4 * Kernel internal timers, kernel timekeeping, basic process system calls
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/tick.h>
39 #include <asm/uaccess.h>
40 #include <asm/unistd.h>
41 #include <asm/div64.h>
42 #include <asm/timex.h>
45 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
47 EXPORT_SYMBOL(jiffies_64);
50 * per-CPU timer vector definitions:
52 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
53 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
54 #define TVN_SIZE (1 << TVN_BITS)
55 #define TVR_SIZE (1 << TVR_BITS)
56 #define TVN_MASK (TVN_SIZE - 1)
57 #define TVR_MASK (TVR_SIZE - 1)
59 typedef struct tvec_s {
60 struct list_head vec[TVN_SIZE];
63 typedef struct tvec_root_s {
64 struct list_head vec[TVR_SIZE];
67 struct tvec_t_base_s {
69 struct timer_list *running_timer;
70 unsigned long timer_jiffies;
76 } ____cacheline_aligned_in_smp;
78 typedef struct tvec_t_base_s tvec_base_t;
80 tvec_base_t boot_tvec_bases;
81 EXPORT_SYMBOL(boot_tvec_bases);
82 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
85 * __round_jiffies - function to round jiffies to a full second
86 * @j: the time in (absolute) jiffies that should be rounded
87 * @cpu: the processor number on which the timeout will happen
89 * __round_jiffies() rounds an absolute time in the future (in jiffies)
90 * up or down to (approximately) full seconds. This is useful for timers
91 * for which the exact time they fire does not matter too much, as long as
92 * they fire approximately every X seconds.
94 * By rounding these timers to whole seconds, all such timers will fire
95 * at the same time, rather than at various times spread out. The goal
96 * of this is to have the CPU wake up less, which saves power.
98 * The exact rounding is skewed for each processor to avoid all
99 * processors firing at the exact same time, which could lead
100 * to lock contention or spurious cache line bouncing.
102 * The return value is the rounded version of the @j parameter.
104 unsigned long __round_jiffies(unsigned long j, int cpu)
107 unsigned long original = j;
110 * We don't want all cpus firing their timers at once hitting the
111 * same lock or cachelines, so we skew each extra cpu with an extra
112 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
114 * The skew is done by adding 3*cpunr, then round, then subtract this
115 * extra offset again.
122 * If the target jiffie is just after a whole second (which can happen
123 * due to delays of the timer irq, long irq off times etc etc) then
124 * we should round down to the whole second, not up. Use 1/4th second
125 * as cutoff for this rounding as an extreme upper bound for this.
127 if (rem < HZ/4) /* round down */
132 /* now that we have rounded, subtract the extra skew again */
135 if (j <= jiffies) /* rounding ate our timeout entirely; */
139 EXPORT_SYMBOL_GPL(__round_jiffies);
142 * __round_jiffies_relative - function to round jiffies to a full second
143 * @j: the time in (relative) jiffies that should be rounded
144 * @cpu: the processor number on which the timeout will happen
146 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
147 * up or down to (approximately) full seconds. This is useful for timers
148 * for which the exact time they fire does not matter too much, as long as
149 * they fire approximately every X seconds.
151 * By rounding these timers to whole seconds, all such timers will fire
152 * at the same time, rather than at various times spread out. The goal
153 * of this is to have the CPU wake up less, which saves power.
155 * The exact rounding is skewed for each processor to avoid all
156 * processors firing at the exact same time, which could lead
157 * to lock contention or spurious cache line bouncing.
159 * The return value is the rounded version of the @j parameter.
161 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
164 * In theory the following code can skip a jiffy in case jiffies
165 * increments right between the addition and the later subtraction.
166 * However since the entire point of this function is to use approximate
167 * timeouts, it's entirely ok to not handle that.
169 return __round_jiffies(j + jiffies, cpu) - jiffies;
171 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
174 * round_jiffies - function to round jiffies to a full second
175 * @j: the time in (absolute) jiffies that should be rounded
177 * round_jiffies() rounds an absolute time in the future (in jiffies)
178 * up or down to (approximately) full seconds. This is useful for timers
179 * for which the exact time they fire does not matter too much, as long as
180 * they fire approximately every X seconds.
182 * By rounding these timers to whole seconds, all such timers will fire
183 * at the same time, rather than at various times spread out. The goal
184 * of this is to have the CPU wake up less, which saves power.
186 * The return value is the rounded version of the @j parameter.
188 unsigned long round_jiffies(unsigned long j)
190 return __round_jiffies(j, raw_smp_processor_id());
192 EXPORT_SYMBOL_GPL(round_jiffies);
195 * round_jiffies_relative - function to round jiffies to a full second
196 * @j: the time in (relative) jiffies that should be rounded
198 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
199 * up or down to (approximately) full seconds. This is useful for timers
200 * for which the exact time they fire does not matter too much, as long as
201 * they fire approximately every X seconds.
203 * By rounding these timers to whole seconds, all such timers will fire
204 * at the same time, rather than at various times spread out. The goal
205 * of this is to have the CPU wake up less, which saves power.
207 * The return value is the rounded version of the @j parameter.
