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/vs_cvirt.h>
38 #include <linux/vserver/sched.h>
40 #include <asm/uaccess.h>
41 #include <asm/unistd.h>
42 #include <asm/div64.h>
43 #include <asm/timex.h>
46 #ifdef CONFIG_TIME_INTERPOLATION
47 static void time_interpolator_update(long delta_nsec);
49 #define time_interpolator_update(x)
52 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
54 EXPORT_SYMBOL(jiffies_64);
57 * per-CPU timer vector definitions:
59 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
60 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
61 #define TVN_SIZE (1 << TVN_BITS)
62 #define TVR_SIZE (1 << TVR_BITS)
63 #define TVN_MASK (TVN_SIZE - 1)
64 #define TVR_MASK (TVR_SIZE - 1)
66 typedef struct tvec_s {
67 struct list_head vec[TVN_SIZE];
70 typedef struct tvec_root_s {
71 struct list_head vec[TVR_SIZE];
74 struct tvec_t_base_s {
76 struct timer_list *running_timer;
77 unsigned long timer_jiffies;
83 } ____cacheline_aligned_in_smp;
85 typedef struct tvec_t_base_s tvec_base_t;
87 tvec_base_t boot_tvec_bases;
88 EXPORT_SYMBOL(boot_tvec_bases);
89 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = &boot_tvec_bases;
91 static inline void set_running_timer(tvec_base_t *base,
92 struct timer_list *timer)
95 base->running_timer = timer;
99 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
101 unsigned long expires = timer->expires;
102 unsigned long idx = expires - base->timer_jiffies;
103 struct list_head *vec;
105 if (idx < TVR_SIZE) {
106 int i = expires & TVR_MASK;
107 vec = base->tv1.vec + i;
108 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
109 int i = (expires >> TVR_BITS) & TVN_MASK;
110 vec = base->tv2.vec + i;
111 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
112 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
113 vec = base->tv3.vec + i;
114 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
115 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
116 vec = base->tv4.vec + i;
117 } else if ((signed long) idx < 0) {
119 * Can happen if you add a timer with expires == jiffies,
120 * or you set a timer to go off in the past
122 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
125 /* If the timeout is larger than 0xffffffff on 64-bit
126 * architectures then we use the maximum timeout:
128 if (idx > 0xffffffffUL) {
130 expires = idx + base->timer_jiffies;
132 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
133 vec = base->tv5.vec + i;
138 list_add_tail(&timer->entry, vec);
142 * init_timer - initialize a timer.
143 * @timer: the timer to be initialized
145 * init_timer() must be done to a timer prior calling *any* of the
146 * other timer functions.
148 void fastcall init_timer(struct timer_list *timer)
150 timer->entry.next = NULL;
151 timer->base = __raw_get_cpu_var(tvec_bases);
153 EXPORT_SYMBOL(init_timer);
155 static inline void detach_timer(struct timer_list *timer,
158 struct list_head *entry = &timer->entry;
160 __list_del(entry->prev, entry->next);
163 entry->prev = LIST_POISON2;
167 * We are using hashed locking: holding per_cpu(tvec_bases).lock
168 * means that all timers which are tied to this base via timer->base are
169 * locked, and the base itself is locked too.
171 * So __run_timers/migrate_timers can safely modify all timers which could
172 * be found on ->tvX lists.
174 * When the timer's base is locked, and the timer removed from list, it is
175 * possible to set timer->base = NULL and drop the lock: the timer remains
178 static tvec_base_t *lock_timer_base(struct timer_list *timer,
179 unsigned long *flags)
185 if (likely(base != NULL)) {
186 spin_lock_irqsave(&base->lock, *flags);
187 if (likely(base == timer->base))
189 /* The timer has migrated to another CPU */
190 spin_unlock_irqrestore(&base->lock, *flags);
196 int __mod_timer(struct timer_list *timer, unsigned long expires)
198 tvec_base_t *base, *new_base;
202 BUG_ON(!timer->function);
204 base = lock_timer_base(timer, &flags);
206 if (timer_pending(timer)) {
207 detach_timer(timer, 0);
211 new_base = __get_cpu_var(tvec_bases);
213 if (base != new_base) {
215 * We are trying to schedule the timer on the local CPU.
216 * However we can't change timer's base while it is running,
217 * otherwise del_timer_sync() can't detect that the timer's
218 * handler yet has not finished. This also guarantees that
219 * the timer is serialized wrt itself.
221 if (likely(base->running_timer != timer)) {
222 /* See the comment in lock_timer_base() */
224 spin_unlock(&base->lock);
226 spin_lock(&base->lock);
231 timer->expires = expires;
232 internal_add_timer(base, timer);
233 spin_unlock_irqrestore(&base->lock, flags);
238 EXPORT_SYMBOL(__mod_timer);
241 * add_timer_on - start a timer on a particular CPU
242 * @timer: the timer to be added
243 * @cpu: the CPU to start it on
245 * This is not very scalable on SMP. Double adds are not possible.
247 void add_timer_on(struct timer_list *timer, int cpu)
249 tvec_base_t *base = per_cpu(tvec_bases, cpu);
252 BUG_ON(timer_pending(timer) || !timer->function);
253 spin_lock_irqsave(&base->lock, flags);
255 internal_add_timer(base, timer);
256 spin_unlock_irqrestore(&base->lock, flags);
261 * mod_timer - modify a timer's timeout
262 * @timer: the timer to be modified
264 * mod_timer is a more efficient way to update the expire field of an
265 * active timer (if the timer is inactive it will be activated)
267 * mod_timer(timer, expires) is equivalent to:
269 * del_timer(timer); timer->expires = expires; add_timer(timer);
271 * Note that if there are multiple unserialized concurrent users of the
272 * same timer, then mod_timer() is the only safe way to modify the timeout,
273 * since add_timer() cannot modify an already running timer.
