V4L/DVB (3320): Check if PLL-description is set
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / timer.c
blob074b4bd5cfd8b62a9b92555514455a4a06f0416e
1 /*
2 * linux/kernel/timer.c
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>
27 #include <linux/mm.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>
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/div64.h>
41 #include <asm/timex.h>
42 #include <asm/io.h>
44 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec);
46 #else
47 #define time_interpolator_update(x)
48 #endif
50 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
52 EXPORT_SYMBOL(jiffies_64);
55 * per-CPU timer vector definitions:
58 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
59 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
60 #define TVN_SIZE (1 << TVN_BITS)
61 #define TVR_SIZE (1 << TVR_BITS)
62 #define TVN_MASK (TVN_SIZE - 1)
63 #define TVR_MASK (TVR_SIZE - 1)
65 struct timer_base_s {
66 spinlock_t lock;
67 struct timer_list *running_timer;
70 typedef struct tvec_s {
71 struct list_head vec[TVN_SIZE];
72 } tvec_t;
74 typedef struct tvec_root_s {
75 struct list_head vec[TVR_SIZE];
76 } tvec_root_t;
78 struct tvec_t_base_s {
79 struct timer_base_s t_base;
80 unsigned long timer_jiffies;
81 tvec_root_t tv1;
82 tvec_t tv2;
83 tvec_t tv3;
84 tvec_t tv4;
85 tvec_t tv5;
86 } ____cacheline_aligned_in_smp;
88 typedef struct tvec_t_base_s tvec_base_t;
89 static DEFINE_PER_CPU(tvec_base_t, tvec_bases);
91 static inline void set_running_timer(tvec_base_t *base,
92 struct timer_list *timer)
94 #ifdef CONFIG_SMP
95 base->t_base.running_timer = timer;
96 #endif
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);
123 } else {
124 int i;
125 /* If the timeout is larger than 0xffffffff on 64-bit
126 * architectures then we use the maximum timeout:
128 if (idx > 0xffffffffUL) {
129 idx = 0xffffffffUL;
130 expires = idx + base->timer_jiffies;
132 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
133 vec = base->tv5.vec + i;
136 * Timers are FIFO:
138 list_add_tail(&timer->entry, vec);
141 typedef struct timer_base_s timer_base_t;
143 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
144 * at compile time, and we need timer->base to lock the timer.
146 timer_base_t __init_timer_base
147 ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED };
148 EXPORT_SYMBOL(__init_timer_base);
150 /***
151 * init_timer - initialize a timer.
152 * @timer: the timer to be initialized
154 * init_timer() must be done to a timer prior calling *any* of the
155 * other timer functions.
157 void fastcall init_timer(struct timer_list *timer)
159 timer->entry.next = NULL;
160 timer->base = &per_cpu(tvec_bases, raw_smp_processor_id()).t_base;
162 EXPORT_SYMBOL(init_timer);
164 static inline void detach_timer(struct timer_list *timer,
165 int clear_pending)
167 struct list_head *entry = &timer->entry;
169 __list_del(entry->prev, entry->next);
170 if (clear_pending)
171 entry->next = NULL;
172 entry->prev = LIST_POISON2;
176 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
177 * means that all timers which are tied to this base via timer->base are
178 * locked, and the base itself is locked too.
180 * So __run_timers/migrate_timers can safely modify all timers which could
181 * be found on ->tvX lists.
183 * When the timer's base is locked, and the timer removed from list, it is
184 * possible to set timer->base = NULL and drop the lock: the timer remains
185 * locked.
187 static timer_base_t *lock_timer_base(struct timer_list *timer,
188 unsigned long *flags)
190 timer_base_t *base;
192 for (;;) {
193 base = timer->base;
194 if (likely(base != NULL)) {
195 spin_lock_irqsave(&base->lock, *flags);
196 if (likely(base == timer->base))
197 return base;
198 /* The timer has migrated to another CPU */
199 spin_unlock_irqrestore(&base->lock, *flags);
201 cpu_relax();
205 int __mod_timer(struct timer_list *timer, unsigned long expires)
207 timer_base_t *base;
208 tvec_base_t *new_base;
209 unsigned long flags;
210 int ret = 0;
212 BUG_ON(!timer->function);
214 base = lock_timer_base(timer, &flags);
216 if (timer_pending(timer)) {
217 detach_timer(timer, 0);
218 ret = 1;
221 new_base = &__get_cpu_var(tvec_bases);
223 if (base != &new_base->t_base) {
225 * We are trying to schedule the timer on the local CPU.
226 * However we can't change timer's base while it is running,
227 * otherwise del_timer_sync() can't detect that the timer's
228 * handler yet has not finished. This also guarantees that
229 * the timer is serialized wrt itself.
231 if (unlikely(base->running_timer == timer)) {
232 /* The timer remains on a former base */
233 new_base = container_of(base, tvec_base_t, t_base);
234 } else {
235 /* See the comment in lock_timer_base() */
236 timer->base = NULL;
237 spin_unlock(&base->lock);
238 spin_lock(&new_base->t_base.lock);
239 timer->base = &new_base->t_base;
243 timer->expires = expires;
244 internal_add_timer(new_base, timer);
245 spin_unlock_irqrestore(&new_base->t_base.lock, flags);
247 return ret;
250 EXPORT_SYMBOL(__mod_timer);
252 /***
253 * add_timer_on - start a timer on a particular CPU
254 * @timer: the timer to be added
255 * @cpu: the CPU to start it on
257 * This is not very scalable on SMP. Double adds are not possible.