209 unsigned long round_jiffies_relative(unsigned long j)
211 return __round_jiffies_relative(j, raw_smp_processor_id());
213 EXPORT_SYMBOL_GPL(round_jiffies_relative);
216 static inline void set_running_timer(tvec_base_t *base,
217 struct timer_list *timer)
220 base->running_timer = timer;
224 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
226 unsigned long expires = timer->expires;
227 unsigned long idx = expires - base->timer_jiffies;
228 struct list_head *vec;
230 if (idx < TVR_SIZE) {
231 int i = expires & TVR_MASK;
232 vec = base->tv1.vec + i;
233 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
234 int i = (expires >> TVR_BITS) & TVN_MASK;
235 vec = base->tv2.vec + i;
236 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
237 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
238 vec = base->tv3.vec + i;
239 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
240 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
241 vec = base->tv4.vec + i;
242 } else if ((signed long) idx < 0) {
244 * Can happen if you add a timer with expires == jiffies,
245 * or you set a timer to go off in the past
247 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
250 /* If the timeout is larger than 0xffffffff on 64-bit
251 * architectures then we use the maximum timeout:
253 if (idx > 0xffffffffUL) {
255 expires = idx + base->timer_jiffies;
257 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
258 vec = base->tv5.vec + i;
263 list_add_tail(&timer->entry, vec);
267 * init_timer - initialize a timer.
268 * @timer: the timer to be initialized
270 * init_timer() must be done to a timer prior calling *any* of the
271 * other timer functions.
273 void fastcall init_timer(struct timer_list *timer)
275 timer->entry.next = NULL;
276 timer->base = __raw_get_cpu_var(tvec_bases);
278 EXPORT_SYMBOL(init_timer);
280 static inline void detach_timer(struct timer_list *timer,
283 struct list_head *entry = &timer->entry;
285 __list_del(entry->prev, entry->next);
288 entry->prev = LIST_POISON2;
292 * We are using hashed locking: holding per_cpu(tvec_bases).lock
293 * means that all timers which are tied to this base via timer->base are
294 * locked, and the base itself is locked too.
296 * So __run_timers/migrate_timers can safely modify all timers which could
297 * be found on ->tvX lists.
299 * When the timer's base is locked, and the timer removed from list, it is
300 * possible to set timer->base = NULL and drop the lock: the timer remains
303 static tvec_base_t *lock_timer_base(struct timer_list *timer,
304 unsigned long *flags)
305 __acquires(timer->base->lock)
311 if (likely(base != NULL)) {
312 spin_lock_irqsave(&base->lock, *flags);
313 if (likely(base == timer->base))
315 /* The timer has migrated to another CPU */
316 spin_unlock_irqrestore(&base->lock, *flags);
322 int __mod_timer(struct timer_list *timer, unsigned long expires)
324 tvec_base_t *base, *new_base;
328 BUG_ON(!timer->function);
330 base = lock_timer_base(timer, &flags);
332 if (timer_pending(timer)) {
333 detach_timer(timer, 0);
337 new_base = __get_cpu_var(tvec_bases);
339 if (base != new_base) {
341 * We are trying to schedule the timer on the local CPU.
342 * However we can't change timer's base while it is running,
343 * otherwise del_timer_sync() can't detect that the timer's
344 * handler yet has not finished. This also guarantees that
345 * the timer is serialized wrt itself.
347 if (likely(base->running_timer != timer)) {
348 /* See the comment in lock_timer_base() */
350 spin_unlock(&base->lock);
352 spin_lock(&base->lock);
357 timer->expires = expires;
358 internal_add_timer(base, timer);
359 spin_unlock_irqrestore(&base->lock, flags);
364 EXPORT_SYMBOL(__mod_timer);
367 * add_timer_on - start a timer on a particular CPU
368 * @timer: the timer to be added
369 * @cpu: the CPU to start it on
371 * This is not very scalable on SMP. Double adds are not possible.
373 void add_timer_on(struct timer_list *timer, int cpu)
375 tvec_base_t *base = per_cpu(tvec_bases, cpu);
378 BUG_ON(timer_pending(timer) || !timer->function);
379 spin_lock_irqsave(&base->lock, flags);
381 internal_add_timer(base, timer);
382 spin_unlock_irqrestore(&base->lock, flags);
387 * mod_timer - modify a timer's timeout
388 * @timer: the timer to be modified
389 * @expires: new timeout in jiffies
391 * mod_timer() is a more efficient way to update the expire field of an
392 * active timer (if the timer is inactive it will be activated)
394 * mod_timer(timer, expires) is equivalent to:
396 * del_timer(timer); timer->expires = expires; add_timer(timer);
398 * Note that if there are multiple unserialized concurrent users of the
399 * same timer, then mod_timer() is the only safe way to modify the timeout,
400 * since add_timer() cannot modify an already running timer.
402 * The function returns whether it has modified a pending timer or not.
403 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
404 * active timer returns 1.)
406 int mod_timer(struct timer_list *timer, unsigned long expires)
408 BUG_ON(!timer->function);
411 * This is a common optimization triggered by the
412 * networking code - if the timer is re-modified
413 * to be the same thing then just return:
415 if (timer->expires == expires && timer_pending(timer))
418 return __mod_timer(timer, expires);
421 EXPORT_SYMBOL(mod_timer);
424 * del_timer - deactive a timer.
425 * @timer: the timer to be deactivated
427 * del_timer() deactivates a timer - this works on both active and inactive
430 * The function returns whether it has deactivated a pending timer or not.
431 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
432 * active timer returns 1.)
434 int del_timer(struct timer_list *timer)
440 if (timer_pending(timer)) {
441 base = lock_timer_base(timer, &flags);
442 if (timer_pending(timer)) {
443 detach_timer(timer, 1);
446 spin_unlock_irqrestore(&base->lock, flags);
452 EXPORT_SYMBOL(del_timer);
456 * try_to_del_timer_sync - Try to deactivate a timer
457 * @timer: timer do del
459 * This function tries to deactivate a timer. Upon successful (ret >= 0)
460 * exit the timer is not queued and the handler is not running on any CPU.
462 * It must not be called from interrupt contexts.
