Enforce that this is accounted to idle !
647 ticks
= jiffies
- ts
->idle_jiffies
;
649 * We might be one off. Do not randomly account a huge number of ticks!
651 if (ticks
&& ticks
< LONG_MAX
)
652 account_idle_ticks(ticks
);
657 * tick_nohz_idle_exit - restart the idle tick from the idle task
659 * Restart the idle tick when the CPU is woken up from idle
660 * This also exit the RCU extended quiescent state. The CPU
661 * can use RCU again after this function is called.
663 void tick_nohz_idle_exit(void)
665 int cpu
= smp_processor_id();
666 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
671 WARN_ON_ONCE(!ts
->inidle
);
675 /* Cancel the timer because CPU already waken up from the C-states*/
676 menu_hrtimer_cancel();
677 if (ts
->idle_active
|| ts
->tick_stopped
)
681 tick_nohz_stop_idle(cpu
, now
);
683 if (ts
->tick_stopped
) {
684 tick_nohz_restart_sched_tick(ts
, now
);
685 tick_nohz_account_idle_ticks(ts
);
690 EXPORT_SYMBOL_GPL(tick_nohz_idle_exit
);
692 static int tick_nohz_reprogram(struct tick_sched
*ts
, ktime_t now
)
694 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
695 return tick_program_event(hrtimer_get_expires(&ts
->sched_timer
), 0);
699 * The nohz low res interrupt handler
701 static void tick_nohz_handler(struct clock_event_device
*dev
)
703 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
704 struct pt_regs
*regs
= get_irq_regs();
705 ktime_t now
= ktime_get();
707 dev
->next_event
.tv64
= KTIME_MAX
;
709 tick_sched_do_timer(now
);
710 tick_sched_handle(ts
, regs
);
712 while (tick_nohz_reprogram(ts
, now
)) {
714 tick_do_update_jiffies64(now
);
719 * tick_nohz_switch_to_nohz - switch to nohz mode
721 static void tick_nohz_switch_to_nohz(void)
723 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
726 if (!tick_nohz_enabled
)
730 if (tick_switch_to_oneshot(tick_nohz_handler
)) {
735 ts
->nohz_mode
= NOHZ_MODE_LOWRES
;
738 * Recycle the hrtimer in ts, so we can share the
739 * hrtimer_forward with the highres code.
741 hrtimer_init(&ts
->sched_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
742 /* Get the next period */
743 next
= tick_init_jiffy_update();
746 hrtimer_set_expires(&ts
->sched_timer
, next
);
747 if (!tick_program_event(next
, 0))
749 next
= ktime_add(next
, tick_period
);
755 * When NOHZ is enabled and the tick is stopped, we need to kick the
756 * tick timer from irq_enter() so that the jiffies update is kept
757 * alive during long running softirqs. That's ugly as hell, but
758 * correctness is key even if we need to fix the offending softirq in
761 * Note, this is different to tick_nohz_restart. We just kick the
762 * timer and do not touch the other magic bits which need to be done
765 static void tick_nohz_kick_tick(int cpu
, ktime_t now
)
768 /* Switch back to 2.6.27 behaviour */
770 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
774 * Do not touch the tick device, when the next expiry is either
775 * already reached or less/equal than the tick period.
777 delta
= ktime_sub(hrtimer_get_expires(&ts
->sched_timer
), now
);
778 if (delta
.tv64
<= tick_period
.tv64
)
781 tick_nohz_restart(ts
, now
);
785 static inline void tick_check_nohz(int cpu
)
787 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
790 if (!ts
->idle_active
&& !ts
->tick_stopped
)
794 tick_nohz_stop_idle(cpu
, now
);
795 if (ts
->tick_stopped
) {
796 tick_nohz_update_jiffies(now
);
797 tick_nohz_kick_tick(cpu
, now
);
803 static inline void tick_nohz_switch_to_nohz(void) { }
804 static inline void tick_check_nohz(int cpu
) { }
809 * Called from irq_enter to notify about the possible interruption of idle()
811 void tick_check_idle(int cpu
)
813 tick_check_oneshot_broadcast(cpu
);
814 tick_check_nohz(cpu
);
818 * High resolution timer specific code
820 #ifdef CONFIG_HIGH_RES_TIMERS
822 * We rearm the timer until we get disabled by the idle code.