259 void add_timer_on(struct timer_list *timer, int cpu)
261 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
262 unsigned long flags;
264 BUG_ON(timer_pending(timer) || !timer->function);
265 spin_lock_irqsave(&base->t_base.lock, flags);
266 timer->base = &base->t_base;
267 internal_add_timer(base, timer);
268 spin_unlock_irqrestore(&base->t_base.lock, flags);
272 /***
273 * mod_timer - modify a timer's timeout
274 * @timer: the timer to be modified
276 * mod_timer is a more efficient way to update the expire field of an
277 * active timer (if the timer is inactive it will be activated)
279 * mod_timer(timer, expires) is equivalent to:
281 * del_timer(timer); timer->expires = expires; add_timer(timer);
283 * Note that if there are multiple unserialized concurrent users of the
284 * same timer, then mod_timer() is the only safe way to modify the timeout,
285 * since add_timer() cannot modify an already running timer.
287 * The function returns whether it has modified a pending timer or not.
288 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
289 * active timer returns 1.)
291 int mod_timer(struct timer_list *timer, unsigned long expires)
293 BUG_ON(!timer->function);
296 * This is a common optimization triggered by the
297 * networking code - if the timer is re-modified
298 * to be the same thing then just return:
300 if (timer->expires == expires && timer_pending(timer))
301 return 1;
303 return __mod_timer(timer, expires);
306 EXPORT_SYMBOL(mod_timer);
308 /***
309 * del_timer - deactive a timer.
310 * @timer: the timer to be deactivated
312 * del_timer() deactivates a timer - this works on both active and inactive
313 * timers.
315 * The function returns whether it has deactivated a pending timer or not.
316 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
317 * active timer returns 1.)
319 int del_timer(struct timer_list *timer)
321 timer_base_t *base;
322 unsigned long flags;
323 int ret = 0;
325 if (timer_pending(timer)) {
326 base = lock_timer_base(timer, &flags);
327 if (timer_pending(timer)) {
328 detach_timer(timer, 1);
329 ret = 1;
331 spin_unlock_irqrestore(&base->lock, flags);
334 return ret;
337 EXPORT_SYMBOL(del_timer);
339 #ifdef CONFIG_SMP
341 * This function tries to deactivate a timer. Upon successful (ret >= 0)
342 * exit the timer is not queued and the handler is not running on any CPU.
344 * It must not be called from interrupt contexts.
346 int try_to_del_timer_sync(struct timer_list *timer)
348 timer_base_t *base;
349 unsigned long flags;
350 int ret = -1;
352 base = lock_timer_base(timer, &flags);
354 if (base->running_timer == timer)
355 goto out;
357 ret = 0;
358 if (timer_pending(timer)) {
359 detach_timer(timer, 1);
360 ret = 1;
362 out:
363 spin_unlock_irqrestore(&base->lock, flags);
365 return ret;
368 /***
369 * del_timer_sync - deactivate a timer and wait for the handler to finish.
370 * @timer: the timer to be deactivated
372 * This function only differs from del_timer() on SMP: besides deactivating
373 * the timer it also makes sure the handler has finished executing on other
374 * CPUs.
376 * Synchronization rules: callers must prevent restarting of the timer,
377 * otherwise this function is meaningless. It must not be called from
378 * interrupt contexts. The caller must not hold locks which would prevent
379 * completion of the timer's handler. The timer's handler must not call
380 * add_timer_on(). Upon exit the timer is not queued and the handler is
381 * not running on any CPU.
383 * The function returns whether it has deactivated a pending timer or not.
385 int del_timer_sync(struct timer_list *timer)
387 for (;;) {
388 int ret = try_to_del_timer_sync(timer);
389 if (ret >= 0)
390 return ret;
394 EXPORT_SYMBOL(del_timer_sync);
395 #endif
397 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
399 /* cascade all the timers from tv up one level */
400 struct list_head *head, *curr;
402 head = tv->vec + index;
403 curr = head->next;
405 * We are removing _all_ timers from the list, so we don't have to
406 * detach them individually, just clear the list afterwards.
408 while (curr != head) {
409 struct timer_list *tmp;
411 tmp = list_entry(curr, struct timer_list, entry);
412 BUG_ON(tmp->base != &base->t_base);
413 curr = curr->next;
414 internal_add_timer(base, tmp);
416 INIT_LIST_HEAD(head);
418 return index;
421 /***
422 * __run_timers - run all expired timers (if any) on this CPU.
423 * @base: the timer vector to be processed.
425 * This function cascades all vectors and executes all expired timer
426 * vectors.