464 int try_to_del_timer_sync(struct timer_list *timer)
470 base = lock_timer_base(timer, &flags);
472 if (base->running_timer == timer)
476 if (timer_pending(timer)) {
477 detach_timer(timer, 1);
481 spin_unlock_irqrestore(&base->lock, flags);
487 * del_timer_sync - deactivate a timer and wait for the handler to finish.
488 * @timer: the timer to be deactivated
490 * This function only differs from del_timer() on SMP: besides deactivating
491 * the timer it also makes sure the handler has finished executing on other
494 * Synchronization rules: Callers must prevent restarting of the timer,
495 * otherwise this function is meaningless. It must not be called from
496 * interrupt contexts. The caller must not hold locks which would prevent
497 * completion of the timer's handler. The timer's handler must not call
498 * add_timer_on(). Upon exit the timer is not queued and the handler is
499 * not running on any CPU.
501 * The function returns whether it has deactivated a pending timer or not.
503 int del_timer_sync(struct timer_list *timer)
506 int ret = try_to_del_timer_sync(timer);
513 EXPORT_SYMBOL(del_timer_sync);
516 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
518 /* cascade all the timers from tv up one level */
519 struct timer_list *timer, *tmp;
520 struct list_head tv_list;
522 list_replace_init(tv->vec + index, &tv_list);
525 * We are removing _all_ timers from the list, so we
526 * don't have to detach them individually.
528 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
529 BUG_ON(timer->base != base);
530 internal_add_timer(base, timer);
536 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
539 * __run_timers - run all expired timers (if any) on this CPU.
540 * @base: the timer vector to be processed.
542 * This function cascades all vectors and executes all expired timer
545 static inline void __run_timers(tvec_base_t *base)
547 struct timer_list *timer;
549 spin_lock_irq(&base->lock);
550 while (time_after_eq(jiffies, base->timer_jiffies)) {
551 struct list_head work_list;
552 struct list_head *head = &work_list;
553 int index = base->timer_jiffies & TVR_MASK;
559 (!cascade(base, &base->tv2, INDEX(0))) &&
560 (!cascade(base, &base->tv3, INDEX(1))) &&
561 !cascade(base, &base->tv4, INDEX(2)))
562 cascade(base, &base->tv5, INDEX(3));
563 ++base->timer_jiffies;
564 list_replace_init(base->tv1.vec + index, &work_list);
565 while (!list_empty(head)) {
566 void (*fn)(unsigned long);
569 timer = list_entry(head->next,struct timer_list,entry);
570 fn = timer->function;
573 set_running_timer(base, timer);
574 detach_timer(timer, 1);
575 spin_unlock_irq(&base->lock);
577 int preempt_count = preempt_count();
579 if (preempt_count != preempt_count()) {
580 printk(KERN_WARNING "huh, entered %p "
581 "with preempt_count %08x, exited"
588 spin_lock_irq(&base->lock);
591 set_running_timer(base, NULL);
592 spin_unlock_irq(&base->lock);
595 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
597 * Find out when the next timer event is due to happen. This
598 * is used on S/390 to stop all activity when a cpus is idle.
599 * This functions needs to be called disabled.
601 static unsigned long __next_timer_interrupt(tvec_base_t *base)
603 unsigned long timer_jiffies = base->timer_jiffies;
604 unsigned long expires = timer_jiffies + (LONG_MAX >> 1);
605 int index, slot, array, found = 0;
606 struct timer_list *nte;
609 /* Look for timer events in tv1. */
610 index = slot = timer_jiffies & TVR_MASK;
612 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
614 expires = nte->expires;
615 /* Look at the cascade bucket(s)? */
616 if (!index || slot < index)
620 slot = (slot + 1) & TVR_MASK;
621 } while (slot != index);
624 /* Calculate the next cascade event */
626 timer_jiffies += TVR_SIZE - index;
627 timer_jiffies >>= TVR_BITS;
630 varray[0] = &base->tv2;
631 varray[1] = &base->tv3;
632 varray[2] = &base->tv4;
633 varray[3] = &base->tv5;
635 for (array = 0; array < 4; array++) {
636 tvec_t *varp = varray[array];
638 index = slot = timer_jiffies & TVN_MASK;
640 list_for_each_entry(nte, varp->vec + slot, entry) {
642 if (time_before(nte->expires, expires))
643 expires = nte->expires;
646 * Do we still search for the first timer or are
647 * we looking up the cascade buckets ?
650 /* Look at the cascade bucket(s)? */
651 if (!index || slot < index)
655 slot = (slot + 1) & TVN_MASK;
656 } while (slot != index);
659 timer_jiffies += TVN_SIZE - index;
660 timer_jiffies >>= TVN_BITS;
666 * Check, if the next hrtimer event is before the next timer wheel
669 static unsigned long cmp_next_hrtimer_event(unsigned long now,
670 unsigned long expires)
672 ktime_t hr_delta = hrtimer_get_next_event();
673 struct timespec tsdelta;
675 if (hr_delta.tv64 == KTIME_MAX)
678 if (hr_delta.tv64 <= TICK_NSEC)
681 tsdelta = ktime_to_timespec(hr_delta);
682 now += timespec_to_jiffies(&tsdelta);
683 if (time_before(now, expires))
689 * next_timer_interrupt - return the jiffy of the next pending timer
691 unsigned long get_next_timer_interrupt(unsigned long now)
693 tvec_base_t *base = __get_cpu_var(tvec_bases);
694 unsigned long expires;
696 spin_lock(&base->lock);
697 expires = __next_timer_interrupt(base);
698 spin_unlock(&base->lock);
700 if (time_before_eq(expires, now))
703 return cmp_next_hrtimer_event(now, expires);
706 #ifdef CONFIG_NO_IDLE_HZ
707 unsigned long next_timer_interrupt(void)
709 return get_next_timer_interrupt(jiffies);
715 /******************************************************************/
719 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
720 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
721 * at zero at system boot time, so wall_to_monotonic will be negative,
722 * however, we will ALWAYS keep the tv_nsec part positive so we can use
723 * the usual normalization.