275 * The function returns whether it has modified a pending timer or not.
276 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
277 * active timer returns 1.)
279 int mod_timer(struct timer_list *timer, unsigned long expires)
281 BUG_ON(!timer->function);
284 * This is a common optimization triggered by the
285 * networking code - if the timer is re-modified
286 * to be the same thing then just return:
288 if (timer->expires == expires && timer_pending(timer))
291 return __mod_timer(timer, expires);
294 EXPORT_SYMBOL(mod_timer);
297 * del_timer - deactive a timer.
298 * @timer: the timer to be deactivated
300 * del_timer() deactivates a timer - this works on both active and inactive
303 * The function returns whether it has deactivated a pending timer or not.
304 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
305 * active timer returns 1.)
307 int del_timer(struct timer_list *timer)
313 if (timer_pending(timer)) {
314 base = lock_timer_base(timer, &flags);
315 if (timer_pending(timer)) {
316 detach_timer(timer, 1);
319 spin_unlock_irqrestore(&base->lock, flags);
325 EXPORT_SYMBOL(del_timer);
329 * This function tries to deactivate a timer. Upon successful (ret >= 0)
330 * exit the timer is not queued and the handler is not running on any CPU.
332 * It must not be called from interrupt contexts.
334 int try_to_del_timer_sync(struct timer_list *timer)
340 base = lock_timer_base(timer, &flags);
342 if (base->running_timer == timer)
346 if (timer_pending(timer)) {
347 detach_timer(timer, 1);
351 spin_unlock_irqrestore(&base->lock, flags);
357 * del_timer_sync - deactivate a timer and wait for the handler to finish.
358 * @timer: the timer to be deactivated
360 * This function only differs from del_timer() on SMP: besides deactivating
361 * the timer it also makes sure the handler has finished executing on other
364 * Synchronization rules: callers must prevent restarting of the timer,
365 * otherwise this function is meaningless. It must not be called from
366 * interrupt contexts. The caller must not hold locks which would prevent
367 * completion of the timer's handler. The timer's handler must not call
368 * add_timer_on(). Upon exit the timer is not queued and the handler is
369 * not running on any CPU.
371 * The function returns whether it has deactivated a pending timer or not.
373 int del_timer_sync(struct timer_list *timer)
376 int ret = try_to_del_timer_sync(timer);
383 EXPORT_SYMBOL(del_timer_sync);
386 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
388 /* cascade all the timers from tv up one level */
389 struct timer_list *timer, *tmp;
390 struct list_head tv_list;
392 list_replace_init(tv->vec + index, &tv_list);
395 * We are removing _all_ timers from the list, so we
396 * don't have to detach them individually.
398 list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
399 BUG_ON(timer->base != base);
400 internal_add_timer(base, timer);
407 * __run_timers - run all expired timers (if any) on this CPU.
408 * @base: the timer vector to be processed.
410 * This function cascades all vectors and executes all expired timer
413 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
415 static inline void __run_timers(tvec_base_t *base)
417 struct timer_list *timer;
419 spin_lock_irq(&base->lock);
420 while (time_after_eq(jiffies, base->timer_jiffies)) {
421 struct list_head work_list;
422 struct list_head *head = &work_list;
423 int index = base->timer_jiffies & TVR_MASK;
429 (!cascade(base, &base->tv2, INDEX(0))) &&
430 (!cascade(base, &base->tv3, INDEX(1))) &&
431 !cascade(base, &base->tv4, INDEX(2)))
432 cascade(base, &base->tv5, INDEX(3));
433 ++base->timer_jiffies;
434 list_replace_init(base->tv1.vec + index, &work_list);
435 while (!list_empty(head)) {
436 void (*fn)(unsigned long);
439 timer = list_entry(head->next,struct timer_list,entry);
440 fn = timer->function;
443 set_running_timer(base, timer);
444 detach_timer(timer, 1);
445 spin_unlock_irq(&base->lock);
447 int preempt_count = preempt_count();
449 if (preempt_count != preempt_count()) {
450 printk(KERN_WARNING "huh, entered %p "
451 "with preempt_count %08x, exited"
458 spin_lock_irq(&base->lock);
461 set_running_timer(base, NULL);
462 spin_unlock_irq(&base->lock);
465 #ifdef CONFIG_NO_IDLE_HZ
467 * Find out when the next timer event is due to happen. This
468 * is used on S/390 to stop all activity when a cpus is idle.
469 * This functions needs to be called disabled.