823 * Called with interrupts disabled.
825 static enum hrtimer_restart
tick_sched_timer(struct hrtimer
*timer
)
827 struct tick_sched
*ts
=
828 container_of(timer
, struct tick_sched
, sched_timer
);
829 struct pt_regs
*regs
= get_irq_regs();
830 ktime_t now
= ktime_get();
832 tick_sched_do_timer(now
);
835 * Do not call, when we are not in irq context and have
836 * no valid regs pointer
839 tick_sched_handle(ts
, regs
);
841 hrtimer_forward(timer
, now
, tick_period
);
843 return HRTIMER_RESTART
;
846 static int sched_skew_tick
;
848 static int __init
skew_tick(char *str
)
850 get_option(&str
, &sched_skew_tick
);
854 early_param("skew_tick", skew_tick
);
857 * tick_setup_sched_timer - setup the tick emulation timer
859 void tick_setup_sched_timer(void)
861 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
862 ktime_t now
= ktime_get();
865 * Emulate tick processing via per-CPU hrtimers:
867 hrtimer_init(&ts
->sched_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_ABS
);
868 ts
->sched_timer
.function
= tick_sched_timer
;
870 /* Get the next period (per cpu) */
871 hrtimer_set_expires(&ts
->sched_timer
, tick_init_jiffy_update());
873 /* Offset the tick to avert jiffies_lock contention. */
874 if (sched_skew_tick
) {
875 u64 offset
= ktime_to_ns(tick_period
) >> 1;
876 do_div(offset
, num_possible_cpus());
877 offset
*= smp_processor_id();
878 hrtimer_add_expires_ns(&ts
->sched_timer
, offset
);
882 hrtimer_forward(&ts
->sched_timer
, now
, tick_period
);
883 hrtimer_start_expires(&ts
->sched_timer
,
884 HRTIMER_MODE_ABS_PINNED
);
885 /* Check, if the timer was already in the past */
886 if (hrtimer_active(&ts
->sched_timer
))
892 if (tick_nohz_enabled
)
893 ts
->nohz_mode
= NOHZ_MODE_HIGHRES
;
896 #endif /* HIGH_RES_TIMERS */
898 #if defined CONFIG_NO_HZ || defined CONFIG_HIGH_RES_TIMERS
899 void tick_cancel_sched_timer(int cpu
)
901 struct tick_sched
*ts
= &per_cpu(tick_cpu_sched
, cpu
);
903 # ifdef CONFIG_HIGH_RES_TIMERS
904 if (ts
->sched_timer
.base
)
905 hrtimer_cancel(&ts
->sched_timer
);
908 memset(ts
, 0, sizeof(*ts
));
913 * Async notification about clocksource changes
915 void tick_clock_notify(void)
919 for_each_possible_cpu(cpu
)
920 set_bit(0, &per_cpu(tick_cpu_sched
, cpu
).check_clocks
);
924 * Async notification about clock event changes
926 void tick_oneshot_notify(void)
928 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
930 set_bit(0, &ts
->check_clocks
);
934 * Check, if a change happened, which makes oneshot possible.
936 * Called cyclic from the hrtimer softirq (driven by the timer
937 * softirq) allow_nohz signals, that we can switch into low-res nohz
938 * mode, because high resolution timers are disabled (either compile
941 int tick_check_oneshot_change(int allow_nohz
)
943 struct tick_sched
*ts
= &__get_cpu_var(tick_cpu_sched
);
945 if (!test_and_clear_bit(0, &ts
->check_clocks
))
948 if (ts
->nohz_mode
!= NOHZ_MODE_INACTIVE
)
951 if (!timekeeping_valid_for_hres() || !tick_is_oneshot_available())
957 tick_nohz_switch_to_nohz();