725 struct timespec xtime __attribute__ ((aligned (16)));
726 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
728 EXPORT_SYMBOL(xtime);
731 /* XXX - all of this timekeeping code should be later moved to time.c */
732 #include <linux/clocksource.h>
733 static struct clocksource *clock; /* pointer to current clocksource */
735 #ifdef CONFIG_GENERIC_TIME
737 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
739 * private function, must hold xtime_lock lock when being
740 * called. Returns the number of nanoseconds since the
741 * last call to update_wall_time() (adjusted by NTP scaling)
743 static inline s64 __get_nsec_offset(void)
745 cycle_t cycle_now, cycle_delta;
748 /* read clocksource: */
749 cycle_now = clocksource_read(clock);
751 /* calculate the delta since the last update_wall_time: */
752 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
754 /* convert to nanoseconds: */
755 ns_offset = cyc2ns(clock, cycle_delta);
761 * __get_realtime_clock_ts - Returns the time of day in a timespec
762 * @ts: pointer to the timespec to be set
764 * Returns the time of day in a timespec. Used by
765 * do_gettimeofday() and get_realtime_clock_ts().
767 static inline void __get_realtime_clock_ts(struct timespec *ts)
773 seq = read_seqbegin(&xtime_lock);
776 nsecs = __get_nsec_offset();
778 } while (read_seqretry(&xtime_lock, seq));
780 timespec_add_ns(ts, nsecs);
784 * getnstimeofday - Returns the time of day in a timespec
785 * @ts: pointer to the timespec to be set
787 * Returns the time of day in a timespec.
789 void getnstimeofday(struct timespec *ts)
791 __get_realtime_clock_ts(ts);
794 EXPORT_SYMBOL(getnstimeofday);
797 * do_gettimeofday - Returns the time of day in a timeval
798 * @tv: pointer to the timeval to be set
800 * NOTE: Users should be converted to using get_realtime_clock_ts()
802 void do_gettimeofday(struct timeval *tv)
806 __get_realtime_clock_ts(&now);
807 tv->tv_sec = now.tv_sec;
808 tv->tv_usec = now.tv_nsec/1000;
811 EXPORT_SYMBOL(do_gettimeofday);
813 * do_settimeofday - Sets the time of day
814 * @tv: pointer to the timespec variable containing the new time
816 * Sets the time of day to the new time and update NTP and notify hrtimers
818 int do_settimeofday(struct timespec *tv)
821 time_t wtm_sec, sec = tv->tv_sec;
822 long wtm_nsec, nsec = tv->tv_nsec;
824 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
827 write_seqlock_irqsave(&xtime_lock, flags);
829 nsec -= __get_nsec_offset();
831 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
832 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
834 set_normalized_timespec(&xtime, sec, nsec);
835 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
840 write_sequnlock_irqrestore(&xtime_lock, flags);
842 /* signal hrtimers about time change */
848 EXPORT_SYMBOL(do_settimeofday);
851 * change_clocksource - Swaps clocksources if a new one is available
853 * Accumulates current time interval and initializes new clocksource
855 static void change_clocksource(void)
857 struct clocksource *new;
861 new = clocksource_get_next();
866 now = clocksource_read(new);
867 nsec = __get_nsec_offset();
868 timespec_add_ns(&xtime, nsec);
871 clock->cycle_last = now;
874 clock->xtime_nsec = 0;
875 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
879 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
883 static inline void change_clocksource(void) { }
887 * timeofday_is_continuous - check to see if timekeeping is free running
889 int timekeeping_is_continuous(void)
895 seq = read_seqbegin(&xtime_lock);
897 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
899 } while (read_seqretry(&xtime_lock, seq));
905 * read_persistent_clock - Return time in seconds from the persistent clock.
907 * Weak dummy function for arches that do not yet support it.
908 * Returns seconds from epoch using the battery backed persistent clock.
909 * Returns zero if unsupported.
911 * XXX - Do be sure to remove it once all arches implement it.
913 unsigned long __attribute__((weak)) read_persistent_clock(void)
919 * timekeeping_init - Initializes the clocksource and common timekeeping values
921 void __init timekeeping_init(void)
924 unsigned long sec = read_persistent_clock();
926 write_seqlock_irqsave(&xtime_lock, flags);
930 clock = clocksource_get_next();
931 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
932 clock->cycle_last = clocksource_read(clock);
936 set_normalized_timespec(&wall_to_monotonic,
937 -xtime.tv_sec, -xtime.tv_nsec);
939 write_sequnlock_irqrestore(&xtime_lock, flags);
942 /* flag for if timekeeping is suspended */
943 static int timekeeping_suspended;
944 /* time in seconds when suspend began */
945 static unsigned long timekeeping_suspend_time;
948 * timekeeping_resume - Resumes the generic timekeeping subsystem.
951 * This is for the generic clocksource timekeeping.
952 * xtime/wall_to_monotonic/jiffies/etc are
953 * still managed by arch specific suspend/resume code.