471 unsigned long next_timer_interrupt(void)
474 struct list_head *list;
475 struct timer_list *nte;
476 unsigned long expires;
477 unsigned long hr_expires = MAX_JIFFY_OFFSET;
482 hr_delta = hrtimer_get_next_event();
483 if (hr_delta.tv64 != KTIME_MAX) {
484 struct timespec tsdelta;
485 tsdelta = ktime_to_timespec(hr_delta);
486 hr_expires = timespec_to_jiffies(&tsdelta);
488 return hr_expires + jiffies;
490 hr_expires += jiffies;
492 base = __get_cpu_var(tvec_bases);
493 spin_lock(&base->lock);
494 expires = base->timer_jiffies + (LONG_MAX >> 1);
497 /* Look for timer events in tv1. */
498 j = base->timer_jiffies & TVR_MASK;
500 list_for_each_entry(nte, base->tv1.vec + j, entry) {
501 expires = nte->expires;
502 if (j < (base->timer_jiffies & TVR_MASK))
503 list = base->tv2.vec + (INDEX(0));
506 j = (j + 1) & TVR_MASK;
507 } while (j != (base->timer_jiffies & TVR_MASK));
510 varray[0] = &base->tv2;
511 varray[1] = &base->tv3;
512 varray[2] = &base->tv4;
513 varray[3] = &base->tv5;
514 for (i = 0; i < 4; i++) {
517 if (list_empty(varray[i]->vec + j)) {
518 j = (j + 1) & TVN_MASK;
521 list_for_each_entry(nte, varray[i]->vec + j, entry)
522 if (time_before(nte->expires, expires))
523 expires = nte->expires;
524 if (j < (INDEX(i)) && i < 3)
525 list = varray[i + 1]->vec + (INDEX(i + 1));
527 } while (j != (INDEX(i)));
532 * The search wrapped. We need to look at the next list
533 * from next tv element that would cascade into tv element
534 * where we found the timer element.
536 list_for_each_entry(nte, list, entry) {
537 if (time_before(nte->expires, expires))
538 expires = nte->expires;
541 spin_unlock(&base->lock);
544 * It can happen that other CPUs service timer IRQs and increment
545 * jiffies, but we have not yet got a local timer tick to process
546 * the timer wheels. In that case, the expiry time can be before
547 * jiffies, but since the high-resolution timer here is relative to
548 * jiffies, the default expression when high-resolution timers are
551 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
553 * would falsely evaluate to true. If that is the case, just
554 * return jiffies so that we can immediately fire the local timer
556 if (time_before(expires, jiffies))
559 if (time_before(hr_expires, expires))
566 /******************************************************************/
569 * Timekeeping variables
571 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
572 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
576 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
577 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
578 * at zero at system boot time, so wall_to_monotonic will be negative,
579 * however, we will ALWAYS keep the tv_nsec part positive so we can use
580 * the usual normalization.
582 struct timespec xtime __attribute__ ((aligned (16)));
583 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
585 EXPORT_SYMBOL(xtime);
587 /* Don't completely fail for HZ > 500. */
588 int tickadj = 500/HZ ? : 1; /* microsecs */
592 * phase-lock loop variables
594 /* TIME_ERROR prevents overwriting the CMOS clock */
595 int time_state = TIME_OK; /* clock synchronization status */
596 int time_status = STA_UNSYNC; /* clock status bits */
597 long time_offset; /* time adjustment (us) */
598 long time_constant = 2; /* pll time constant */
599 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
600 long time_precision = 1; /* clock precision (us) */
601 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
602 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
603 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
604 /* frequency offset (scaled ppm)*/
605 static long time_adj; /* tick adjust (scaled 1 / HZ) */
606 long time_reftime; /* time at last adjustment (s) */
608 long time_next_adjust;
611 * this routine handles the overflow of the microsecond field
613 * The tricky bits of code to handle the accurate clock support
614 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
615 * They were originally developed for SUN and DEC kernels.
616 * All the kudos should go to Dave for this stuff.
619 static void second_overflow(void)
623 /* Bump the maxerror field */
624 time_maxerror += time_tolerance >> SHIFT_USEC;
625 if (time_maxerror > NTP_PHASE_LIMIT) {
626 time_maxerror = NTP_PHASE_LIMIT;
627 time_status |= STA_UNSYNC;
631 * Leap second processing. If in leap-insert state at the end of the
632 * day, the system clock is set back one second; if in leap-delete
633 * state, the system clock is set ahead one second. The microtime()
634 * routine or external clock driver will insure that reported time is
635 * always monotonic. The ugly divides should be replaced.
637 switch (time_state) {
639 if (time_status & STA_INS)
640 time_state = TIME_INS;
641 else if (time_status & STA_DEL)
642 time_state = TIME_DEL;
645 if (xtime.tv_sec % 86400 == 0) {
647 wall_to_monotonic.tv_sec++;
649 * The timer interpolator will make time change
650 * gradually instead of an immediate jump by one second
652 time_interpolator_update(-NSEC_PER_SEC);
653 time_state = TIME_OOP;
655 printk(KERN_NOTICE "Clock: inserting leap second "
660 if ((xtime.tv_sec + 1) % 86400 == 0) {
662 wall_to_monotonic.tv_sec--;
664 * Use of time interpolator for a gradual change of
667 time_interpolator_update(NSEC_PER_SEC);
668 time_state = TIME_WAIT;
670 printk(KERN_NOTICE "Clock: deleting leap second "
675 time_state = TIME_WAIT;
678 if (!(time_status & (STA_INS | STA_DEL)))
679 time_state = TIME_OK;
683 * Compute the phase adjustment for the next second. In PLL mode, the
684 * offset is reduced by a fixed factor times the time constant. In FLL
685 * mode the offset is used directly. In either mode, the maximum phase
686 * adjustment for each second is clamped so as to spread the adjustment
687 * over not more than the number of seconds between updates.
690 if (!(time_status & STA_FLL))
691 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
692 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
693 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
694 time_offset -= ltemp;
695 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
698 * Compute the frequency estimate and additional phase adjustment due
699 * to frequency error for the next second.