428 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
430 static inline void __run_timers(tvec_base_t *base)
432 struct timer_list *timer;
434 spin_lock_irq(&base->t_base.lock);
435 while (time_after_eq(jiffies, base->timer_jiffies)) {
436 struct list_head work_list = LIST_HEAD_INIT(work_list);
437 struct list_head *head = &work_list;
438 int index = base->timer_jiffies & TVR_MASK;
441 * Cascade timers:
443 if (!index &&
444 (!cascade(base, &base->tv2, INDEX(0))) &&
445 (!cascade(base, &base->tv3, INDEX(1))) &&
446 !cascade(base, &base->tv4, INDEX(2)))
447 cascade(base, &base->tv5, INDEX(3));
448 ++base->timer_jiffies;
449 list_splice_init(base->tv1.vec + index, &work_list);
450 while (!list_empty(head)) {
451 void (*fn)(unsigned long);
452 unsigned long data;
454 timer = list_entry(head->next,struct timer_list,entry);
455 fn = timer->function;
456 data = timer->data;
458 set_running_timer(base, timer);
459 detach_timer(timer, 1);
460 spin_unlock_irq(&base->t_base.lock);
462 int preempt_count = preempt_count();
463 fn(data);
464 if (preempt_count != preempt_count()) {
465 printk(KERN_WARNING "huh, entered %p "
466 "with preempt_count %08x, exited"
467 " with %08x?\n",
468 fn, preempt_count,
469 preempt_count());
470 BUG();
473 spin_lock_irq(&base->t_base.lock);
476 set_running_timer(base, NULL);
477 spin_unlock_irq(&base->t_base.lock);
480 #ifdef CONFIG_NO_IDLE_HZ
482 * Find out when the next timer event is due to happen. This
483 * is used on S/390 to stop all activity when a cpus is idle.
484 * This functions needs to be called disabled.
486 unsigned long next_timer_interrupt(void)
488 tvec_base_t *base;
489 struct list_head *list;
490 struct timer_list *nte;
491 unsigned long expires;
492 tvec_t *varray[4];
493 int i, j;
495 base = &__get_cpu_var(tvec_bases);
496 spin_lock(&base->t_base.lock);
497 expires = base->timer_jiffies + (LONG_MAX >> 1);
498 list = 0;
500 /* Look for timer events in tv1. */
501 j = base->timer_jiffies & TVR_MASK;
502 do {
503 list_for_each_entry(nte, base->tv1.vec + j, entry) {
504 expires = nte->expires;
505 if (j < (base->timer_jiffies & TVR_MASK))
506 list = base->tv2.vec + (INDEX(0));
507 goto found;
509 j = (j + 1) & TVR_MASK;
510 } while (j != (base->timer_jiffies & TVR_MASK));
512 /* Check tv2-tv5. */
513 varray[0] = &base->tv2;
514 varray[1] = &base->tv3;
515 varray[2] = &base->tv4;
516 varray[3] = &base->tv5;
517 for (i = 0; i < 4; i++) {
518 j = INDEX(i);
519 do {
520 if (list_empty(varray[i]->vec + j)) {
521 j = (j + 1) & TVN_MASK;
522 continue;
524 list_for_each_entry(nte, varray[i]->vec + j, entry)
525 if (time_before(nte->expires, expires))
526 expires = nte->expires;
527 if (j < (INDEX(i)) && i < 3)
528 list = varray[i + 1]->vec + (INDEX(i + 1));
529 goto found;
530 } while (j != (INDEX(i)));
532 found:
533 if (list) {
535 * The search wrapped. We need to look at the next list
536 * from next tv element that would cascade into tv element
537 * where we found the timer element.
539 list_for_each_entry(nte, list, entry) {
540 if (time_before(nte->expires, expires))
541 expires = nte->expires;
544 spin_unlock(&base->t_base.lock);
545 return expires;
547 #endif
549 /******************************************************************/
552 * Timekeeping variables
554 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
555 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
558 * The current time
559 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
560 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
561 * at zero at system boot time, so wall_to_monotonic will be negative,
562 * however, we will ALWAYS keep the tv_nsec part positive so we can use
563 * the usual normalization.
565 struct timespec xtime __attribute__ ((aligned (16)));
566 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
568 EXPORT_SYMBOL(xtime);
570 /* Don't completely fail for HZ > 500. */
571 int tickadj = 500/HZ ? : 1; /* microsecs */
575 * phase-lock loop variables
577 /* TIME_ERROR prevents overwriting the CMOS clock */
578 int time_state = TIME_OK; /* clock synchronization status */
579 int time_status = STA_UNSYNC; /* clock status bits */
580 long time_offset; /* time adjustment (us) */
581 long time_constant = 2; /* pll time constant */
582 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
583 long time_precision = 1; /* clock precision (us) */
584 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
585 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
586 static long time_phase; /* phase offset (scaled us) */
587 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
588 /* frequency offset (scaled ppm)*/
589 static long time_adj; /* tick adjust (scaled 1 / HZ) */
590 long time_reftime; /* time at last adjustment (s) */
591 long time_adjust;
592 long time_next_adjust;
595 * this routine handles the overflow of the microsecond field
597 * The tricky bits of code to handle the accurate clock support
598 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
599 * They were originally developed for SUN and DEC kernels.
600 * All the kudos should go to Dave for this stuff.
603 static void second_overflow(void)
605 long ltemp;
607 /* Bump the maxerror field */
608 time_maxerror += time_tolerance >> SHIFT_USEC;
609 if (time_maxerror > NTP_PHASE_LIMIT) {
610 time_maxerror = NTP_PHASE_LIMIT;
611 time_status |= STA_UNSYNC;
615 * Leap second processing. If in leap-insert state at the end of the
616 * day, the system clock is set back one second; if in leap-delete
617 * state, the system clock is set ahead one second. The microtime()
618 * routine or external clock driver will insure that reported time is
619 * always monotonic. The ugly divides should be replaced.
