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>
38 #include <linux/kallsyms.h>
40 #include <asm/uaccess.h>
41 #include <asm/unistd.h>
42 #include <asm/div64.h>
43 #include <asm/timex.h>
46 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
48 EXPORT_SYMBOL(jiffies_64);
51 * per-CPU timer vector definitions:
53 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
54 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
55 #define TVN_SIZE (1 << TVN_BITS)
56 #define TVR_SIZE (1 << TVR_BITS)
57 #define TVN_MASK (TVN_SIZE - 1)
58 #define TVR_MASK (TVR_SIZE - 1)
60 typedef struct tvec_s {
61 struct list_head vec[TVN_SIZE];
64 typedef struct tvec_root_s {
65 struct list_head vec[TVR_SIZE];
68 struct tvec_t_base_s {
70 struct timer_list *running_timer;
71 unsigned long timer_jiffies;
77 } ____cacheline_aligned_in_smp;
79 typedef struct tvec_t_base_s tvec_base_t;
81 tvec_base_t boot_tvec_bases;
82 EXPORT_SYMBOL(boot_tvec_bases);
83 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
86 * __round_jiffies - function to round jiffies to a full second
87 * @j: the time in (absolute) jiffies that should be rounded
88 * @cpu: the processor number on which the timeout will happen
90 * __round_jiffies() rounds an absolute time in the future (in jiffies)
91 * up or down to (approximately) full seconds. This is useful for timers
92 * for which the exact time they fire does not matter too much, as long as
93 * they fire approximately every X seconds.
95 * By rounding these timers to whole seconds, all such timers will fire
96 * at the same time, rather than at various times spread out. The goal
97 * of this is to have the CPU wake up less, which saves power.
99 * The exact rounding is skewed for each processor to avoid all
100 * processors firing at the exact same time, which could lead
101 * to lock contention or spurious cache line bouncing.
103 * The return value is the rounded version of the @j parameter.
105 unsigned long __round_jiffies(unsigned long j, int cpu)
108 unsigned long original = j;
111 * We don't want all cpus firing their timers at once hitting the
112 * same lock or cachelines, so we skew each extra cpu with an extra
113 * 3 jiffies. This 3 jiffies came originally from the mm/ code which
115 * The skew is done by adding 3*cpunr, then round, then subtract this
116 * extra offset again.
123 * If the target jiffie is just after a whole second (which can happen
124 * due to delays of the timer irq, long irq off times etc etc) then
125 * we should round down to the whole second, not up. Use 1/4th second
126 * as cutoff for this rounding as an extreme upper bound for this.
128 if (rem < HZ/4) /* round down */
133 /* now that we have rounded, subtract the extra skew again */
136 if (j <= jiffies) /* rounding ate our timeout entirely; */
140 EXPORT_SYMBOL_GPL(__round_jiffies);
143 * __round_jiffies_relative - function to round jiffies to a full second
144 * @j: the time in (relative) jiffies that should be rounded
145 * @cpu: the processor number on which the timeout will happen
147 * __round_jiffies_relative() rounds a time delta in the future (in jiffies)
148 * up or down to (approximately) full seconds. This is useful for timers
149 * for which the exact time they fire does not matter too much, as long as
150 * they fire approximately every X seconds.
152 * By rounding these timers to whole seconds, all such timers will fire
153 * at the same time, rather than at various times spread out. The goal
154 * of this is to have the CPU wake up less, which saves power.
156 * The exact rounding is skewed for each processor to avoid all
157 * processors firing at the exact same time, which could lead
158 * to lock contention or spurious cache line bouncing.
160 * The return value is the rounded version of the @j parameter.
162 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
165 * In theory the following code can skip a jiffy in case jiffies
166 * increments right between the addition and the later subtraction.
167 * However since the entire point of this function is to use approximate
168 * timeouts, it's entirely ok to not handle that.
170 return __round_jiffies(j + jiffies, cpu) - jiffies;
172 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
175 * round_jiffies - function to round jiffies to a full second
176 * @j: the time in (absolute) jiffies that should be rounded
178 * round_jiffies() rounds an absolute time in the future (in jiffies)
179 * up or down to (approximately) full seconds. This is useful for timers
180 * for which the exact time they fire does not matter too much, as long as
181 * they fire approximately every X seconds.
183 * By rounding these timers to whole seconds, all such timers will fire
184 * at the same time, rather than at various times spread out. The goal
185 * of this is to have the CPU wake up less, which saves power.
187 * The return value is the rounded version of the @j parameter.
189 unsigned long round_jiffies(unsigned long j)
191 return __round_jiffies(j, raw_smp_processor_id());
193 EXPORT_SYMBOL_GPL(round_jiffies);
196 * round_jiffies_relative - function to round jiffies to a full second
197 * @j: the time in (relative) jiffies that should be rounded
199 * round_jiffies_relative() rounds a time delta in the future (in jiffies)
200 * up or down to (approximately) full seconds. This is useful for timers
201 * for which the exact time they fire does not matter too much, as long as
202 * they fire approximately every X seconds.
204 * By rounding these timers to whole seconds, all such timers will fire
205 * at the same time, rather than at various times spread out. The goal
206 * of this is to have the CPU wake up less, which saves power.
208 * The return value is the rounded version of the @j parameter.