955 static int timekeeping_resume(struct sys_device *dev)
958 unsigned long now = read_persistent_clock();
960 write_seqlock_irqsave(&xtime_lock, flags);
962 if (now && (now > timekeeping_suspend_time)) {
963 unsigned long sleep_length = now - timekeeping_suspend_time;
965 xtime.tv_sec += sleep_length;
966 wall_to_monotonic.tv_sec -= sleep_length;
968 /* re-base the last cycle value */
969 clock->cycle_last = clocksource_read(clock);
971 timekeeping_suspended = 0;
972 write_sequnlock_irqrestore(&xtime_lock, flags);
974 touch_softlockup_watchdog();
975 /* Resume hrtimers */
981 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
985 write_seqlock_irqsave(&xtime_lock, flags);
986 timekeeping_suspended = 1;
987 timekeeping_suspend_time = read_persistent_clock();
988 write_sequnlock_irqrestore(&xtime_lock, flags);
992 /* sysfs resume/suspend bits for timekeeping */
993 static struct sysdev_class timekeeping_sysclass = {
994 .resume = timekeeping_resume,
995 .suspend = timekeeping_suspend,
996 set_kset_name("timekeeping"),
999 static struct sys_device device_timer = {
1001 .cls = &timekeeping_sysclass,
1004 static int __init timekeeping_init_device(void)
1006 int error = sysdev_class_register(&timekeeping_sysclass);
1008 error = sysdev_register(&device_timer);
1012 device_initcall(timekeeping_init_device);
1015 * If the error is already larger, we look ahead even further
1016 * to compensate for late or lost adjustments.
1018 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
1022 u32 look_ahead, adj;
1026 * Use the current error value to determine how much to look ahead.
1027 * The larger the error the slower we adjust for it to avoid problems
1028 * with losing too many ticks, otherwise we would overadjust and
1029 * produce an even larger error. The smaller the adjustment the
1030 * faster we try to adjust for it, as lost ticks can do less harm
1031 * here. This is tuned so that an error of about 1 msec is adusted
1032 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1034 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1035 error2 = abs(error2);
1036 for (look_ahead = 0; error2 > 0; look_ahead++)
1040 * Now calculate the error in (1 << look_ahead) ticks, but first
1041 * remove the single look ahead already included in the error.
1043 tick_error = current_tick_length() >>
1044 (TICK_LENGTH_SHIFT - clock->shift + 1);
1045 tick_error -= clock->xtime_interval >> 1;
1046 error = ((error - tick_error) >> look_ahead) + tick_error;
1048 /* Finally calculate the adjustment shift value. */
1053 *interval = -*interval;
1057 for (adj = 0; error > i; adj++)
1066 * Adjust the multiplier to reduce the error value,
1067 * this is optimized for the most common adjustments of -1,0,1,
1068 * for other values we can do a bit more work.
1070 static void clocksource_adjust(struct clocksource *clock, s64 offset)
1072 s64 error, interval = clock->cycle_interval;
1075 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1076 if (error > interval) {
1078 if (likely(error <= interval))
1081 adj = clocksource_bigadjust(error, &interval, &offset);
1082 } else if (error < -interval) {
1084 if (likely(error >= -interval)) {
1086 interval = -interval;
1089 adj = clocksource_bigadjust(error, &interval, &offset);
1094 clock->xtime_interval += interval;
1095 clock->xtime_nsec -= offset;
1096 clock->error -= (interval - offset) <<
1097 (TICK_LENGTH_SHIFT - clock->shift);
1101 * update_wall_time - Uses the current clocksource to increment the wall time
1103 * Called from the timer interrupt, must hold a write on xtime_lock.
1105 static void update_wall_time(void)
1109 /* Make sure we're fully resumed: */
1110 if (unlikely(timekeeping_suspended))
1113 #ifdef CONFIG_GENERIC_TIME
1114 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1116 offset = clock->cycle_interval;
1118 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1120 /* normally this loop will run just once, however in the
1121 * case of lost or late ticks, it will accumulate correctly.
1123 while (offset >= clock->cycle_interval) {
1124 /* accumulate one interval */
1125 clock->xtime_nsec += clock->xtime_interval;
1126 clock->cycle_last += clock->cycle_interval;
1127 offset -= clock->cycle_interval;
1129 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1130 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1135 /* interpolator bits */
1136 time_interpolator_update(clock->xtime_interval
1139 /* accumulate error between NTP and clock interval */
1140 clock->error += current_tick_length();
1141 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1144 /* correct the clock when NTP error is too big */
1145 clocksource_adjust(clock, offset);
1147 /* store full nanoseconds into xtime */
1148 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1149 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1151 /* check to see if there is a new clocksource to use */
1152 change_clocksource();
1156 * Called from the timer interrupt handler to charge one tick to the current
1157 * process. user_tick is 1 if the tick is user time, 0 for system.
1159 void update_process_times(int user_tick)
1161 struct task_struct *p = current;
1162 int cpu = smp_processor_id();
1164 /* Note: this timer irq context must be accounted for as well. */
1166 account_user_time(p, jiffies_to_cputime(1));
1168 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1170 if (rcu_pending(cpu))
1171 rcu_check_callbacks(cpu, user_tick);
1173 run_posix_cpu_timers(p);
1177 * Nr of active tasks - counted in fixed-point numbers
1179 static unsigned long count_active_tasks(void)
1181 return nr_active() * FIXED_1;
1185 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1186 * imply that avenrun[] is the standard name for this kind of thing.
1187 * Nothing else seems to be standardized: the fractional size etc
1188 * all seem to differ on different machines.
1190 * Requires xtime_lock to access.
1192 unsigned long avenrun[3];
1194 EXPORT_SYMBOL(avenrun);
1197 * calc_load - given tick count, update the avenrun load estimates.
1198 * This is called while holding a write_lock on xtime_lock.