621 switch (time_state) {
622 case TIME_OK:
623 if (time_status & STA_INS)
624 time_state = TIME_INS;
625 else if (time_status & STA_DEL)
626 time_state = TIME_DEL;
627 break;
628 case TIME_INS:
629 if (xtime.tv_sec % 86400 == 0) {
630 xtime.tv_sec--;
631 wall_to_monotonic.tv_sec++;
633 * The timer interpolator will make time change
634 * gradually instead of an immediate jump by one second
636 time_interpolator_update(-NSEC_PER_SEC);
637 time_state = TIME_OOP;
638 clock_was_set();
639 printk(KERN_NOTICE "Clock: inserting leap second "
640 "23:59:60 UTC\n");
642 break;
643 case TIME_DEL:
644 if ((xtime.tv_sec + 1) % 86400 == 0) {
645 xtime.tv_sec++;
646 wall_to_monotonic.tv_sec--;
648 * Use of time interpolator for a gradual change of
649 * time
651 time_interpolator_update(NSEC_PER_SEC);
652 time_state = TIME_WAIT;
653 clock_was_set();
654 printk(KERN_NOTICE "Clock: deleting leap second "
655 "23:59:59 UTC\n");
657 break;
658 case TIME_OOP:
659 time_state = TIME_WAIT;
660 break;
661 case TIME_WAIT:
662 if (!(time_status & (STA_INS | STA_DEL)))
663 time_state = TIME_OK;
667 * Compute the phase adjustment for the next second. In PLL mode, the
668 * offset is reduced by a fixed factor times the time constant. In FLL
669 * mode the offset is used directly. In either mode, the maximum phase
670 * adjustment for each second is clamped so as to spread the adjustment
671 * over not more than the number of seconds between updates.
673 ltemp = time_offset;
674 if (!(time_status & STA_FLL))
675 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
676 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
677 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
678 time_offset -= ltemp;
679 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
682 * Compute the frequency estimate and additional phase adjustment due
683 * to frequency error for the next second. When the PPS signal is
684 * engaged, gnaw on the watchdog counter and update the frequency
685 * computed by the pll and the PPS signal.
687 pps_valid++;
688 if (pps_valid == PPS_VALID) { /* PPS signal lost */
689 pps_jitter = MAXTIME;
690 pps_stabil = MAXFREQ;
691 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
692 STA_PPSWANDER | STA_PPSERROR);
694 ltemp = time_freq + pps_freq;
695 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
697 #if HZ == 100
699 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
700 * get 128.125; => only 0.125% error (p. 14)
702 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
703 #endif
704 #if HZ == 250
706 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
707 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
709 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
710 #endif
711 #if HZ == 1000
713 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
714 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
716 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
717 #endif
720 /* in the NTP reference this is called "hardclock()" */
721 static void update_wall_time_one_tick(void)
723 long time_adjust_step, delta_nsec;
725 if ((time_adjust_step = time_adjust) != 0 ) {
727 * We are doing an adjtime thing. Prepare time_adjust_step to
728 * be within bounds. Note that a positive time_adjust means we
729 * want the clock to run faster.
731 * Limit the amount of the step to be in the range
732 * -tickadj .. +tickadj
734 time_adjust_step = min(time_adjust_step, (long)tickadj);
735 time_adjust_step = max(time_adjust_step, (long)-tickadj);
737 /* Reduce by this step the amount of time left */
738 time_adjust -= time_adjust_step;
740 delta_nsec = tick_nsec + time_adjust_step * 1000;
742 * Advance the phase, once it gets to one microsecond, then
743 * advance the tick more.
745 time_phase += time_adj;
746 if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) {
747 long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10));
748 time_phase -= ltemp << (SHIFT_SCALE - 10);
749 delta_nsec += ltemp;
751 xtime.tv_nsec += delta_nsec;
752 time_interpolator_update(delta_nsec);
754 /* Changes by adjtime() do not take effect till next tick. */
755 if (time_next_adjust != 0) {
756 time_adjust = time_next_adjust;
757 time_next_adjust = 0;
762 * Using a loop looks inefficient, but "ticks" is
763 * usually just one (we shouldn't be losing ticks,
764 * we're doing this this way mainly for interrupt
765 * latency reasons, not because we think we'll
766 * have lots of lost timer ticks
768 static void update_wall_time(unsigned long ticks)
770 do {
771 ticks--;
772 update_wall_time_one_tick();
773 if (xtime.tv_nsec >= 1000000000) {
774 xtime.tv_nsec -= 1000000000;
775 xtime.tv_sec++;
776 second_overflow();
778 } while (ticks);
782 * Called from the timer interrupt handler to charge one tick to the current
783 * process. user_tick is 1 if the tick is user time, 0 for system.
785 void update_process_times(int user_tick)
787 struct task_struct *p = current;
788 int cpu = smp_processor_id();
790 /* Note: this timer irq context must be accounted for as well. */
791 if (user_tick)
792 account_user_time(p, jiffies_to_cputime(1));
793 else
794 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
795 run_local_timers();
796 if (rcu_pending(cpu))
797 rcu_check_callbacks(cpu, user_tick);
798 scheduler_tick();
799 run_posix_cpu_timers(p);
803 * Nr of active tasks - counted in fixed-point numbers
805 static unsigned long count_active_tasks(void)
807 return (nr_running() + nr_uninterruptible()) * FIXED_1;
811 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
812 * imply that avenrun[] is the standard name for this kind of thing.
