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>
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/div64.h>
41 #include <asm/timex.h>
44 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec);
47 #define time_interpolator_update(x)
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)
67 struct timer_list *running_timer;
70 typedef struct tvec_s {
71 struct list_head vec[TVN_SIZE];
74 typedef struct tvec_root_s {
75 struct list_head vec[TVR_SIZE];
78 struct tvec_t_base_s {
79 struct timer_base_s t_base;
80 unsigned long timer_jiffies;
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)
95 base->t_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);
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);
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,
167 struct list_head *entry = &timer->entry;
169 __list_del(entry->prev, entry->next);
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
187 static timer_base_t *lock_timer_base(struct timer_list *timer,
188 unsigned long *flags)
194 if (likely(base != NULL)) {
195 spin_lock_irqsave(&base->lock, *flags);
196 if (likely(base == timer->base))
198 /* The timer has migrated to another CPU */
199 spin_unlock_irqrestore(&base->lock, *flags);
205 int __mod_timer(struct timer_list *timer, unsigned long expires)
208 tvec_base_t *new_base;
212 BUG_ON(!timer->function);
214 base = lock_timer_base(timer, &flags);
216 if (timer_pending(timer)) {
217 detach_timer(timer, 0);
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);
235 /* See the comment in lock_timer_base() */
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);
250 EXPORT_SYMBOL(__mod_timer);
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);
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);
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))
303 return __mod_timer(timer, expires);
306 EXPORT_SYMBOL(mod_timer);
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
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)
325 if (timer_pending(timer)) {
326 base = lock_timer_base(timer, &flags);
327 if (timer_pending(timer)) {
328 detach_timer(timer, 1);
331 spin_unlock_irqrestore(&base->lock, flags);
337 EXPORT_SYMBOL(del_timer);
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)
352 base = lock_timer_base(timer, &flags);
354 if (base->running_timer == timer)
358 if (timer_pending(timer)) {
359 detach_timer(timer, 1);
363 spin_unlock_irqrestore(&base->lock, flags);
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
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)
388 int ret = try_to_del_timer_sync(timer);
394 EXPORT_SYMBOL(del_timer_sync);
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;
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);
414 internal_add_timer(base, tmp);
416 INIT_LIST_HEAD(head);
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
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;
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);
454 timer = list_entry(head->next,struct timer_list,entry);
455 fn = timer->function;
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();
464 if (preempt_count != preempt_count()) {
465 printk(KERN_WARNING "huh, entered %p "
466 "with preempt_count %08x, exited"
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)
489 struct list_head *list;
490 struct timer_list *nte;
491 unsigned long expires;
495 base = &__get_cpu_var(tvec_bases);
496 spin_lock(&base->t_base.lock);
497 expires = base->timer_jiffies + (LONG_MAX >> 1);
500 /* Look for timer events in tv1. */
501 j = base->timer_jiffies & TVR_MASK;
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));
509 j = (j + 1) & TVR_MASK;
510 } while (j != (base->timer_jiffies & TVR_MASK));
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++) {
520 if (list_empty(varray[i]->vec + j)) {
521 j = (j + 1) & TVN_MASK;
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));
530 } while (j != (INDEX(i)));
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);
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) */
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) */
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)
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) {
623 if (time_status & STA_INS)
624 time_state = TIME_INS;
625 else if (time_status & STA_DEL)
626 time_state = TIME_DEL;
629 if (xtime.tv_sec % 86400 == 0) {
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;
639 printk(KERN_NOTICE "Clock: inserting leap second "
644 if ((xtime.tv_sec + 1) % 86400 == 0) {
646 wall_to_monotonic.tv_sec--;
648 * Use of time interpolator for a gradual change of
651 time_interpolator_update(NSEC_PER_SEC);
652 time_state = TIME_WAIT;
654 printk(KERN_NOTICE "Clock: deleting leap second "
659 time_state = 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.
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.
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));
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);
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);
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);
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);
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)
772 update_wall_time_one_tick();
773 if (xtime.tv_nsec >= 1000000000) {
774 xtime.tv_nsec -= 1000000000;
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. */
792 account_user_time(p, jiffies_to_cputime(1));
794 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
796 if (rcu_pending(cpu))
797 rcu_check_callbacks(cpu, user_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;
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);
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 hrtimer_run_queues();
862 if (time_after_eq(jiffies, base->timer_jiffies))
867 * Called by the local, per-CPU timer interrupt on SMP.
869 void run_local_timers(void)
871 raise_softirq(TIMER_SOFTIRQ);
875 * Called by the timer interrupt. xtime_lock must already be taken
878 static inline void update_times(void)
882 ticks = jiffies - wall_jiffies;
884 wall_jiffies += ticks;
885 update_wall_time(ticks);
891 * The 64-bit jiffies value is not atomic - you MUST NOT read it
892 * without sampling the sequence number in xtime_lock.