210 unsigned long round_jiffies_relative(unsigned long j)
212 return __round_jiffies_relative(j, raw_smp_processor_id());
214 EXPORT_SYMBOL_GPL(round_jiffies_relative);
217 static inline void set_running_timer(tvec_base_t *base,
218 struct timer_list *timer)
221 base->running_timer = timer;
225 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
227 unsigned long expires = timer->expires;
228 unsigned long idx = expires - base->timer_jiffies;
229 struct list_head *vec;
231 if (idx < TVR_SIZE) {
232 int i = expires & TVR_MASK;
233 vec = base->tv1.vec + i;
234 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
235 int i = (expires >> TVR_BITS) & TVN_MASK;
236 vec = base->tv2.vec + i;
237 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
238 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
239 vec = base->tv3.vec + i;
240 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
241 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
242 vec = base->tv4.vec + i;
243 } else if ((signed long) idx < 0) {
245 * Can happen if you add a timer with expires == jiffies,
246 * or you set a timer to go off in the past
248 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
251 /* If the timeout is larger than 0xffffffff on 64-bit
252 * architectures then we use the maximum timeout:
254 if (idx > 0xffffffffUL) {
256 expires = idx + base->timer_jiffies;
258 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
259 vec = base->tv5.vec + i;
264 list_add_tail(&timer->entry, vec);
267 #ifdef CONFIG_TIMER_STATS
268 void __timer_stats_timer_set_start_info(struct timer_list *timer, void *addr)
270 if (timer->start_site)
273 timer->start_site = addr;
274 memcpy(timer->start_comm, current->comm, TASK_COMM_LEN);
275 timer->start_pid = current->pid;
280 * init_timer - initialize a timer.
281 * @timer: the timer to be initialized
283 * init_timer() must be done to a timer prior calling *any* of the
284 * other timer functions.
286 void fastcall init_timer(struct timer_list *timer)
288 timer->entry.next = NULL;
289 timer->base = __raw_get_cpu_var(tvec_bases);
290 #ifdef CONFIG_TIMER_STATS
291 timer->start_site = NULL;
292 timer->start_pid = -1;
293 memset(timer->start_comm, 0, TASK_COMM_LEN);
296 EXPORT_SYMBOL(init_timer);
298 static inline void detach_timer(struct timer_list *timer,
301 struct list_head *entry = &timer->entry;
303 __list_del(entry->prev, entry->next);
306 entry->prev = LIST_POISON2;
310 * We are using hashed locking: holding per_cpu(tvec_bases).lock
311 * means that all timers which are tied to this base via timer->base are
312 * locked, and the base itself is locked too.
314 * So __run_timers/migrate_timers can safely modify all timers which could
315 * be found on ->tvX lists.
317 * When the timer's base is locked, and the timer removed from list, it is
318 * possible to set timer->base = NULL and drop the lock: the timer remains
321 static tvec_base_t *lock_timer_base(struct timer_list *timer,
322 unsigned long *flags)
323 __acquires(timer->base->lock)
329 if (likely(base != NULL)) {
330 spin_lock_irqsave(&base->lock, *flags);
331 if (likely(base == timer->base))
333 /* The timer has migrated to another CPU */
334 spin_unlock_irqrestore(&base->lock, *flags);
340 int __mod_timer(struct timer_list *timer, unsigned long expires)
342 tvec_base_t *base, *new_base;
346 timer_stats_timer_set_start_info(timer);
347 BUG_ON(!timer->function);
349 base = lock_timer_base(timer, &flags);
351 if (timer_pending(timer)) {
352 detach_timer(timer, 0);
356 new_base = __get_cpu_var(tvec_bases);
358 if (base != new_base) {
360 * We are trying to schedule the timer on the local CPU.
361 * However we can't change timer's base while it is running,
362 * otherwise del_timer_sync() can't detect that the timer's
363 * handler yet has not finished. This also guarantees that
364 * the timer is serialized wrt itself.
366 if (likely(base->running_timer != timer)) {
367 /* See the comment in lock_timer_base() */
369 spin_unlock(&base->lock);
371 spin_lock(&base->lock);
376 timer->expires = expires;
377 internal_add_timer(base, timer);
378 spin_unlock_irqrestore(&base->lock, flags);
383 EXPORT_SYMBOL(__mod_timer);
386 * add_timer_on - start a timer on a particular CPU
387 * @timer: the timer to be added
388 * @cpu: the CPU to start it on
390 * This is not very scalable on SMP. Double adds are not possible.
392 void add_timer_on(struct timer_list *timer, int cpu)
394 tvec_base_t *base = per_cpu(tvec_bases, cpu);
397 timer_stats_timer_set_start_info(timer);
398 BUG_ON(timer_pending(timer) || !timer->function);
399 spin_lock_irqsave(&base->lock, flags);
401 internal_add_timer(base, timer);
402 spin_unlock_irqrestore(&base->lock, flags);
407 * mod_timer - modify a timer's timeout
408 * @timer: the timer to be modified
409 * @expires: new timeout in jiffies
411 * mod_timer() is a more efficient way to update the expire field of an
412 * active timer (if the timer is inactive it will be activated)
414 * mod_timer(timer, expires) is equivalent to:
416 * del_timer(timer); timer->expires = expires; add_timer(timer);
418 * Note that if there are multiple unserialized concurrent users of the
419 * same timer, then mod_timer() is the only safe way to modify the timeout,
420 * since add_timer() cannot modify an already running timer.
422 * The function returns whether it has modified a pending timer or not.
423 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
424 * active timer returns 1.)
426 int mod_timer(struct timer_list *timer, unsigned long expires)
428 BUG_ON(!timer->function);
430 timer_stats_timer_set_start_info(timer);
432 * This is a common optimization triggered by the
433 * networking code - if the timer is re-modified
434 * to be the same thing then just return:
436 if (timer->expires == expires && timer_pending(timer))
439 return __mod_timer(timer, expires);
442 EXPORT_SYMBOL(mod_timer);
445 * del_timer - deactive a timer.
446 * @timer: the timer to be deactivated
448 * del_timer() deactivates a timer - this works on both active and inactive
451 * The function returns whether it has deactivated a pending timer or not.
452 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
453 * active timer returns 1.)