813 * Nothing else seems to be standardized: the fractional size etc
814 * all seem to differ on different machines.
816 * Requires xtime_lock to access.
818 unsigned long avenrun[3];
820 EXPORT_SYMBOL(avenrun);
823 * calc_load - given tick count, update the avenrun load estimates.
824 * This is called while holding a write_lock on xtime_lock.
826 static inline void calc_load(unsigned long ticks)
828 unsigned long active_tasks; /* fixed-point */
829 static int count = LOAD_FREQ;
831 count -= ticks;
832 if (count < 0) {
833 count += LOAD_FREQ;
834 active_tasks = count_active_tasks();
835 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
836 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
837 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
841 /* jiffies at the most recent update of wall time */
842 unsigned long wall_jiffies = INITIAL_JIFFIES;
845 * This read-write spinlock protects us from races in SMP while
846 * playing with xtime and avenrun.
848 #ifndef ARCH_HAVE_XTIME_LOCK
849 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
851 EXPORT_SYMBOL(xtime_lock);
852 #endif
855 * This function runs timers and the timer-tq in bottom half context.
857 static void run_timer_softirq(struct softirq_action *h)
859 tvec_base_t *base = &__get_cpu_var(tvec_bases);
861 if (time_after_eq(jiffies, base->timer_jiffies))
862 __run_timers(base);
866 * Called by the local, per-CPU timer interrupt on SMP.
868 void run_local_timers(void)
870 raise_softirq(TIMER_SOFTIRQ);
874 * Called by the timer interrupt. xtime_lock must already be taken
875 * by the timer IRQ!
877 static inline void update_times(void)
879 unsigned long ticks;
881 ticks = jiffies - wall_jiffies;
882 if (ticks) {
883 wall_jiffies += ticks;
884 update_wall_time(ticks);
886 calc_load(ticks);
890 * The 64-bit jiffies value is not atomic - you MUST NOT read it
891 * without sampling the sequence number in xtime_lock.
892 * jiffies is defined in the linker script...
895 void do_timer(struct pt_regs *regs)
897 jiffies_64++;
898 update_times();
899 softlockup_tick(regs);
902 #ifdef __ARCH_WANT_SYS_ALARM
905 * For backwards compatibility? This can be done in libc so Alpha
906 * and all newer ports shouldn't need it.
908 asmlinkage unsigned long sys_alarm(unsigned int seconds)
910 struct itimerval it_new, it_old;
911 unsigned int oldalarm;
913 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
914 it_new.it_value.tv_sec = seconds;
915 it_new.it_value.tv_usec = 0;
916 do_setitimer(ITIMER_REAL, &it_new, &it_old);
917 oldalarm = it_old.it_value.tv_sec;
918 /* ehhh.. We can't return 0 if we have an alarm pending.. */
919 /* And we'd better return too much than too little anyway */
920 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
921 oldalarm++;
922 return oldalarm;
925 #endif
927 #ifndef __alpha__
930 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
931 * should be moved into arch/i386 instead?
935 * sys_getpid - return the thread group id of the current process
937 * Note, despite the name, this returns the tgid not the pid. The tgid and
938 * the pid are identical unless CLONE_THREAD was specified on clone() in
939 * which case the tgid is the same in all threads of the same group.
941 * This is SMP safe as current->tgid does not change.
943 asmlinkage long sys_getpid(void)
945 return current->tgid;
949 * Accessing ->group_leader->real_parent is not SMP-safe, it could
950 * change from under us. However, rather than getting any lock
951 * we can use an optimistic algorithm: get the parent
952 * pid, and go back and check that the parent is still
953 * the same. If it has changed (which is extremely unlikely
954 * indeed), we just try again..
956 * NOTE! This depends on the fact that even if we _do_
957 * get an old value of "parent", we can happily dereference
958 * the pointer (it was and remains a dereferencable kernel pointer
959 * no matter what): we just can't necessarily trust the result
960 * until we know that the parent pointer is valid.
962 * NOTE2: ->group_leader never changes from under us.
964 asmlinkage long sys_getppid(void)
966 int pid;
967 struct task_struct *me = current;
968 struct task_struct *parent;
970 parent = me->group_leader->real_parent;
971 for (;;) {
972 pid = parent->tgid;
973 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
975 struct task_struct *old = parent;
978 * Make sure we read the pid before re-reading the
979 * parent pointer:
981 smp_rmb();
982 parent = me->group_leader->real_parent;
983 if (old != parent)
984 continue;
986 #endif
987 break;
989 return pid;
992 asmlinkage long sys_getuid(void)
994 /* Only we change this so SMP safe */
995 return current->uid;
998 asmlinkage long sys_geteuid(void)
1000 /* Only we change this so SMP safe */
1001 return current->euid;
1004 asmlinkage long sys_getgid(void)
1006 /* Only we change this so SMP safe */
1007 return current->gid;
1010 asmlinkage long sys_getegid(void)
1012 /* Only we change this so SMP safe */
1013 return current->egid;
1016 #endif
1018 static void process_timeout(unsigned long __data)
1020 wake_up_process((task_t *)__data);
1024 * schedule_timeout - sleep until timeout
1025 * @timeout: timeout value in jiffies
1027 * Make the current task sleep until @timeout jiffies have
1028 * elapsed. The routine will return immediately unless
1029 * the current task state has been set (see set_current_state()).