702 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
706 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
707 * get 128.125; => only 0.125% error (p. 14)
709 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
713 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
714 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
716 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
720 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
721 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
723 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
728 * Returns how many microseconds we need to add to xtime this tick
729 * in doing an adjustment requested with adjtime.
731 static long adjtime_adjustment(void)
733 long time_adjust_step;
735 time_adjust_step = time_adjust;
736 if (time_adjust_step) {
738 * We are doing an adjtime thing. Prepare time_adjust_step to
739 * be within bounds. Note that a positive time_adjust means we
740 * want the clock to run faster.
742 * Limit the amount of the step to be in the range
743 * -tickadj .. +tickadj
745 time_adjust_step = min(time_adjust_step, (long)tickadj);
746 time_adjust_step = max(time_adjust_step, (long)-tickadj);
748 return time_adjust_step;
751 /* in the NTP reference this is called "hardclock()" */
752 static void update_ntp_one_tick(void)
754 long time_adjust_step;
756 time_adjust_step = adjtime_adjustment();
757 if (time_adjust_step)
758 /* Reduce by this step the amount of time left */
759 time_adjust -= time_adjust_step;
761 /* Changes by adjtime() do not take effect till next tick. */
762 if (time_next_adjust != 0) {
763 time_adjust = time_next_adjust;
764 time_next_adjust = 0;
769 * Return how long ticks are at the moment, that is, how much time
770 * update_wall_time_one_tick will add to xtime next time we call it
771 * (assuming no calls to do_adjtimex in the meantime).
772 * The return value is in fixed-point nanoseconds shifted by the
773 * specified number of bits to the right of the binary point.
774 * This function has no side-effects.
776 u64 current_tick_length(void)
781 /* calculate the finest interval NTP will allow.
782 * ie: nanosecond value shifted by (SHIFT_SCALE - 10)
784 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
785 ret = (u64)delta_nsec << TICK_LENGTH_SHIFT;
786 ret += (s64)time_adj << (TICK_LENGTH_SHIFT - (SHIFT_SCALE - 10));
791 /* XXX - all of this timekeeping code should be later moved to time.c */
792 #include <linux/clocksource.h>
793 static struct clocksource *clock; /* pointer to current clocksource */
795 #ifdef CONFIG_GENERIC_TIME
797 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
799 * private function, must hold xtime_lock lock when being
800 * called. Returns the number of nanoseconds since the
801 * last call to update_wall_time() (adjusted by NTP scaling)
803 static inline s64 __get_nsec_offset(void)
805 cycle_t cycle_now, cycle_delta;
808 /* read clocksource: */
809 cycle_now = clocksource_read(clock);
811 /* calculate the delta since the last update_wall_time: */
812 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
814 /* convert to nanoseconds: */
815 ns_offset = cyc2ns(clock, cycle_delta);
821 * __get_realtime_clock_ts - Returns the time of day in a timespec
822 * @ts: pointer to the timespec to be set
824 * Returns the time of day in a timespec. Used by
825 * do_gettimeofday() and get_realtime_clock_ts().
827 static inline void __get_realtime_clock_ts(struct timespec *ts)
833 seq = read_seqbegin(&xtime_lock);
836 nsecs = __get_nsec_offset();
838 } while (read_seqretry(&xtime_lock, seq));
840 timespec_add_ns(ts, nsecs);
844 * getnstimeofday - Returns the time of day in a timespec
845 * @ts: pointer to the timespec to be set
847 * Returns the time of day in a timespec.
849 void getnstimeofday(struct timespec *ts)
851 __get_realtime_clock_ts(ts);
854 EXPORT_SYMBOL(getnstimeofday);
857 * do_gettimeofday - Returns the time of day in a timeval
858 * @tv: pointer to the timeval to be set
860 * NOTE: Users should be converted to using get_realtime_clock_ts()
862 void do_gettimeofday(struct timeval *tv)
866 __get_realtime_clock_ts(&now);
867 tv->tv_sec = now.tv_sec;
868 tv->tv_usec = now.tv_nsec/1000;
871 EXPORT_SYMBOL(do_gettimeofday);
873 * do_settimeofday - Sets the time of day
874 * @tv: pointer to the timespec variable containing the new time
876 * Sets the time of day to the new time and update NTP and notify hrtimers
878 int do_settimeofday(struct timespec *tv)
881 time_t wtm_sec, sec = tv->tv_sec;
882 long wtm_nsec, nsec = tv->tv_nsec;
884 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
887 write_seqlock_irqsave(&xtime_lock, flags);
889 nsec -= __get_nsec_offset();
891 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
892 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
894 set_normalized_timespec(&xtime, sec, nsec);
895 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
900 write_sequnlock_irqrestore(&xtime_lock, flags);
902 /* signal hrtimers about time change */
908 EXPORT_SYMBOL(do_settimeofday);
911 * change_clocksource - Swaps clocksources if a new one is available
913 * Accumulates current time interval and initializes new clocksource
915 static int change_clocksource(void)
917 struct clocksource *new;
920 new = clocksource_get_next();
922 now = clocksource_read(new);
923 nsec = __get_nsec_offset();
924 timespec_add_ns(&xtime, nsec);
927 clock->cycle_last = now;
928 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
931 } else if (clock->update_callback) {
932 return clock->update_callback();
937 #define change_clocksource() (0)
941 * timeofday_is_continuous - check to see if timekeeping is free running
943 int timekeeping_is_continuous(void)
949 seq = read_seqbegin(&xtime_lock);
951 ret = clock->is_continuous;
953 } while (read_seqretry(&xtime_lock, seq));
959 * timekeeping_init - Initializes the clocksource and common timekeeping values
961 void __init timekeeping_init(void)
965 write_seqlock_irqsave(&xtime_lock, flags);
966 clock = clocksource_get_next();
967 clocksource_calculate_interval(clock, tick_nsec);
968 clock->cycle_last = clocksource_read(clock);
970 write_sequnlock_irqrestore(&xtime_lock, flags);
974 static int timekeeping_suspended;
976 * timekeeping_resume - Resumes the generic timekeeping subsystem.