455 int del_timer(struct timer_list *timer)
461 timer_stats_timer_clear_start_info(timer);
462 if (timer_pending(timer)) {
463 base = lock_timer_base(timer, &flags);
464 if (timer_pending(timer)) {
465 detach_timer(timer, 1);
468 spin_unlock_irqrestore(&base->lock, flags);
474 EXPORT_SYMBOL(del_timer);
478 * try_to_del_timer_sync - Try to deactivate a timer
479 * @timer: timer do del
481 * This function tries to deactivate a timer. Upon successful (ret >= 0)
482 * exit the timer is not queued and the handler is not running on any CPU.
484 * It must not be called from interrupt contexts.
486 int try_to_del_timer_sync(struct timer_list *timer)
492 base = lock_timer_base(timer, &flags);
494 if (base->running_timer == timer)
498 if (timer_pending(timer)) {
499 detach_timer(timer, 1);
503 spin_unlock_irqrestore(&base->lock, flags);
509 * del_timer_sync - deactivate a timer and wait for the handler to finish.
510 * @timer: the timer to be deactivated
512 * This function only differs from del_timer() on SMP: besides deactivating
513 * the timer it also makes sure the handler has finished executing on other
516 * Synchronization rules: Callers must prevent restarting of the timer,
517 * otherwise this function is meaningless. It must not be called from
518 * interrupt contexts. The caller must not hold locks which would prevent
519 * completion of the timer's handler. The timer's handler must not call
520 * add_timer_on(). Upon exit the timer is not queued and the handler is
521 * not running on any CPU.
523 * The function returns whether it has deactivated a pending timer or not.
525 int del_timer_sync(struct timer_list *timer)
528 int ret = try_to_del_timer_sync(timer);
535 EXPORT_SYMBOL(del_timer_sync);
538 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
540 /* cascade all the timers from tv up one level */
541 struct timer_list *timer, *tmp;
542 struct list_head tv_list;
544 list_replace_init(tv->vec + index, &tv_list);
547 * We are removing _all_ timers from the list, so we
548 * don't have to detach them individually.
550 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
551 BUG_ON(timer->base != base);
552 internal_add_timer(base, timer);
558 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
561 * __run_timers - run all expired timers (if any) on this CPU.
562 * @base: the timer vector to be processed.
564 * This function cascades all vectors and executes all expired timer
567 static inline void __run_timers(tvec_base_t *base)
569 struct timer_list *timer;
571 spin_lock_irq(&base->lock);
572 while (time_after_eq(jiffies, base->timer_jiffies)) {
573 struct list_head work_list;
574 struct list_head *head = &work_list;
575 int index = base->timer_jiffies & TVR_MASK;
581 (!cascade(base, &base->tv2, INDEX(0))) &&
582 (!cascade(base, &base->tv3, INDEX(1))) &&
583 !cascade(base, &base->tv4, INDEX(2)))
584 cascade(base, &base->tv5, INDEX(3));
585 ++base->timer_jiffies;
586 list_replace_init(base->tv1.vec + index, &work_list);
587 while (!list_empty(head)) {
588 void (*fn)(unsigned long);
591 timer = list_entry(head->next,struct timer_list,entry);
592 fn = timer->function;
595 timer_stats_account_timer(timer);
597 set_running_timer(base, timer);
598 detach_timer(timer, 1);
599 spin_unlock_irq(&base->lock);
601 int preempt_count = preempt_count();
603 if (preempt_count != preempt_count()) {
604 printk(KERN_WARNING "huh, entered %p "
605 "with preempt_count %08x, exited"
612 spin_lock_irq(&base->lock);
615 set_running_timer(base, NULL);
616 spin_unlock_irq(&base->lock);
619 #if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
621 * Find out when the next timer event is due to happen. This
622 * is used on S/390 to stop all activity when a cpus is idle.
623 * This functions needs to be called disabled.
625 static unsigned long __next_timer_interrupt(tvec_base_t *base)
627 unsigned long timer_jiffies = base->timer_jiffies;
628 unsigned long expires = timer_jiffies + (LONG_MAX >> 1);
629 int index, slot, array, found = 0;
630 struct timer_list *nte;
633 /* Look for timer events in tv1. */
634 index = slot = timer_jiffies & TVR_MASK;
636 list_for_each_entry(nte, base->tv1.vec + slot, entry) {
638 expires = nte->expires;
639 /* Look at the cascade bucket(s)? */
640 if (!index || slot < index)
644 slot = (slot + 1) & TVR_MASK;
645 } while (slot != index);
648 /* Calculate the next cascade event */
650 timer_jiffies += TVR_SIZE - index;
651 timer_jiffies >>= TVR_BITS;
654 varray[0] = &base->tv2;
655 varray[1] = &base->tv3;
656 varray[2] = &base->tv4;
657 varray[3] = &base->tv5;
659 for (array = 0; array < 4; array++) {
660 tvec_t *varp = varray[array];
662 index = slot = timer_jiffies & TVN_MASK;
664 list_for_each_entry(nte, varp->vec + slot, entry) {
666 if (time_before(nte->expires, expires))
667 expires = nte->expires;
670 * Do we still search for the first timer or are
671 * we looking up the cascade buckets ?