1031 * You can set the task state as follows -
1033 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1034 * pass before the routine returns. The routine will return 0
1036 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1037 * delivered to the current task. In this case the remaining time
1038 * in jiffies will be returned, or 0 if the timer expired in time
1040 * The current task state is guaranteed to be TASK_RUNNING when this
1041 * routine returns.
1043 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1044 * the CPU away without a bound on the timeout. In this case the return
1045 * value will be %MAX_SCHEDULE_TIMEOUT.
1047 * In all cases the return value is guaranteed to be non-negative.
1049 fastcall signed long __sched schedule_timeout(signed long timeout)
1051 struct timer_list timer;
1052 unsigned long expire;
1054 switch (timeout)
1056 case MAX_SCHEDULE_TIMEOUT:
1058 * These two special cases are useful to be comfortable
1059 * in the caller. Nothing more. We could take
1060 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1061 * but I' d like to return a valid offset (>=0) to allow
1062 * the caller to do everything it want with the retval.
1064 schedule();
1065 goto out;
1066 default:
1068 * Another bit of PARANOID. Note that the retval will be
1069 * 0 since no piece of kernel is supposed to do a check
1070 * for a negative retval of schedule_timeout() (since it
1071 * should never happens anyway). You just have the printk()
1072 * that will tell you if something is gone wrong and where.
1074 if (timeout < 0)
1076 printk(KERN_ERR "schedule_timeout: wrong timeout "
1077 "value %lx from %p\n", timeout,
1078 __builtin_return_address(0));
1079 current->state = TASK_RUNNING;
1080 goto out;
1084 expire = timeout + jiffies;
1086 setup_timer(&timer, process_timeout, (unsigned long)current);
1087 __mod_timer(&timer, expire);
1088 schedule();
1089 del_singleshot_timer_sync(&timer);
1091 timeout = expire - jiffies;
1093 out:
1094 return timeout < 0 ? 0 : timeout;
1096 EXPORT_SYMBOL(schedule_timeout);
1099 * We can use __set_current_state() here because schedule_timeout() calls
1100 * schedule() unconditionally.
1102 signed long __sched schedule_timeout_interruptible(signed long timeout)
1104 __set_current_state(TASK_INTERRUPTIBLE);
1105 return schedule_timeout(timeout);
1107 EXPORT_SYMBOL(schedule_timeout_interruptible);
1109 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1111 __set_current_state(TASK_UNINTERRUPTIBLE);
1112 return schedule_timeout(timeout);
1114 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1116 /* Thread ID - the internal kernel "pid" */
1117 asmlinkage long sys_gettid(void)
1119 return current->pid;
1122 static long __sched nanosleep_restart(struct restart_block *restart)
1124 unsigned long expire = restart->arg0, now = jiffies;
1125 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1126 long ret;
1128 /* Did it expire while we handled signals? */
1129 if (!time_after(expire, now))
1130 return 0;
1132 expire = schedule_timeout_interruptible(expire - now);
1134 ret = 0;
1135 if (expire) {
1136 struct timespec t;
1137 jiffies_to_timespec(expire, &t);
1139 ret = -ERESTART_RESTARTBLOCK;
1140 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1141 ret = -EFAULT;
1142 /* The 'restart' block is already filled in */
1144 return ret;
1147 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1149 struct timespec t;
1150 unsigned long expire;
1151 long ret;
1153 if (copy_from_user(&t, rqtp, sizeof(t)))
1154 return -EFAULT;
1156 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1157 return -EINVAL;
1159 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1160 expire = schedule_timeout_interruptible(expire);
1162 ret = 0;
1163 if (expire) {
1164 struct restart_block *restart;
1165 jiffies_to_timespec(expire, &t);
1166 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1167 return -EFAULT;
1169 restart = &current_thread_info()->restart_block;
1170 restart->fn = nanosleep_restart;
1171 restart->arg0 = jiffies + expire;
1172 restart->arg1 = (unsigned long) rmtp;
1173 ret = -ERESTART_RESTARTBLOCK;
1175 return ret;
1179 * sys_sysinfo - fill in sysinfo struct
1181 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1183 struct sysinfo val;
1184 unsigned long mem_total, sav_total;
1185 unsigned int mem_unit, bitcount;
1186 unsigned long seq;
1188 memset((char *)&val, 0, sizeof(struct sysinfo));
1190 do {
1191 struct timespec tp;
1192 seq = read_seqbegin(&xtime_lock);
1195 * This is annoying. The below is the same thing
1196 * posix_get_clock_monotonic() does, but it wants to
1197 * take the lock which we want to cover the loads stuff
1198 * too.
1201 getnstimeofday(&tp);
1202 tp.tv_sec += wall_to_monotonic.tv_sec;
1203 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1204 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1205 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1206 tp.tv_sec++;
1208 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1210 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1211 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1212 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1214 val.procs = nr_threads;
1215 } while (read_seqretry(&xtime_lock, seq));
1217 si_meminfo(&val);
1218 si_swapinfo(&val);
1221 * If the sum of all the available memory (i.e. ram + swap)
1222 * is less than can be stored in a 32 bit unsigned long then
1223 * we can be binary compatible with 2.2.x kernels. If not,
1224 * well, in that case 2.2.x was broken anyways...