893 * jiffies is defined in the linker script...
896 void do_timer(struct pt_regs *regs)
900 softlockup_tick(regs);
903 #ifdef __ARCH_WANT_SYS_ALARM
906 * For backwards compatibility? This can be done in libc so Alpha
907 * and all newer ports shouldn't need it.
909 asmlinkage unsigned long sys_alarm(unsigned int seconds)
911 struct itimerval it_new, it_old;
912 unsigned int oldalarm;
914 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
915 it_new.it_value.tv_sec = seconds;
916 it_new.it_value.tv_usec = 0;
917 do_setitimer(ITIMER_REAL, &it_new, &it_old);
918 oldalarm = it_old.it_value.tv_sec;
919 /* ehhh.. We can't return 0 if we have an alarm pending.. */
920 /* And we'd better return too much than too little anyway */
921 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
931 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
932 * should be moved into arch/i386 instead?
936 * sys_getpid - return the thread group id of the current process
938 * Note, despite the name, this returns the tgid not the pid. The tgid and
939 * the pid are identical unless CLONE_THREAD was specified on clone() in
940 * which case the tgid is the same in all threads of the same group.
942 * This is SMP safe as current->tgid does not change.
944 asmlinkage long sys_getpid(void)
946 return current->tgid;
950 * Accessing ->group_leader->real_parent is not SMP-safe, it could
951 * change from under us. However, rather than getting any lock
952 * we can use an optimistic algorithm: get the parent
953 * pid, and go back and check that the parent is still
954 * the same. If it has changed (which is extremely unlikely
955 * indeed), we just try again..
957 * NOTE! This depends on the fact that even if we _do_
958 * get an old value of "parent", we can happily dereference
959 * the pointer (it was and remains a dereferencable kernel pointer
960 * no matter what): we just can't necessarily trust the result
961 * until we know that the parent pointer is valid.
963 * NOTE2: ->group_leader never changes from under us.
965 asmlinkage long sys_getppid(void)
968 struct task_struct *me = current;
969 struct task_struct *parent;
971 parent = me->group_leader->real_parent;
974 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
976 struct task_struct *old = parent;
979 * Make sure we read the pid before re-reading the
983 parent = me->group_leader->real_parent;
993 asmlinkage long sys_getuid(void)
995 /* Only we change this so SMP safe */
999 asmlinkage long sys_geteuid(void)
1001 /* Only we change this so SMP safe */
1002 return current->euid;
1005 asmlinkage long sys_getgid(void)
1007 /* Only we change this so SMP safe */
1008 return current->gid;
1011 asmlinkage long sys_getegid(void)
1013 /* Only we change this so SMP safe */
1014 return current->egid;
1019 static void process_timeout(unsigned long __data)
1021 wake_up_process((task_t *)__data);
1025 * schedule_timeout - sleep until timeout
1026 * @timeout: timeout value in jiffies
1028 * Make the current task sleep until @timeout jiffies have
1029 * elapsed. The routine will return immediately unless
1030 * the current task state has been set (see set_current_state()).
1032 * You can set the task state as follows -
1034 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1035 * pass before the routine returns. The routine will return 0
1037 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1038 * delivered to the current task. In this case the remaining time
1039 * in jiffies will be returned, or 0 if the timer expired in time
1041 * The current task state is guaranteed to be TASK_RUNNING when this
1044 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1045 * the CPU away without a bound on the timeout. In this case the return
1046 * value will be %MAX_SCHEDULE_TIMEOUT.
1048 * In all cases the return value is guaranteed to be non-negative.
1050 fastcall signed long __sched schedule_timeout(signed long timeout)
1052 struct timer_list timer;
1053 unsigned long expire;
1057 case MAX_SCHEDULE_TIMEOUT:
1059 * These two special cases are useful to be comfortable
1060 * in the caller. Nothing more. We could take
1061 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1062 * but I' d like to return a valid offset (>=0) to allow
1063 * the caller to do everything it want with the retval.
1069 * Another bit of PARANOID. Note that the retval will be
1070 * 0 since no piece of kernel is supposed to do a check
1071 * for a negative retval of schedule_timeout() (since it
1072 * should never happens anyway). You just have the printk()
1073 * that will tell you if something is gone wrong and where.
1077 printk(KERN_ERR "schedule_timeout: wrong timeout "
1078 "value %lx from %p\n", timeout,
1079 __builtin_return_address(0));
1080 current->state = TASK_RUNNING;
1085 expire = timeout + jiffies;
1087 setup_timer(&timer, process_timeout, (unsigned long)current);
1088 __mod_timer(&timer, expire);
1090 del_singleshot_timer_sync(&timer);
1092 timeout = expire - jiffies;
1095 return timeout < 0 ? 0 : timeout;
1097 EXPORT_SYMBOL(schedule_timeout);
1100 * We can use __set_current_state() here because schedule_timeout() calls
1101 * schedule() unconditionally.