674 /* Look at the cascade bucket(s)? */
675 if (!index || slot < index)
679 slot = (slot + 1) & TVN_MASK;
680 } while (slot != index);
683 timer_jiffies += TVN_SIZE - index;
684 timer_jiffies >>= TVN_BITS;
690 * Check, if the next hrtimer event is before the next timer wheel
693 static unsigned long cmp_next_hrtimer_event(unsigned long now,
694 unsigned long expires)
696 ktime_t hr_delta = hrtimer_get_next_event();
697 struct timespec tsdelta;
699 if (hr_delta.tv64 == KTIME_MAX)
702 if (hr_delta.tv64 <= TICK_NSEC)
705 tsdelta = ktime_to_timespec(hr_delta);
706 now += timespec_to_jiffies(&tsdelta);
707 if (time_before(now, expires))
713 * next_timer_interrupt - return the jiffy of the next pending timer
714 * @now: current time (in jiffies)
716 unsigned long get_next_timer_interrupt(unsigned long now)
718 tvec_base_t *base = __get_cpu_var(tvec_bases);
719 unsigned long expires;
721 spin_lock(&base->lock);
722 expires = __next_timer_interrupt(base);
723 spin_unlock(&base->lock);
725 if (time_before_eq(expires, now))
728 return cmp_next_hrtimer_event(now, expires);
731 #ifdef CONFIG_NO_IDLE_HZ
732 unsigned long next_timer_interrupt(void)
734 return get_next_timer_interrupt(jiffies);
740 /******************************************************************/
744 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
745 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
746 * at zero at system boot time, so wall_to_monotonic will be negative,
747 * however, we will ALWAYS keep the tv_nsec part positive so we can use
748 * the usual normalization.
750 struct timespec xtime __attribute__ ((aligned (16)));
751 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
753 EXPORT_SYMBOL(xtime);
756 /* XXX - all of this timekeeping code should be later moved to time.c */
757 #include <linux/clocksource.h>
758 static struct clocksource *clock; /* pointer to current clocksource */
760 #ifdef CONFIG_GENERIC_TIME
762 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
764 * private function, must hold xtime_lock lock when being
765 * called. Returns the number of nanoseconds since the
766 * last call to update_wall_time() (adjusted by NTP scaling)
768 static inline s64 __get_nsec_offset(void)
770 cycle_t cycle_now, cycle_delta;
773 /* read clocksource: */
774 cycle_now = clocksource_read(clock);
776 /* calculate the delta since the last update_wall_time: */
777 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
779 /* convert to nanoseconds: */
780 ns_offset = cyc2ns(clock, cycle_delta);
786 * __get_realtime_clock_ts - Returns the time of day in a timespec
787 * @ts: pointer to the timespec to be set
789 * Returns the time of day in a timespec. Used by
790 * do_gettimeofday() and get_realtime_clock_ts().
792 static inline void __get_realtime_clock_ts(struct timespec *ts)
798 seq = read_seqbegin(&xtime_lock);
801 nsecs = __get_nsec_offset();
803 } while (read_seqretry(&xtime_lock, seq));
805 timespec_add_ns(ts, nsecs);
809 * getnstimeofday - Returns the time of day in a timespec
810 * @ts: pointer to the timespec to be set
812 * Returns the time of day in a timespec.
814 void getnstimeofday(struct timespec *ts)
816 __get_realtime_clock_ts(ts);
819 EXPORT_SYMBOL(getnstimeofday);
822 * do_gettimeofday - Returns the time of day in a timeval
823 * @tv: pointer to the timeval to be set
825 * NOTE: Users should be converted to using get_realtime_clock_ts()
827 void do_gettimeofday(struct timeval *tv)
831 __get_realtime_clock_ts(&now);
832 tv->tv_sec = now.tv_sec;
833 tv->tv_usec = now.tv_nsec/1000;
836 EXPORT_SYMBOL(do_gettimeofday);
838 * do_settimeofday - Sets the time of day
839 * @tv: pointer to the timespec variable containing the new time
841 * Sets the time of day to the new time and update NTP and notify hrtimers
843 int do_settimeofday(struct timespec *tv)
846 time_t wtm_sec, sec = tv->tv_sec;
847 long wtm_nsec, nsec = tv->tv_nsec;
849 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
852 write_seqlock_irqsave(&xtime_lock, flags);
854 nsec -= __get_nsec_offset();
856 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
857 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
859 set_normalized_timespec(&xtime, sec, nsec);
860 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
865 write_sequnlock_irqrestore(&xtime_lock, flags);
867 /* signal hrtimers about time change */
873 EXPORT_SYMBOL(do_settimeofday);
876 * change_clocksource - Swaps clocksources if a new one is available
878 * Accumulates current time interval and initializes new clocksource
880 static void change_clocksource(void)
882 struct clocksource *new;
886 new = clocksource_get_next();
891 now = clocksource_read(new);
892 nsec = __get_nsec_offset();
893 timespec_add_ns(&xtime, nsec);
896 clock->cycle_last = now;
899 clock->xtime_nsec = 0;
900 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
904 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
908 static inline void change_clocksource(void) { }
912 * timekeeping_is_continuous - check to see if timekeeping is free running
914 int timekeeping_is_continuous(void)
920 seq = read_seqbegin(&xtime_lock);
922 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
924 } while (read_seqretry(&xtime_lock, seq));
930 * read_persistent_clock - Return time in seconds from the persistent clock.
932 * Weak dummy function for arches that do not yet support it.
933 * Returns seconds from epoch using the battery backed persistent clock.
934 * Returns zero if unsupported.
936 * XXX - Do be sure to remove it once all arches implement it.
938 unsigned long __attribute__((weak)) read_persistent_clock(void)
944 * timekeeping_init - Initializes the clocksource and common timekeeping values
946 void __init timekeeping_init(void)
949 unsigned long sec = read_persistent_clock();
951 write_seqlock_irqsave(&xtime_lock, flags);
955 clock = clocksource_get_next();
956 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
957 clock->cycle_last = clocksource_read(clock);
961 set_normalized_timespec(&wall_to_monotonic,
962 -xtime.tv_sec, -xtime.tv_nsec);
964 write_sequnlock_irqrestore(&xtime_lock, flags);
967 /* flag for if timekeeping is suspended */
968 static int timekeeping_suspended;
969 /* time in seconds when suspend began */
970 static unsigned long timekeeping_suspend_time;
973 * timekeeping_resume - Resumes the generic timekeeping subsystem.