1200 static inline void calc_load(unsigned long ticks)
1202 unsigned long active_tasks; /* fixed-point */
1203 static int count = LOAD_FREQ;
1206 if (unlikely(count < 0)) {
1207 active_tasks = count_active_tasks();
1209 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1210 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1211 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1213 } while (count < 0);
1218 * This read-write spinlock protects us from races in SMP while
1219 * playing with xtime and avenrun.
1221 __attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1223 EXPORT_SYMBOL(xtime_lock);
1226 * This function runs timers and the timer-tq in bottom half context.
1228 static void run_timer_softirq(struct softirq_action *h)
1230 tvec_base_t *base = __get_cpu_var(tvec_bases);
1232 hrtimer_run_queues();
1233 if (time_after_eq(jiffies, base->timer_jiffies))
1238 * Called by the local, per-CPU timer interrupt on SMP.
1240 void run_local_timers(void)
1242 raise_softirq(TIMER_SOFTIRQ);
1247 * Called by the timer interrupt. xtime_lock must already be taken
1250 static inline void update_times(unsigned long ticks)
1257 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1258 * without sampling the sequence number in xtime_lock.
1259 * jiffies is defined in the linker script...
1262 void do_timer(unsigned long ticks)
1264 jiffies_64 += ticks;
1265 update_times(ticks);
1268 #ifdef __ARCH_WANT_SYS_ALARM
1271 * For backwards compatibility? This can be done in libc so Alpha
1272 * and all newer ports shouldn't need it.
1274 asmlinkage unsigned long sys_alarm(unsigned int seconds)
1276 return alarm_setitimer(seconds);
1284 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1285 * should be moved into arch/i386 instead?
1289 * sys_getpid - return the thread group id of the current process
1291 * Note, despite the name, this returns the tgid not the pid. The tgid and
1292 * the pid are identical unless CLONE_THREAD was specified on clone() in
1293 * which case the tgid is the same in all threads of the same group.
1295 * This is SMP safe as current->tgid does not change.
1297 asmlinkage long sys_getpid(void)
1299 return current->tgid;
1303 * Accessing ->real_parent is not SMP-safe, it could
1304 * change from under us. However, we can use a stale
1305 * value of ->real_parent under rcu_read_lock(), see
1306 * release_task()->call_rcu(delayed_put_task_struct).
1308 asmlinkage long sys_getppid(void)
1313 pid = rcu_dereference(current->real_parent)->tgid;
1319 asmlinkage long sys_getuid(void)
1321 /* Only we change this so SMP safe */
1322 return current->uid;
1325 asmlinkage long sys_geteuid(void)
1327 /* Only we change this so SMP safe */
1328 return current->euid;
1331 asmlinkage long sys_getgid(void)
1333 /* Only we change this so SMP safe */
1334 return current->gid;
1337 asmlinkage long sys_getegid(void)
1339 /* Only we change this so SMP safe */
1340 return current->egid;
1345 static void process_timeout(unsigned long __data)
1347 wake_up_process((struct task_struct *)__data);
1351 * schedule_timeout - sleep until timeout
1352 * @timeout: timeout value in jiffies
1354 * Make the current task sleep until @timeout jiffies have
1355 * elapsed. The routine will return immediately unless
1356 * the current task state has been set (see set_current_state()).
1358 * You can set the task state as follows -
1360 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1361 * pass before the routine returns. The routine will return 0
1363 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1364 * delivered to the current task. In this case the remaining time
1365 * in jiffies will be returned, or 0 if the timer expired in time
1367 * The current task state is guaranteed to be TASK_RUNNING when this
1370 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1371 * the CPU away without a bound on the timeout. In this case the return
1372 * value will be %MAX_SCHEDULE_TIMEOUT.
1374 * In all cases the return value is guaranteed to be non-negative.
1376 fastcall signed long __sched schedule_timeout(signed long timeout)
1378 struct timer_list timer;
1379 unsigned long expire;
1383 case MAX_SCHEDULE_TIMEOUT:
1385 * These two special cases are useful to be comfortable
1386 * in the caller. Nothing more. We could take
1387 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1388 * but I' d like to return a valid offset (>=0) to allow
1389 * the caller to do everything it want with the retval.
1395 * Another bit of PARANOID. Note that the retval will be
1396 * 0 since no piece of kernel is supposed to do a check
1397 * for a negative retval of schedule_timeout() (since it
1398 * should never happens anyway). You just have the printk()
1399 * that will tell you if something is gone wrong and where.
1402 printk(KERN_ERR "schedule_timeout: wrong timeout "
1403 "value %lx\n", timeout);
1405 current->state = TASK_RUNNING;
1410 expire = timeout + jiffies;
1412 setup_timer(&timer, process_timeout, (unsigned long)current);
1413 __mod_timer(&timer, expire);
1415 del_singleshot_timer_sync(&timer);
1417 timeout = expire - jiffies;
1420 return timeout < 0 ? 0 : timeout;
1422 EXPORT_SYMBOL(schedule_timeout);
1425 * We can use __set_current_state() here because schedule_timeout() calls
1426 * schedule() unconditionally.