979 * This is for the generic clocksource timekeeping.
980 * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
981 * still managed by arch specific suspend/resume code.
983 static int timekeeping_resume(struct sys_device *dev)
987 write_seqlock_irqsave(&xtime_lock, flags);
988 /* restart the last cycle value */
989 clock->cycle_last = clocksource_read(clock);
991 timekeeping_suspended = 0;
992 write_sequnlock_irqrestore(&xtime_lock, flags);
996 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
1000 write_seqlock_irqsave(&xtime_lock, flags);
1001 timekeeping_suspended = 1;
1002 write_sequnlock_irqrestore(&xtime_lock, flags);
1006 /* sysfs resume/suspend bits for timekeeping */
1007 static struct sysdev_class timekeeping_sysclass = {
1008 .resume = timekeeping_resume,
1009 .suspend = timekeeping_suspend,
1010 set_kset_name("timekeeping"),
1013 static struct sys_device device_timer = {
1015 .cls = &timekeeping_sysclass,
1018 static int __init timekeeping_init_device(void)
1020 int error = sysdev_class_register(&timekeeping_sysclass);
1022 error = sysdev_register(&device_timer);
1026 device_initcall(timekeeping_init_device);
1029 * If the error is already larger, we look ahead even further
1030 * to compensate for late or lost adjustments.
1032 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, s64 *offset)
1035 u32 look_ahead, adj;
1039 * Use the current error value to determine how much to look ahead.
1040 * The larger the error the slower we adjust for it to avoid problems
1041 * with losing too many ticks, otherwise we would overadjust and
1042 * produce an even larger error. The smaller the adjustment the
1043 * faster we try to adjust for it, as lost ticks can do less harm
1044 * here. This is tuned so that an error of about 1 msec is adusted
1045 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1047 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
1048 error2 = abs(error2);
1049 for (look_ahead = 0; error2 > 0; look_ahead++)
1053 * Now calculate the error in (1 << look_ahead) ticks, but first
1054 * remove the single look ahead already included in the error.
1056 tick_error = current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
1057 tick_error -= clock->xtime_interval >> 1;
1058 error = ((error - tick_error) >> look_ahead) + tick_error;
1060 /* Finally calculate the adjustment shift value. */
1065 *interval = -*interval;
1069 for (adj = 0; error > i; adj++)
1078 * Adjust the multiplier to reduce the error value,
1079 * this is optimized for the most common adjustments of -1,0,1,
1080 * for other values we can do a bit more work.
1082 static void clocksource_adjust(struct clocksource *clock, s64 offset)
1084 s64 error, interval = clock->cycle_interval;
1087 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
1088 if (error > interval) {
1090 if (likely(error <= interval))
1093 adj = clocksource_bigadjust(error, &interval, &offset);
1094 } else if (error < -interval) {
1096 if (likely(error >= -interval)) {
1098 interval = -interval;
1101 adj = clocksource_bigadjust(error, &interval, &offset);
1106 clock->xtime_interval += interval;
1107 clock->xtime_nsec -= offset;
1108 clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
1112 * update_wall_time - Uses the current clocksource to increment the wall time
1114 * Called from the timer interrupt, must hold a write on xtime_lock.
1116 static void update_wall_time(void)
1120 /* Make sure we're fully resumed: */
1121 if (unlikely(timekeeping_suspended))
1124 #ifdef CONFIG_GENERIC_TIME
1125 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
1127 offset = clock->cycle_interval;
1129 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
1131 /* normally this loop will run just once, however in the
1132 * case of lost or late ticks, it will accumulate correctly.
1134 while (offset >= clock->cycle_interval) {
1135 /* accumulate one interval */
1136 clock->xtime_nsec += clock->xtime_interval;
1137 clock->cycle_last += clock->cycle_interval;
1138 offset -= clock->cycle_interval;
1140 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
1141 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
1146 /* interpolator bits */
1147 time_interpolator_update(clock->xtime_interval
1149 /* increment the NTP state machine */
1150 update_ntp_one_tick();
1152 /* accumulate error between NTP and clock interval */
1153 clock->error += current_tick_length();
1154 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
1157 /* correct the clock when NTP error is too big */
1158 clocksource_adjust(clock, offset);
1160 /* store full nanoseconds into xtime */
1161 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
1162 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
1164 /* check to see if there is a new clocksource to use */
1165 if (change_clocksource()) {
1167 clock->xtime_nsec = 0;
1168 clocksource_calculate_interval(clock, tick_nsec);
1173 * Called from the timer interrupt handler to charge one tick to the current
1174 * process. user_tick is 1 if the tick is user time, 0 for system.