976 * This is for the generic clocksource timekeeping.
977 * xtime/wall_to_monotonic/jiffies/etc are
978 * still managed by arch specific suspend/resume code.
980 static int timekeeping_resume(struct sys_device *dev)
983 unsigned long now = read_persistent_clock();
985 write_seqlock_irqsave(&xtime_lock, flags);
987 if (now && (now > timekeeping_suspend_time)) {
988 unsigned long sleep_length = now - timekeeping_suspend_time;
990 xtime.tv_sec += sleep_length;
991 wall_to_monotonic.tv_sec -= sleep_length;
993 /* re-base the last cycle value */
994 clock->cycle_last = clocksource_read(clock);
996 timekeeping_suspended = 0;
997 write_sequnlock_irqrestore(&xtime_lock, flags);
999 touch_softlockup_watchdog();
1001 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
1003 /* Resume hrtimers */
1009 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
1011 unsigned long flags;
1013 write_seqlock_irqsave(&xtime_lock, flags);
1014 timekeeping_suspended = 1;
1015 timekeeping_suspend_time = read_persistent_clock();
1016 write_sequnlock_irqrestore(&xtime_lock, flags);
1018 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1023 /* sysfs resume/suspend bits for timekeeping */
1024 static struct sysdev_class timekeeping_sysclass = {
1025 .resume = timekeeping_resume,
1026 .suspend = timekeeping_suspend,
1027 set_kset_name("timekeeping"),
1030 static struct sys_device device_timer = {
1032 .cls = &timekeeping_sysclass,
1035 static int __init timekeeping_init_device(void)
1037 int error = sysdev_class_register(&timekeeping_sysclass);
1039 error = sysdev_register(&device_timer);
1043 device_initcall(timekeeping_init_device);
1046 * If the error is already larger, we look ahead even further
1047 * to compensate for late or lost adjustments.
1049 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
1053 u32 look_ahead, adj;
1057 * Use the current error value to determine how much to look ahead.
1058 * The larger the error the slower we adjust for it to avoid problems
1059 * with losing too many ticks, otherwise we would overadjust and
1060 * produce an even larger error. The smaller the adjustment the
1061 * faster we try to adjust for it, as lost ticks can do less harm
1062 * here. This is tuned so that an error of about 1 msec is adusted
1063 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1065 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1066 error2 = abs(error2);
1067 for (look_ahead = 0; error2 > 0; look_ahead++)
1071 * Now calculate the error in (1 << look_ahead) ticks, but first
1072 * remove the single look ahead already included in the error.
1074 tick_error = current_tick_length() >>
1075 (TICK_LENGTH_SHIFT - clock->shift + 1);
1076 tick_error -= clock->xtime_interval >> 1;
1077 error = ((error - tick_error) >> look_ahead) + tick_error;
1079 /* Finally calculate the adjustment shift value. */
1084 *interval = -*interval;
1088 for (adj = 0; error > i; adj++)
1097 * Adjust the multiplier to reduce the error value,
1098 * this is optimized for the most common adjustments of -1,0,1,
1099 * for other values we can do a bit more work.
1101 static void clocksource_adjust(struct clocksource *clock, s64 offset)
1103 s64 error, interval = clock->cycle_interval;
1106 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1107 if (error > interval) {
1109 if (likely(error <= interval))
1112 adj = clocksource_bigadjust(error, &interval, &offset);
1113 } else if (error < -interval) {
1115 if (likely(error >= -interval)) {
1117 interval = -interval;
1120 adj = clocksource_bigadjust(error, &interval, &offset);
1125 clock->xtime_interval += interval;
1126 clock->xtime_nsec -= offset;
1127 clock->error -= (interval - offset) <<
1128 (TICK_LENGTH_SHIFT - clock->shift);
1132 * update_wall_time - Uses the current clocksource to increment the wall time
1134 * Called from the timer interrupt, must hold a write on xtime_lock.
1136 static void update_wall_time(void)
1140 /* Make sure we're fully resumed: */
1141 if (unlikely(timekeeping_suspended))
1144 #ifdef CONFIG_GENERIC_TIME
1145 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1147 offset = clock->cycle_interval;
1149 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1151 /* normally this loop will run just once, however in the
1152 * case of lost or late ticks, it will accumulate correctly.
1154 while (offset >= clock->cycle_interval) {
1155 /* accumulate one interval */
1156 clock->xtime_nsec += clock->xtime_interval;
1157 clock->cycle_last += clock->cycle_interval;
1158 offset -= clock->cycle_interval;
1160 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1161 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1166 /* interpolator bits */
1167 time_interpolator_update(clock->xtime_interval
1170 /* accumulate error between NTP and clock interval */
1171 clock->error += current_tick_length();
1172 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1175 /* correct the clock when NTP error is too big */
1176 clocksource_adjust(clock, offset);
1178 /* store full nanoseconds into xtime */
1179 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1180 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1182 /* check to see if there is a new clocksource to use */
1183 change_clocksource();
1184 update_vsyscall(&xtime, clock);
1188 * Called from the timer interrupt handler to charge one tick to the current
1189 * process. user_tick is 1 if the tick is user time, 0 for system.
1191 void update_process_times(int user_tick)
1193 struct task_struct *p = current;
1194 int cpu = smp_processor_id();
1196 /* Note: this timer irq context must be accounted for as well. */
1198 account_user_time(p, jiffies_to_cputime(1));
1200 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1202 if (rcu_pending(cpu))
1203 rcu_check_callbacks(cpu, user_tick);
1205 run_posix_cpu_timers(p);
1209 * Nr of active tasks - counted in fixed-point numbers
1211 static unsigned long count_active_tasks(void)
1213 return nr_active() * FIXED_1;
1217 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1218 * imply that avenrun[] is the standard name for this kind of thing.