1428 signed long __sched schedule_timeout_interruptible(signed long timeout)
1430 __set_current_state(TASK_INTERRUPTIBLE);
1431 return schedule_timeout(timeout);
1433 EXPORT_SYMBOL(schedule_timeout_interruptible);
1435 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1437 __set_current_state(TASK_UNINTERRUPTIBLE);
1438 return schedule_timeout(timeout);
1440 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1442 /* Thread ID - the internal kernel "pid" */
1443 asmlinkage long sys_gettid(void)
1445 return current->pid;
1449 * do_sysinfo - fill in sysinfo struct
1450 * @info: pointer to buffer to fill
1452 int do_sysinfo(struct sysinfo *info)
1454 unsigned long mem_total, sav_total;
1455 unsigned int mem_unit, bitcount;
1458 memset(info, 0, sizeof(struct sysinfo));
1462 seq = read_seqbegin(&xtime_lock);
1465 * This is annoying. The below is the same thing
1466 * posix_get_clock_monotonic() does, but it wants to
1467 * take the lock which we want to cover the loads stuff
1471 getnstimeofday(&tp);
1472 tp.tv_sec += wall_to_monotonic.tv_sec;
1473 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1474 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1475 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1478 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1480 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1481 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1482 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1484 info->procs = nr_threads;
1485 } while (read_seqretry(&xtime_lock, seq));
1491 * If the sum of all the available memory (i.e. ram + swap)
1492 * is less than can be stored in a 32 bit unsigned long then
1493 * we can be binary compatible with 2.2.x kernels. If not,
1494 * well, in that case 2.2.x was broken anyways...
1496 * -Erik Andersen <andersee@debian.org>
1499 mem_total = info->totalram + info->totalswap;
1500 if (mem_total < info->totalram || mem_total < info->totalswap)
1503 mem_unit = info->mem_unit;
1504 while (mem_unit > 1) {
1507 sav_total = mem_total;
1509 if (mem_total < sav_total)
1514 * If mem_total did not overflow, multiply all memory values by
1515 * info->mem_unit and set it to 1. This leaves things compatible
1516 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1521 info->totalram <<= bitcount;
1522 info->freeram <<= bitcount;
1523 info->sharedram <<= bitcount;
1524 info->bufferram <<= bitcount;
1525 info->totalswap <<= bitcount;
1526 info->freeswap <<= bitcount;
1527 info->totalhigh <<= bitcount;
1528 info->freehigh <<= bitcount;
1534 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1540 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1547 * lockdep: we want to track each per-CPU base as a separate lock-class,
1548 * but timer-bases are kmalloc()-ed, so we need to attach separate
1551 static struct lock_class_key base_lock_keys[NR_CPUS];
1553 static int __devinit init_timers_cpu(int cpu)
1557 static char __devinitdata tvec_base_done[NR_CPUS];
1559 if (!tvec_base_done[cpu]) {
1560 static char boot_done;
1564 * The APs use this path later in boot
1566 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1570 memset(base, 0, sizeof(*base));
1571 per_cpu(tvec_bases, cpu) = base;
1574 * This is for the boot CPU - we use compile-time
1575 * static initialisation because per-cpu memory isn't
1576 * ready yet and because the memory allocators are not
1577 * initialised either.
1580 base = &boot_tvec_bases;
1582 tvec_base_done[cpu] = 1;
1584 base = per_cpu(tvec_bases, cpu);
1587 spin_lock_init(&base->lock);
1588 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1590 for (j = 0; j < TVN_SIZE; j++) {
1591 INIT_LIST_HEAD(base->tv5.vec + j);
1592 INIT_LIST_HEAD(base->tv4.vec + j);
1593 INIT_LIST_HEAD(base->tv3.vec + j);
1594 INIT_LIST_HEAD(base->tv2.vec + j);
1596 for (j = 0; j < TVR_SIZE; j++)
1597 INIT_LIST_HEAD(base->tv1.vec + j);
1599 base->timer_jiffies = jiffies;
1603 #ifdef CONFIG_HOTPLUG_CPU
1604 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1606 struct timer_list *timer;
1608 while (!list_empty(head)) {
1609 timer = list_entry(head->next, struct timer_list, entry);
1610 detach_timer(timer, 0);
1611 timer->base = new_base;
1612 internal_add_timer(new_base, timer);
1616 static void __devinit migrate_timers(int cpu)
1618 tvec_base_t *old_base;
1619 tvec_base_t *new_base;
1622 BUG_ON(cpu_online(cpu));
1623 old_base = per_cpu(tvec_bases, cpu);
1624 new_base = get_cpu_var(tvec_bases);
1626 local_irq_disable();
1627 spin_lock(&new_base->lock);
1628 spin_lock(&old_base->lock);
1630 BUG_ON(old_base->running_timer);
1632 for (i = 0; i < TVR_SIZE; i++)
1633 migrate_timer_list(new_base, old_base->tv1.vec + i);
1634 for (i = 0; i < TVN_SIZE; i++) {
1635 migrate_timer_list(new_base, old_base->tv2.vec + i);
1636 migrate_timer_list(new_base, old_base->tv3.vec + i);
1637 migrate_timer_list(new_base, old_base->tv4.vec + i);
1638 migrate_timer_list(new_base, old_base->tv5.vec + i);
1641 spin_unlock(&old_base->lock);
1642 spin_unlock(&new_base->lock);
1644 put_cpu_var(tvec_bases);
1646 #endif /* CONFIG_HOTPLUG_CPU */
1648 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1649 unsigned long action, void *hcpu)
1651 long cpu = (long)hcpu;
1653 case CPU_UP_PREPARE:
1654 if (init_timers_cpu(cpu) < 0)
1657 #ifdef CONFIG_HOTPLUG_CPU
1659 migrate_timers(cpu);
1668 static struct notifier_block __cpuinitdata timers_nb = {
1669 .notifier_call = timer_cpu_notify,
1673 void __init init_timers(void)
1675 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1676 (void *)(long)smp_processor_id());
1678 BUG_ON(err == NOTIFY_BAD);
1679 register_cpu_notifier(&timers_nb);
1680 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1683 #ifdef CONFIG_TIME_INTERPOLATION
1685 struct time_interpolator *time_interpolator __read_mostly;
1686 static struct time_interpolator *time_interpolator_list __read_mostly;
1687 static DEFINE_SPINLOCK(time_interpolator_lock);
1689 static inline cycles_t time_interpolator_get_cycles(unsigned int src)
1691 unsigned long (*x)(void);
1695 case TIME_SOURCE_FUNCTION:
1696 x = time_interpolator->addr;
1699 case TIME_SOURCE_MMIO64 :
1700 return readq_relaxed((void __iomem *)time_interpolator->addr);
1702 case TIME_SOURCE_MMIO32 :
1703 return readl_relaxed((void __iomem *)time_interpolator->addr);
1705 default: return get_cycles();
1709 static inline u64 time_interpolator_get_counter(int writelock)
1711 unsigned int src = time_interpolator->source;
1713 if (time_interpolator->jitter)
1719 lcycle = time_interpolator->last_cycle;
1720 now = time_interpolator_get_cycles(src);
1721 if (lcycle && time_after(lcycle, now))
1724 /* When holding the xtime write lock, there's no need
1725 * to add the overhead of the cmpxchg. Readers are
1726 * force to retry until the write lock is released.