1176 void update_process_times(int user_tick)
1178 struct task_struct *p = current;
1179 int cpu = smp_processor_id();
1181 /* Note: this timer irq context must be accounted for as well. */
1183 account_user_time(p, jiffies_to_cputime(1));
1185 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
1187 if (rcu_pending(cpu))
1188 rcu_check_callbacks(cpu, user_tick);
1190 run_posix_cpu_timers(p);
1194 * Nr of active tasks - counted in fixed-point numbers
1196 static unsigned long count_active_tasks(void)
1198 return nr_active() * FIXED_1;
1202 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1203 * imply that avenrun[] is the standard name for this kind of thing.
1204 * Nothing else seems to be standardized: the fractional size etc
1205 * all seem to differ on different machines.
1207 * Requires xtime_lock to access.
1209 unsigned long avenrun[3];
1211 EXPORT_SYMBOL(avenrun);
1214 * calc_load - given tick count, update the avenrun load estimates.
1215 * This is called while holding a write_lock on xtime_lock.
1217 static inline void calc_load(unsigned long ticks)
1219 unsigned long active_tasks; /* fixed-point */
1220 static int count = LOAD_FREQ;
1225 active_tasks = count_active_tasks();
1226 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
1227 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
1228 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
1232 /* jiffies at the most recent update of wall time */
1233 unsigned long wall_jiffies = INITIAL_JIFFIES;
1236 * This read-write spinlock protects us from races in SMP while
1237 * playing with xtime and avenrun.
1239 #ifndef ARCH_HAVE_XTIME_LOCK
1240 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
1242 EXPORT_SYMBOL(xtime_lock);
1246 * This function runs timers and the timer-tq in bottom half context.
1248 static void run_timer_softirq(struct softirq_action *h)
1250 tvec_base_t *base = __get_cpu_var(tvec_bases);
1252 hrtimer_run_queues();
1253 if (time_after_eq(jiffies, base->timer_jiffies))
1258 * Called by the local, per-CPU timer interrupt on SMP.
1260 void run_local_timers(void)
1262 raise_softirq(TIMER_SOFTIRQ);
1267 * Called by the timer interrupt. xtime_lock must already be taken
1270 static inline void update_times(void)
1272 unsigned long ticks;
1274 ticks = jiffies - wall_jiffies;
1275 wall_jiffies += ticks;
1281 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1282 * without sampling the sequence number in xtime_lock.
1283 * jiffies is defined in the linker script...
1286 void do_timer(struct pt_regs *regs)
1289 /* prevent loading jiffies before storing new jiffies_64 value. */
1294 #ifdef __ARCH_WANT_SYS_ALARM
1297 * For backwards compatibility? This can be done in libc so Alpha
1298 * and all newer ports shouldn't need it.
1300 asmlinkage unsigned long sys_alarm(unsigned int seconds)
1302 return alarm_setitimer(seconds);
1309 * sys_getpid - return the thread group id of the current process
1311 * Note, despite the name, this returns the tgid not the pid. The tgid and
1312 * the pid are identical unless CLONE_THREAD was specified on clone() in
1313 * which case the tgid is the same in all threads of the same group.
1315 * This is SMP safe as current->tgid does not change.
1317 asmlinkage long sys_getpid(void)
1319 return vx_map_tgid(current->tgid);
1323 * Accessing ->parent is not SMP-safe, it could
1324 * change from under us. However, we can use a stale
1325 * value of ->real_parent under rcu_read_lock(), see
1326 * release_task()->call_rcu(delayed_put_task_struct).
1328 asmlinkage long sys_getppid(void)
1333 pid = rcu_dereference(current->parent)->tgid;
1335 return vx_map_pid(pid);
1341 * The Alpha uses getxpid, getxuid, and getxgid instead.
1344 asmlinkage long do_getxpid(long *ppid)
1346 *ppid = sys_getppid();
1347 return sys_getpid();
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.
1436 printk(KERN_ERR "schedule_timeout: wrong timeout "
1437 "value %lx from %p\n", timeout,
1438 __builtin_return_address(0));
1439 current->state = TASK_RUNNING;
1444 expire = timeout + jiffies;
1446 setup_timer(&timer, process_timeout, (unsigned long)current);
1447 __mod_timer(&timer, expire);
1449 del_singleshot_timer_sync(&timer);
1451 timeout = expire - jiffies;
1454 return timeout < 0 ? 0 : timeout;
1456 EXPORT_SYMBOL(schedule_timeout);
1459 * We can use __set_current_state() here because schedule_timeout() calls
1460 * schedule() unconditionally.
1462 signed long __sched schedule_timeout_interruptible(signed long timeout)
1464 __set_current_state(TASK_INTERRUPTIBLE);
1465 return schedule_timeout(timeout);
1467 EXPORT_SYMBOL(schedule_timeout_interruptible);
1469 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1471 __set_current_state(TASK_UNINTERRUPTIBLE);
1472 return schedule_timeout(timeout);
1474 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1476 /* Thread ID - the internal kernel "pid" */
1477 asmlinkage long sys_gettid(void)
1479 return current->pid;
1483 * sys_sysinfo - fill in sysinfo struct
1485 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1488 unsigned long mem_total, sav_total;
1489 unsigned int mem_unit, bitcount;
1492 memset((char *)&val, 0, sizeof(struct sysinfo));
1496 seq = read_seqbegin(&xtime_lock);
1499 * This is annoying. The below is the same thing
1500 * posix_get_clock_monotonic() does, but it wants to
1501 * take the lock which we want to cover the loads stuff
1505 getnstimeofday(&tp);
1506 tp.tv_sec += wall_to_monotonic.tv_sec;
1507 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1508 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1509 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1512 if (vx_flags(VXF_VIRT_UPTIME, 0))
1513 vx_vsi_uptime(&tp, NULL);
1514 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1516 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1517 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1518 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1520 val.procs = nr_threads;
1521 } while (read_seqretry(&xtime_lock, seq));
1527 * If the sum of all the available memory (i.e. ram + swap)
1528 * is less than can be stored in a 32 bit unsigned long then
1529 * we can be binary compatible with 2.2.x kernels. If not,
1530 * well, in that case 2.2.x was broken anyways...