1219 * Nothing else seems to be standardized: the fractional size etc
1220 * all seem to differ on different machines.
1222 * Requires xtime_lock to access.
1224 unsigned long avenrun[3];
1226 EXPORT_SYMBOL(avenrun);
1229 * calc_load - given tick count, update the avenrun load estimates.
1230 * This is called while holding a write_lock on xtime_lock.
1232 static inline void calc_load(unsigned long ticks)
1234 unsigned long active_tasks; /* fixed-point */
1235 static int count = LOAD_FREQ;
1238 if (unlikely(count < 0)) {
1239 active_tasks = count_active_tasks();
1241 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1242 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1243 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1245 } while (count < 0);
1250 * This read-write spinlock protects us from races in SMP while
1251 * playing with xtime and avenrun.
1253 __attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1255 EXPORT_SYMBOL(xtime_lock);
1258 * This function runs timers and the timer-tq in bottom half context.
1260 static void run_timer_softirq(struct softirq_action *h)
1262 tvec_base_t *base = __get_cpu_var(tvec_bases);
1264 hrtimer_run_queues();
1266 if (time_after_eq(jiffies, base->timer_jiffies))
1271 * Called by the local, per-CPU timer interrupt on SMP.
1273 void run_local_timers(void)
1275 raise_softirq(TIMER_SOFTIRQ);
1280 * Called by the timer interrupt. xtime_lock must already be taken
1283 static inline void update_times(unsigned long ticks)
1290 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1291 * without sampling the sequence number in xtime_lock.
1292 * jiffies is defined in the linker script...
1295 void do_timer(unsigned long ticks)
1297 jiffies_64 += ticks;
1298 update_times(ticks);
1301 #ifdef __ARCH_WANT_SYS_ALARM
1304 * For backwards compatibility? This can be done in libc so Alpha
1305 * and all newer ports shouldn't need it.
1307 asmlinkage unsigned long sys_alarm(unsigned int seconds)
1309 return alarm_setitimer(seconds);
1317 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1318 * should be moved into arch/i386 instead?
1322 * sys_getpid - return the thread group id of the current process
1324 * Note, despite the name, this returns the tgid not the pid. The tgid and
1325 * the pid are identical unless CLONE_THREAD was specified on clone() in
1326 * which case the tgid is the same in all threads of the same group.
1328 * This is SMP safe as current->tgid does not change.
1330 asmlinkage long sys_getpid(void)
1332 return current->tgid;
1336 * Accessing ->real_parent is not SMP-safe, it could
1337 * change from under us. However, we can use a stale
1338 * value of ->real_parent under rcu_read_lock(), see
1339 * release_task()->call_rcu(delayed_put_task_struct).
1341 asmlinkage long sys_getppid(void)
1346 pid = rcu_dereference(current->real_parent)->tgid;
1352 asmlinkage long sys_getuid(void)
1354 /* Only we change this so SMP safe */
1355 return current->uid;
1358 asmlinkage long sys_geteuid(void)
1360 /* Only we change this so SMP safe */
1361 return current->euid;
1364 asmlinkage long sys_getgid(void)
1366 /* Only we change this so SMP safe */
1367 return current->gid;
1370 asmlinkage long sys_getegid(void)
1372 /* Only we change this so SMP safe */
1373 return current->egid;
1378 static void process_timeout(unsigned long __data)
1380 wake_up_process((struct task_struct *)__data);
1384 * schedule_timeout - sleep until timeout
1385 * @timeout: timeout value in jiffies
1387 * Make the current task sleep until @timeout jiffies have
1388 * elapsed. The routine will return immediately unless
1389 * the current task state has been set (see set_current_state()).
1391 * You can set the task state as follows -
1393 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1394 * pass before the routine returns. The routine will return 0
1396 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1397 * delivered to the current task. In this case the remaining time
1398 * in jiffies will be returned, or 0 if the timer expired in time
1400 * The current task state is guaranteed to be TASK_RUNNING when this
1403 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1404 * the CPU away without a bound on the timeout. In this case the return
1405 * value will be %MAX_SCHEDULE_TIMEOUT.
1407 * In all cases the return value is guaranteed to be non-negative.
1409 fastcall signed long __sched schedule_timeout(signed long timeout)
1411 struct timer_list timer;
1412 unsigned long expire;
1416 case MAX_SCHEDULE_TIMEOUT:
1418 * These two special cases are useful to be comfortable
1419 * in the caller. Nothing more. We could take
1420 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1421 * but I' d like to return a valid offset (>=0) to allow
1422 * the caller to do everything it want with the retval.
1428 * Another bit of PARANOID. Note that the retval will be
1429 * 0 since no piece of kernel is supposed to do a check
1430 * for a negative retval of schedule_timeout() (since it
1431 * should never happens anyway). You just have the printk()
1432 * that will tell you if something is gone wrong and where.
1435 printk(KERN_ERR "schedule_timeout: wrong timeout "
1436 "value %lx\n", timeout);
1438 current->state = TASK_RUNNING;
1443 expire = timeout + jiffies;
1445 setup_timer(&timer, process_timeout, (unsigned long)current);
1446 __mod_timer(&timer, expire);
1448 del_singleshot_timer_sync(&timer);
1450 timeout = expire - jiffies;
1453 return timeout < 0 ? 0 : timeout;
1455 EXPORT_SYMBOL(schedule_timeout);
1458 * We can use __set_current_state() here because schedule_timeout() calls
1459 * schedule() unconditionally.