1226 * -Erik Andersen <andersee@debian.org>
1229 mem_total = val.totalram + val.totalswap;
1230 if (mem_total < val.totalram || mem_total < val.totalswap)
1231 goto out;
1232 bitcount = 0;
1233 mem_unit = val.mem_unit;
1234 while (mem_unit > 1) {
1235 bitcount++;
1236 mem_unit >>= 1;
1237 sav_total = mem_total;
1238 mem_total <<= 1;
1239 if (mem_total < sav_total)
1240 goto out;
1244 * If mem_total did not overflow, multiply all memory values by
1245 * val.mem_unit and set it to 1. This leaves things compatible
1246 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1247 * kernels...
1250 val.mem_unit = 1;
1251 val.totalram <<= bitcount;
1252 val.freeram <<= bitcount;
1253 val.sharedram <<= bitcount;
1254 val.bufferram <<= bitcount;
1255 val.totalswap <<= bitcount;
1256 val.freeswap <<= bitcount;
1257 val.totalhigh <<= bitcount;
1258 val.freehigh <<= bitcount;
1260 out:
1261 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1262 return -EFAULT;
1264 return 0;
1267 static void __devinit init_timers_cpu(int cpu)
1269 int j;
1270 tvec_base_t *base;
1272 base = &per_cpu(tvec_bases, cpu);
1273 spin_lock_init(&base->t_base.lock);
1274 for (j = 0; j < TVN_SIZE; j++) {
1275 INIT_LIST_HEAD(base->tv5.vec + j);
1276 INIT_LIST_HEAD(base->tv4.vec + j);
1277 INIT_LIST_HEAD(base->tv3.vec + j);
1278 INIT_LIST_HEAD(base->tv2.vec + j);
1280 for (j = 0; j < TVR_SIZE; j++)
1281 INIT_LIST_HEAD(base->tv1.vec + j);
1283 base->timer_jiffies = jiffies;
1286 #ifdef CONFIG_HOTPLUG_CPU
1287 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1289 struct timer_list *timer;
1291 while (!list_empty(head)) {
1292 timer = list_entry(head->next, struct timer_list, entry);
1293 detach_timer(timer, 0);
1294 timer->base = &new_base->t_base;
1295 internal_add_timer(new_base, timer);
1299 static void __devinit migrate_timers(int cpu)
1301 tvec_base_t *old_base;
1302 tvec_base_t *new_base;
1303 int i;
1305 BUG_ON(cpu_online(cpu));
1306 old_base = &per_cpu(tvec_bases, cpu);
1307 new_base = &get_cpu_var(tvec_bases);
1309 local_irq_disable();
1310 spin_lock(&new_base->t_base.lock);
1311 spin_lock(&old_base->t_base.lock);
1313 if (old_base->t_base.running_timer)
1314 BUG();
1315 for (i = 0; i < TVR_SIZE; i++)
1316 migrate_timer_list(new_base, old_base->tv1.vec + i);
1317 for (i = 0; i < TVN_SIZE; i++) {
1318 migrate_timer_list(new_base, old_base->tv2.vec + i);
1319 migrate_timer_list(new_base, old_base->tv3.vec + i);
1320 migrate_timer_list(new_base, old_base->tv4.vec + i);
1321 migrate_timer_list(new_base, old_base->tv5.vec + i);
1324 spin_unlock(&old_base->t_base.lock);
1325 spin_unlock(&new_base->t_base.lock);
1326 local_irq_enable();
1327 put_cpu_var(tvec_bases);
1329 #endif /* CONFIG_HOTPLUG_CPU */
1331 static int __devinit timer_cpu_notify(struct notifier_block *self,
1332 unsigned long action, void *hcpu)
1334 long cpu = (long)hcpu;
1335 switch(action) {
1336 case CPU_UP_PREPARE:
1337 init_timers_cpu(cpu);
1338 break;
1339 #ifdef CONFIG_HOTPLUG_CPU
1340 case CPU_DEAD:
1341 migrate_timers(cpu);
1342 break;
1343 #endif
1344 default:
1345 break;
1347 return NOTIFY_OK;
1350 static struct notifier_block __devinitdata timers_nb = {
1351 .notifier_call = timer_cpu_notify,
1355 void __init init_timers(void)
1357 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1358 (void *)(long)smp_processor_id());
1359 register_cpu_notifier(&timers_nb);
1360 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1363 #ifdef CONFIG_TIME_INTERPOLATION
1365 struct time_interpolator *time_interpolator;
1366 static struct time_interpolator *time_interpolator_list;
1367 static DEFINE_SPINLOCK(time_interpolator_lock);
1369 static inline u64 time_interpolator_get_cycles(unsigned int src)
1371 unsigned long (*x)(void);
1373 switch (src)
1375 case TIME_SOURCE_FUNCTION:
1376 x = time_interpolator->addr;
1377 return x();
1379 case TIME_SOURCE_MMIO64 :
1380 return readq((void __iomem *) time_interpolator->addr);
1382 case TIME_SOURCE_MMIO32 :
1383 return readl((void __iomem *) time_interpolator->addr);
1385 default: return get_cycles();
1389 static inline u64 time_interpolator_get_counter(int writelock)
1391 unsigned int src = time_interpolator->source;
1393 if (time_interpolator->jitter)
1395 u64 lcycle;
1396 u64 now;
1398 do {
1399 lcycle = time_interpolator->last_cycle;
1400 now = time_interpolator_get_cycles(src);
1401 if (lcycle && time_after(lcycle, now))
1402 return lcycle;
1404 /* When holding the xtime write lock, there's no need
1405 * to add the overhead of the cmpxchg. Readers are
1406 * force to retry until the write lock is released.