1729 time_interpolator->last_cycle = now;
1732 /* Keep track of the last timer value returned. The use of cmpxchg here
1733 * will cause contention in an SMP environment.
1735 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1739 return time_interpolator_get_cycles(src);
1742 void time_interpolator_reset(void)
1744 time_interpolator->offset = 0;
1745 time_interpolator->last_counter = time_interpolator_get_counter(1);
1748 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1750 unsigned long time_interpolator_get_offset(void)
1752 /* If we do not have a time interpolator set up then just return zero */
1753 if (!time_interpolator)
1756 return time_interpolator->offset +
1757 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1760 #define INTERPOLATOR_ADJUST 65536
1761 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1763 void time_interpolator_update(long delta_nsec)
1766 unsigned long offset;
1768 /* If there is no time interpolator set up then do nothing */
1769 if (!time_interpolator)
1773 * The interpolator compensates for late ticks by accumulating the late
1774 * time in time_interpolator->offset. A tick earlier than expected will
1775 * lead to a reset of the offset and a corresponding jump of the clock
1776 * forward. Again this only works if the interpolator clock is running
1777 * slightly slower than the regular clock and the tuning logic insures
1781 counter = time_interpolator_get_counter(1);
1782 offset = time_interpolator->offset +
1783 GET_TI_NSECS(counter, time_interpolator);
1785 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1786 time_interpolator->offset = offset - delta_nsec;
1788 time_interpolator->skips++;
1789 time_interpolator->ns_skipped += delta_nsec - offset;
1790 time_interpolator->offset = 0;
1792 time_interpolator->last_counter = counter;
1794 /* Tuning logic for time interpolator invoked every minute or so.
1795 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1796 * Increase interpolator clock speed if we skip too much time.
1798 if (jiffies % INTERPOLATOR_ADJUST == 0)
1800 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1801 time_interpolator->nsec_per_cyc--;
1802 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1803 time_interpolator->nsec_per_cyc++;
1804 time_interpolator->skips = 0;
1805 time_interpolator->ns_skipped = 0;
1810 is_better_time_interpolator(struct time_interpolator *new)
1812 if (!time_interpolator)
1814 return new->frequency > 2*time_interpolator->frequency ||
1815 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1819 register_time_interpolator(struct time_interpolator *ti)
1821 unsigned long flags;
1824 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1826 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1827 spin_lock(&time_interpolator_lock);
1828 write_seqlock_irqsave(&xtime_lock, flags);
1829 if (is_better_time_interpolator(ti)) {
1830 time_interpolator = ti;
1831 time_interpolator_reset();
1833 write_sequnlock_irqrestore(&xtime_lock, flags);
1835 ti->next = time_interpolator_list;
1836 time_interpolator_list = ti;
1837 spin_unlock(&time_interpolator_lock);
1841 unregister_time_interpolator(struct time_interpolator *ti)
1843 struct time_interpolator *curr, **prev;
1844 unsigned long flags;
1846 spin_lock(&time_interpolator_lock);
1847 prev = &time_interpolator_list;
1848 for (curr = *prev; curr; curr = curr->next) {
1856 write_seqlock_irqsave(&xtime_lock, flags);
1857 if (ti == time_interpolator) {
1858 /* we lost the best time-interpolator: */
1859 time_interpolator = NULL;
1860 /* find the next-best interpolator */
1861 for (curr = time_interpolator_list; curr; curr = curr->next)
1862 if (is_better_time_interpolator(curr))
1863 time_interpolator = curr;
1864 time_interpolator_reset();
1866 write_sequnlock_irqrestore(&xtime_lock, flags);
1867 spin_unlock(&time_interpolator_lock);
1869 #endif /* CONFIG_TIME_INTERPOLATION */
1872 * msleep - sleep safely even with waitqueue interruptions
1873 * @msecs: Time in milliseconds to sleep for
1875 void msleep(unsigned int msecs)
1877 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1880 timeout = schedule_timeout_uninterruptible(timeout);
1883 EXPORT_SYMBOL(msleep);
1886 * msleep_interruptible - sleep waiting for signals
1887 * @msecs: Time in milliseconds to sleep for
1889 unsigned long msleep_interruptible(unsigned int msecs)
1891 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1893 while (timeout && !signal_pending(current))
1894 timeout = schedule_timeout_interruptible(timeout);
1895 return jiffies_to_msecs(timeout);
1898 EXPORT_SYMBOL(msleep_interruptible);