1461 signed long __sched schedule_timeout_interruptible(signed long timeout)
1463 __set_current_state(TASK_INTERRUPTIBLE);
1464 return schedule_timeout(timeout);
1466 EXPORT_SYMBOL(schedule_timeout_interruptible);
1468 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1470 __set_current_state(TASK_UNINTERRUPTIBLE);
1471 return schedule_timeout(timeout);
1473 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1475 /* Thread ID - the internal kernel "pid" */
1476 asmlinkage long sys_gettid(void)
1478 return current->pid;
1482 * do_sysinfo - fill in sysinfo struct
1483 * @info: pointer to buffer to fill
1485 int do_sysinfo(struct sysinfo *info)
1487 unsigned long mem_total, sav_total;
1488 unsigned int mem_unit, bitcount;
1491 memset(info, 0, sizeof(struct sysinfo));
1495 seq = read_seqbegin(&xtime_lock);
1498 * This is annoying. The below is the same thing
1499 * posix_get_clock_monotonic() does, but it wants to
1500 * take the lock which we want to cover the loads stuff
1504 getnstimeofday(&tp);
1505 tp.tv_sec += wall_to_monotonic.tv_sec;
1506 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1507 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1508 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1511 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1513 info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1514 info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1515 info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1517 info->procs = nr_threads;
1518 } while (read_seqretry(&xtime_lock, seq));
1524 * If the sum of all the available memory (i.e. ram + swap)
1525 * is less than can be stored in a 32 bit unsigned long then
1526 * we can be binary compatible with 2.2.x kernels. If not,
1527 * well, in that case 2.2.x was broken anyways...
1529 * -Erik Andersen <andersee@debian.org>
1532 mem_total = info->totalram + info->totalswap;
1533 if (mem_total < info->totalram || mem_total < info->totalswap)
1536 mem_unit = info->mem_unit;
1537 while (mem_unit > 1) {
1540 sav_total = mem_total;
1542 if (mem_total < sav_total)
1547 * If mem_total did not overflow, multiply all memory values by
1548 * info->mem_unit and set it to 1. This leaves things compatible
1549 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1554 info->totalram <<= bitcount;
1555 info->freeram <<= bitcount;
1556 info->sharedram <<= bitcount;
1557 info->bufferram <<= bitcount;
1558 info->totalswap <<= bitcount;
1559 info->freeswap <<= bitcount;
1560 info->totalhigh <<= bitcount;
1561 info->freehigh <<= bitcount;
1567 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1573 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1580 * lockdep: we want to track each per-CPU base as a separate lock-class,
1581 * but timer-bases are kmalloc()-ed, so we need to attach separate
1584 static struct lock_class_key base_lock_keys[NR_CPUS];
1586 static int __devinit init_timers_cpu(int cpu)
1590 static char __devinitdata tvec_base_done[NR_CPUS];
1592 if (!tvec_base_done[cpu]) {
1593 static char boot_done;
1597 * The APs use this path later in boot
1599 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1603 memset(base, 0, sizeof(*base));
1604 per_cpu(tvec_bases, cpu) = base;
1607 * This is for the boot CPU - we use compile-time
1608 * static initialisation because per-cpu memory isn't
1609 * ready yet and because the memory allocators are not
1610 * initialised either.
1613 base = &boot_tvec_bases;
1615 tvec_base_done[cpu] = 1;
1617 base = per_cpu(tvec_bases, cpu);
1620 spin_lock_init(&base->lock);
1621 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1623 for (j = 0; j < TVN_SIZE; j++) {
1624 INIT_LIST_HEAD(base->tv5.vec + j);
1625 INIT_LIST_HEAD(base->tv4.vec + j);
1626 INIT_LIST_HEAD(base->tv3.vec + j);
1627 INIT_LIST_HEAD(base->tv2.vec + j);
1629 for (j = 0; j < TVR_SIZE; j++)
1630 INIT_LIST_HEAD(base->tv1.vec + j);
1632 base->timer_jiffies = jiffies;
1636 #ifdef CONFIG_HOTPLUG_CPU
1637 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1639 struct timer_list *timer;
1641 while (!list_empty(head)) {
1642 timer = list_entry(head->next, struct timer_list, entry);
1643 detach_timer(timer, 0);
1644 timer->base = new_base;
1645 internal_add_timer(new_base, timer);
1649 static void __devinit migrate_timers(int cpu)
1651 tvec_base_t *old_base;
1652 tvec_base_t *new_base;
1655 BUG_ON(cpu_online(cpu));
1656 old_base = per_cpu(tvec_bases, cpu);
1657 new_base = get_cpu_var(tvec_bases);
1659 local_irq_disable();
1660 double_spin_lock(&new_base->lock, &old_base->lock,
1661 smp_processor_id() < cpu);
1663 BUG_ON(old_base->running_timer);
1665 for (i = 0; i < TVR_SIZE; i++)
1666 migrate_timer_list(new_base, old_base->tv1.vec + i);
1667 for (i = 0; i < TVN_SIZE; i++) {
1668 migrate_timer_list(new_base, old_base->tv2.vec + i);
1669 migrate_timer_list(new_base, old_base->tv3.vec + i);
1670 migrate_timer_list(new_base, old_base->tv4.vec + i);
1671 migrate_timer_list(new_base, old_base->tv5.vec + i);
1674 double_spin_unlock(&new_base->lock, &old_base->lock,
1675 smp_processor_id() < cpu);
1677 put_cpu_var(tvec_bases);
1679 #endif /* CONFIG_HOTPLUG_CPU */
1681 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1682 unsigned long action, void *hcpu)
1684 long cpu = (long)hcpu;
1686 case CPU_UP_PREPARE:
1687 if (init_timers_cpu(cpu) < 0)
1690 #ifdef CONFIG_HOTPLUG_CPU
1692 migrate_timers(cpu);
1701 static struct notifier_block __cpuinitdata timers_nb = {
1702 .notifier_call = timer_cpu_notify,
1706 void __init init_timers(void)
1708 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1709 (void *)(long)smp_processor_id());
1713 BUG_ON(err == NOTIFY_BAD);
1714 register_cpu_notifier(&timers_nb);
1715 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1718 #ifdef CONFIG_TIME_INTERPOLATION
1720 struct time_interpolator *time_interpolator __read_mostly;
1721 static struct time_interpolator *time_interpolator_list __read_mostly;
1722 static DEFINE_SPINLOCK(time_interpolator_lock);
1724 static inline cycles_t time_interpolator_get_cycles(unsigned int src)
1726 unsigned long (*x)(void);
1730 case TIME_SOURCE_FUNCTION:
1731 x = time_interpolator->addr;
1734 case TIME_SOURCE_MMIO64 :
1735 return readq_relaxed((void __iomem *)time_interpolator->addr);
1737 case TIME_SOURCE_MMIO32 :
1738 return readl_relaxed((void __iomem *)time_interpolator->addr);
1740 default: return get_cycles();
1744 static inline u64 time_interpolator_get_counter(int writelock)
1746 unsigned int src = time_interpolator->source;
1748 if (time_interpolator->jitter)
1754 lcycle = time_interpolator->last_cycle;
1755 now = time_interpolator_get_cycles(src);
1756 if (lcycle && time_after(lcycle, now))
1759 /* When holding the xtime write lock, there's no need
1760 * to add the overhead of the cmpxchg. Readers are
1761 * force to retry until the write lock is released.