1532 * -Erik Andersen <andersee@debian.org>
1535 mem_total = val.totalram + val.totalswap;
1536 if (mem_total < val.totalram || mem_total < val.totalswap)
1539 mem_unit = val.mem_unit;
1540 while (mem_unit > 1) {
1543 sav_total = mem_total;
1545 if (mem_total < sav_total)
1550 * If mem_total did not overflow, multiply all memory values by
1551 * val.mem_unit and set it to 1. This leaves things compatible
1552 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1557 val.totalram <<= bitcount;
1558 val.freeram <<= bitcount;
1559 val.sharedram <<= bitcount;
1560 val.bufferram <<= bitcount;
1561 val.totalswap <<= bitcount;
1562 val.freeswap <<= bitcount;
1563 val.totalhigh <<= bitcount;
1564 val.freehigh <<= bitcount;
1567 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1574 * lockdep: we want to track each per-CPU base as a separate lock-class,
1575 * but timer-bases are kmalloc()-ed, so we need to attach separate
1578 static struct lock_class_key base_lock_keys[NR_CPUS];
1580 static int __devinit init_timers_cpu(int cpu)
1584 static char __devinitdata tvec_base_done[NR_CPUS];
1586 if (!tvec_base_done[cpu]) {
1587 static char boot_done;
1591 * The APs use this path later in boot
1593 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1597 memset(base, 0, sizeof(*base));
1598 per_cpu(tvec_bases, cpu) = base;
1601 * This is for the boot CPU - we use compile-time
1602 * static initialisation because per-cpu memory isn't
1603 * ready yet and because the memory allocators are not
1604 * initialised either.
1607 base = &boot_tvec_bases;
1609 tvec_base_done[cpu] = 1;
1611 base = per_cpu(tvec_bases, cpu);
1614 spin_lock_init(&base->lock);
1615 lockdep_set_class(&base->lock, base_lock_keys + cpu);
1617 for (j = 0; j < TVN_SIZE; j++) {
1618 INIT_LIST_HEAD(base->tv5.vec + j);
1619 INIT_LIST_HEAD(base->tv4.vec + j);
1620 INIT_LIST_HEAD(base->tv3.vec + j);
1621 INIT_LIST_HEAD(base->tv2.vec + j);
1623 for (j = 0; j < TVR_SIZE; j++)
1624 INIT_LIST_HEAD(base->tv1.vec + j);
1626 base->timer_jiffies = jiffies;
1630 #ifdef CONFIG_HOTPLUG_CPU
1631 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1633 struct timer_list *timer;
1635 while (!list_empty(head)) {
1636 timer = list_entry(head->next, struct timer_list, entry);
1637 detach_timer(timer, 0);
1638 timer->base = new_base;
1639 internal_add_timer(new_base, timer);
1643 static void __devinit migrate_timers(int cpu)
1645 tvec_base_t *old_base;
1646 tvec_base_t *new_base;
1649 BUG_ON(cpu_online(cpu));
1650 old_base = per_cpu(tvec_bases, cpu);
1651 new_base = get_cpu_var(tvec_bases);
1653 local_irq_disable();
1654 spin_lock(&new_base->lock);
1655 spin_lock(&old_base->lock);
1657 BUG_ON(old_base->running_timer);
1659 for (i = 0; i < TVR_SIZE; i++)
1660 migrate_timer_list(new_base, old_base->tv1.vec + i);
1661 for (i = 0; i < TVN_SIZE; i++) {
1662 migrate_timer_list(new_base, old_base->tv2.vec + i);
1663 migrate_timer_list(new_base, old_base->tv3.vec + i);
1664 migrate_timer_list(new_base, old_base->tv4.vec + i);
1665 migrate_timer_list(new_base, old_base->tv5.vec + i);
1668 spin_unlock(&old_base->lock);
1669 spin_unlock(&new_base->lock);
1671 put_cpu_var(tvec_bases);
1673 #endif /* CONFIG_HOTPLUG_CPU */
1675 static int __cpuinit timer_cpu_notify(struct notifier_block *self,
1676 unsigned long action, void *hcpu)
1678 long cpu = (long)hcpu;
1680 case CPU_UP_PREPARE:
1681 if (init_timers_cpu(cpu) < 0)
1684 #ifdef CONFIG_HOTPLUG_CPU
1686 migrate_timers(cpu);
1695 static struct notifier_block __cpuinitdata timers_nb = {
1696 .notifier_call = timer_cpu_notify,
1700 void __init init_timers(void)
1702 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1703 (void *)(long)smp_processor_id());
1704 register_cpu_notifier(&timers_nb);
1705 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1708 #ifdef CONFIG_TIME_INTERPOLATION
1710 struct time_interpolator *time_interpolator __read_mostly;
1711 static struct time_interpolator *time_interpolator_list __read_mostly;
1712 static DEFINE_SPINLOCK(time_interpolator_lock);
1714 static inline u64 time_interpolator_get_cycles(unsigned int src)
1716 unsigned long (*x)(void);
1720 case TIME_SOURCE_FUNCTION:
1721 x = time_interpolator->addr;
1724 case TIME_SOURCE_MMIO64 :
1725 return readq_relaxed((void __iomem *)time_interpolator->addr);
1727 case TIME_SOURCE_MMIO32 :
1728 return readl_relaxed((void __iomem *)time_interpolator->addr);
1730 default: return get_cycles();
1734 static inline u64 time_interpolator_get_counter(int writelock)
1736 unsigned int src = time_interpolator->source;
1738 if (time_interpolator->jitter)
1744 lcycle = time_interpolator->last_cycle;
1745 now = time_interpolator_get_cycles(src);
1746 if (lcycle && time_after(lcycle, now))
1749 /* When holding the xtime write lock, there's no need
1750 * to add the overhead of the cmpxchg. Readers are
1751 * force to retry until the write lock is released.