1408 if (writelock) {
1409 time_interpolator->last_cycle = now;
1410 return now;
1412 /* Keep track of the last timer value returned. The use of cmpxchg here
1413 * will cause contention in an SMP environment.
1415 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1416 return now;
1418 else
1419 return time_interpolator_get_cycles(src);
1422 void time_interpolator_reset(void)
1424 time_interpolator->offset = 0;
1425 time_interpolator->last_counter = time_interpolator_get_counter(1);
1428 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1430 unsigned long time_interpolator_get_offset(void)
1432 /* If we do not have a time interpolator set up then just return zero */
1433 if (!time_interpolator)
1434 return 0;
1436 return time_interpolator->offset +
1437 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1440 #define INTERPOLATOR_ADJUST 65536
1441 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1443 static void time_interpolator_update(long delta_nsec)
1445 u64 counter;
1446 unsigned long offset;
1448 /* If there is no time interpolator set up then do nothing */
1449 if (!time_interpolator)
1450 return;
1453 * The interpolator compensates for late ticks by accumulating the late
1454 * time in time_interpolator->offset. A tick earlier than expected will
1455 * lead to a reset of the offset and a corresponding jump of the clock
1456 * forward. Again this only works if the interpolator clock is running
1457 * slightly slower than the regular clock and the tuning logic insures
1458 * that.
1461 counter = time_interpolator_get_counter(1);
1462 offset = time_interpolator->offset +
1463 GET_TI_NSECS(counter, time_interpolator);
1465 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1466 time_interpolator->offset = offset - delta_nsec;
1467 else {
1468 time_interpolator->skips++;
1469 time_interpolator->ns_skipped += delta_nsec - offset;
1470 time_interpolator->offset = 0;
1472 time_interpolator->last_counter = counter;
1474 /* Tuning logic for time interpolator invoked every minute or so.
1475 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1476 * Increase interpolator clock speed if we skip too much time.
1478 if (jiffies % INTERPOLATOR_ADJUST == 0)
1480 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1481 time_interpolator->nsec_per_cyc--;
1482 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1483 time_interpolator->nsec_per_cyc++;
1484 time_interpolator->skips = 0;
1485 time_interpolator->ns_skipped = 0;
1489 static inline int
1490 is_better_time_interpolator(struct time_interpolator *new)
1492 if (!time_interpolator)
1493 return 1;
1494 return new->frequency > 2*time_interpolator->frequency ||
1495 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1498 void
1499 register_time_interpolator(struct time_interpolator *ti)
1501 unsigned long flags;
1503 /* Sanity check */
1504 if (ti->frequency == 0 || ti->mask == 0)
1505 BUG();
1507 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1508 spin_lock(&time_interpolator_lock);
1509 write_seqlock_irqsave(&xtime_lock, flags);
1510 if (is_better_time_interpolator(ti)) {
1511 time_interpolator = ti;
1512 time_interpolator_reset();
1514 write_sequnlock_irqrestore(&xtime_lock, flags);
1516 ti->next = time_interpolator_list;
1517 time_interpolator_list = ti;
1518 spin_unlock(&time_interpolator_lock);
1521 void
1522 unregister_time_interpolator(struct time_interpolator *ti)
1524 struct time_interpolator *curr, **prev;
1525 unsigned long flags;
1527 spin_lock(&time_interpolator_lock);
1528 prev = &time_interpolator_list;
1529 for (curr = *prev; curr; curr = curr->next) {
1530 if (curr == ti) {
1531 *prev = curr->next;
1532 break;
1534 prev = &curr->next;
1537 write_seqlock_irqsave(&xtime_lock, flags);
1538 if (ti == time_interpolator) {
1539 /* we lost the best time-interpolator: */
1540 time_interpolator = NULL;
1541 /* find the next-best interpolator */
1542 for (curr = time_interpolator_list; curr; curr = curr->next)
1543 if (is_better_time_interpolator(curr))
1544 time_interpolator = curr;
1545 time_interpolator_reset();
1547 write_sequnlock_irqrestore(&xtime_lock, flags);
1548 spin_unlock(&time_interpolator_lock);
1550 #endif /* CONFIG_TIME_INTERPOLATION */
1553 * msleep - sleep safely even with waitqueue interruptions
1554 * @msecs: Time in milliseconds to sleep for
1556 void msleep(unsigned int msecs)
1558 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1560 while (timeout)
1561 timeout = schedule_timeout_uninterruptible(timeout);
1564 EXPORT_SYMBOL(msleep);
1567 * msleep_interruptible - sleep waiting for signals
1568 * @msecs: Time in milliseconds to sleep for
1570 unsigned long msleep_interruptible(unsigned int msecs)
1572 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1574 while (timeout && !signal_pending(current))
1575 timeout = schedule_timeout_interruptible(timeout);
1576 return jiffies_to_msecs(timeout);
1579 EXPORT_SYMBOL(msleep_interruptible);