1764 time_interpolator->last_cycle = now;
1767 /* Keep track of the last timer value returned. The use of cmpxchg here
1768 * will cause contention in an SMP environment.
1770 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1774 return time_interpolator_get_cycles(src);
1777 void time_interpolator_reset(void)
1779 time_interpolator->offset = 0;
1780 time_interpolator->last_counter = time_interpolator_get_counter(1);
1783 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1785 unsigned long time_interpolator_get_offset(void)
1787 /* If we do not have a time interpolator set up then just return zero */
1788 if (!time_interpolator)
1791 return time_interpolator->offset +
1792 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1795 #define INTERPOLATOR_ADJUST 65536
1796 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1798 void time_interpolator_update(long delta_nsec)
1801 unsigned long offset;
1803 /* If there is no time interpolator set up then do nothing */
1804 if (!time_interpolator)
1808 * The interpolator compensates for late ticks by accumulating the late
1809 * time in time_interpolator->offset. A tick earlier than expected will
1810 * lead to a reset of the offset and a corresponding jump of the clock
1811 * forward. Again this only works if the interpolator clock is running
1812 * slightly slower than the regular clock and the tuning logic insures
1816 counter = time_interpolator_get_counter(1);
1817 offset = time_interpolator->offset +
1818 GET_TI_NSECS(counter, time_interpolator);
1820 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1821 time_interpolator->offset = offset - delta_nsec;
1823 time_interpolator->skips++;
1824 time_interpolator->ns_skipped += delta_nsec - offset;
1825 time_interpolator->offset = 0;
1827 time_interpolator->last_counter = counter;
1829 /* Tuning logic for time interpolator invoked every minute or so.
1830 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1831 * Increase interpolator clock speed if we skip too much time.
1833 if (jiffies % INTERPOLATOR_ADJUST == 0)
1835 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1836 time_interpolator->nsec_per_cyc--;
1837 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1838 time_interpolator->nsec_per_cyc++;
1839 time_interpolator->skips = 0;
1840 time_interpolator->ns_skipped = 0;
1845 is_better_time_interpolator(struct time_interpolator *new)
1847 if (!time_interpolator)
1849 return new->frequency > 2*time_interpolator->frequency ||
1850 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1854 register_time_interpolator(struct time_interpolator *ti)
1856 unsigned long flags;
1859 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1861 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1862 spin_lock(&time_interpolator_lock);
1863 write_seqlock_irqsave(&xtime_lock, flags);
1864 if (is_better_time_interpolator(ti)) {
1865 time_interpolator = ti;
1866 time_interpolator_reset();
1868 write_sequnlock_irqrestore(&xtime_lock, flags);
1870 ti->next = time_interpolator_list;
1871 time_interpolator_list = ti;
1872 spin_unlock(&time_interpolator_lock);
1876 unregister_time_interpolator(struct time_interpolator *ti)
1878 struct time_interpolator *curr, **prev;
1879 unsigned long flags;
1881 spin_lock(&time_interpolator_lock);
1882 prev = &time_interpolator_list;
1883 for (curr = *prev; curr; curr = curr->next) {
1891 write_seqlock_irqsave(&xtime_lock, flags);
1892 if (ti == time_interpolator) {
1893 /* we lost the best time-interpolator: */
1894 time_interpolator = NULL;
1895 /* find the next-best interpolator */
1896 for (curr = time_interpolator_list; curr; curr = curr->next)
1897 if (is_better_time_interpolator(curr))
1898 time_interpolator = curr;
1899 time_interpolator_reset();
1901 write_sequnlock_irqrestore(&xtime_lock, flags);
1902 spin_unlock(&time_interpolator_lock);
1904 #endif /* CONFIG_TIME_INTERPOLATION */
1907 * msleep - sleep safely even with waitqueue interruptions
1908 * @msecs: Time in milliseconds to sleep for
1910 void msleep(unsigned int msecs)
1912 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1915 timeout = schedule_timeout_uninterruptible(timeout);
1918 EXPORT_SYMBOL(msleep);
1921 * msleep_interruptible - sleep waiting for signals
1922 * @msecs: Time in milliseconds to sleep for
1924 unsigned long msleep_interruptible(unsigned int msecs)
1926 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1928 while (timeout && !signal_pending(current))
1929 timeout = schedule_timeout_interruptible(timeout);
1930 return jiffies_to_msecs(timeout);
1933 EXPORT_SYMBOL(msleep_interruptible);