1754 time_interpolator->last_cycle = now;
1757 /* Keep track of the last timer value returned. The use of cmpxchg here
1758 * will cause contention in an SMP environment.
1760 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1764 return time_interpolator_get_cycles(src);
1767 void time_interpolator_reset(void)
1769 time_interpolator->offset = 0;
1770 time_interpolator->last_counter = time_interpolator_get_counter(1);
1773 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1775 unsigned long time_interpolator_get_offset(void)
1777 /* If we do not have a time interpolator set up then just return zero */
1778 if (!time_interpolator)
1781 return time_interpolator->offset +
1782 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1785 #define INTERPOLATOR_ADJUST 65536
1786 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1788 static void time_interpolator_update(long delta_nsec)
1791 unsigned long offset;
1793 /* If there is no time interpolator set up then do nothing */
1794 if (!time_interpolator)
1798 * The interpolator compensates for late ticks by accumulating the late
1799 * time in time_interpolator->offset. A tick earlier than expected will
1800 * lead to a reset of the offset and a corresponding jump of the clock
1801 * forward. Again this only works if the interpolator clock is running
1802 * slightly slower than the regular clock and the tuning logic insures
1806 counter = time_interpolator_get_counter(1);
1807 offset = time_interpolator->offset +
1808 GET_TI_NSECS(counter, time_interpolator);
1810 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1811 time_interpolator->offset = offset - delta_nsec;
1813 time_interpolator->skips++;
1814 time_interpolator->ns_skipped += delta_nsec - offset;
1815 time_interpolator->offset = 0;
1817 time_interpolator->last_counter = counter;
1819 /* Tuning logic for time interpolator invoked every minute or so.
1820 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1821 * Increase interpolator clock speed if we skip too much time.
1823 if (jiffies % INTERPOLATOR_ADJUST == 0)
1825 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1826 time_interpolator->nsec_per_cyc--;
1827 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1828 time_interpolator->nsec_per_cyc++;
1829 time_interpolator->skips = 0;
1830 time_interpolator->ns_skipped = 0;
1835 is_better_time_interpolator(struct time_interpolator *new)
1837 if (!time_interpolator)
1839 return new->frequency > 2*time_interpolator->frequency ||
1840 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1844 register_time_interpolator(struct time_interpolator *ti)
1846 unsigned long flags;
1849 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1851 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1852 spin_lock(&time_interpolator_lock);
1853 write_seqlock_irqsave(&xtime_lock, flags);
1854 if (is_better_time_interpolator(ti)) {
1855 time_interpolator = ti;
1856 time_interpolator_reset();
1858 write_sequnlock_irqrestore(&xtime_lock, flags);
1860 ti->next = time_interpolator_list;
1861 time_interpolator_list = ti;
1862 spin_unlock(&time_interpolator_lock);
1866 unregister_time_interpolator(struct time_interpolator *ti)
1868 struct time_interpolator *curr, **prev;
1869 unsigned long flags;
1871 spin_lock(&time_interpolator_lock);
1872 prev = &time_interpolator_list;
1873 for (curr = *prev; curr; curr = curr->next) {
1881 write_seqlock_irqsave(&xtime_lock, flags);
1882 if (ti == time_interpolator) {
1883 /* we lost the best time-interpolator: */
1884 time_interpolator = NULL;
1885 /* find the next-best interpolator */
1886 for (curr = time_interpolator_list; curr; curr = curr->next)
1887 if (is_better_time_interpolator(curr))
1888 time_interpolator = curr;
1889 time_interpolator_reset();
1891 write_sequnlock_irqrestore(&xtime_lock, flags);
1892 spin_unlock(&time_interpolator_lock);
1894 #endif /* CONFIG_TIME_INTERPOLATION */
1897 * msleep - sleep safely even with waitqueue interruptions
1898 * @msecs: Time in milliseconds to sleep for
1900 void msleep(unsigned int msecs)
1902 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1905 timeout = schedule_timeout_uninterruptible(timeout);
1908 EXPORT_SYMBOL(msleep);
1911 * msleep_interruptible - sleep waiting for signals
1912 * @msecs: Time in milliseconds to sleep for
1914 unsigned long msleep_interruptible(unsigned int msecs)
1916 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1918 while (timeout && !signal_pending(current))
1919 timeout = schedule_timeout_interruptible(timeout);
1920 return jiffies_to_msecs(timeout);
1923 EXPORT_SYMBOL(msleep_interruptible);