1103 signed long __sched schedule_timeout_interruptible(signed long timeout)
1105 __set_current_state(TASK_INTERRUPTIBLE);
1106 return schedule_timeout(timeout);
1108 EXPORT_SYMBOL(schedule_timeout_interruptible);
1110 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1112 __set_current_state(TASK_UNINTERRUPTIBLE);
1113 return schedule_timeout(timeout);
1115 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1117 /* Thread ID - the internal kernel "pid" */
1118 asmlinkage long sys_gettid(void)
1120 return current->pid;
1124 * sys_sysinfo - fill in sysinfo struct
1126 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1129 unsigned long mem_total, sav_total;
1130 unsigned int mem_unit, bitcount;
1133 memset((char *)&val, 0, sizeof(struct sysinfo));
1137 seq = read_seqbegin(&xtime_lock);
1140 * This is annoying. The below is the same thing
1141 * posix_get_clock_monotonic() does, but it wants to
1142 * take the lock which we want to cover the loads stuff
1146 getnstimeofday(&tp);
1147 tp.tv_sec += wall_to_monotonic.tv_sec;
1148 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1149 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1150 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1153 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1155 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1156 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1157 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1159 val.procs = nr_threads;
1160 } while (read_seqretry(&xtime_lock, seq));
1166 * If the sum of all the available memory (i.e. ram + swap)
1167 * is less than can be stored in a 32 bit unsigned long then
1168 * we can be binary compatible with 2.2.x kernels. If not,
1169 * well, in that case 2.2.x was broken anyways...
1171 * -Erik Andersen <andersee@debian.org>
1174 mem_total = val.totalram + val.totalswap;
1175 if (mem_total < val.totalram || mem_total < val.totalswap)
1178 mem_unit = val.mem_unit;
1179 while (mem_unit > 1) {
1182 sav_total = mem_total;
1184 if (mem_total < sav_total)
1189 * If mem_total did not overflow, multiply all memory values by
1190 * val.mem_unit and set it to 1. This leaves things compatible
1191 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1196 val.totalram <<= bitcount;
1197 val.freeram <<= bitcount;
1198 val.sharedram <<= bitcount;
1199 val.bufferram <<= bitcount;
1200 val.totalswap <<= bitcount;
1201 val.freeswap <<= bitcount;
1202 val.totalhigh <<= bitcount;
1203 val.freehigh <<= bitcount;
1206 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1212 static void __devinit init_timers_cpu(int cpu)
1217 base = &per_cpu(tvec_bases, cpu);
1218 spin_lock_init(&base->t_base.lock);
1219 for (j = 0; j < TVN_SIZE; j++) {
1220 INIT_LIST_HEAD(base->tv5.vec + j);
1221 INIT_LIST_HEAD(base->tv4.vec + j);
1222 INIT_LIST_HEAD(base->tv3.vec + j);
1223 INIT_LIST_HEAD(base->tv2.vec + j);
1225 for (j = 0; j < TVR_SIZE; j++)
1226 INIT_LIST_HEAD(base->tv1.vec + j);
1228 base->timer_jiffies = jiffies;
1231 #ifdef CONFIG_HOTPLUG_CPU
1232 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1234 struct timer_list *timer;
1236 while (!list_empty(head)) {
1237 timer = list_entry(head->next, struct timer_list, entry);
1238 detach_timer(timer, 0);
1239 timer->base = &new_base->t_base;
1240 internal_add_timer(new_base, timer);
1244 static void __devinit migrate_timers(int cpu)
1246 tvec_base_t *old_base;
1247 tvec_base_t *new_base;
1250 BUG_ON(cpu_online(cpu));
1251 old_base = &per_cpu(tvec_bases, cpu);
1252 new_base = &get_cpu_var(tvec_bases);
1254 local_irq_disable();
1255 spin_lock(&new_base->t_base.lock);
1256 spin_lock(&old_base->t_base.lock);
1258 if (old_base->t_base.running_timer)
1260 for (i = 0; i < TVR_SIZE; i++)
1261 migrate_timer_list(new_base, old_base->tv1.vec + i);
1262 for (i = 0; i < TVN_SIZE; i++) {
1263 migrate_timer_list(new_base, old_base->tv2.vec + i);
1264 migrate_timer_list(new_base, old_base->tv3.vec + i);
1265 migrate_timer_list(new_base, old_base->tv4.vec + i);
1266 migrate_timer_list(new_base, old_base->tv5.vec + i);
1269 spin_unlock(&old_base->t_base.lock);
1270 spin_unlock(&new_base->t_base.lock);
1272 put_cpu_var(tvec_bases);
1274 #endif /* CONFIG_HOTPLUG_CPU */
1276 static int __devinit timer_cpu_notify(struct notifier_block *self,
1277 unsigned long action, void *hcpu)
1279 long cpu = (long)hcpu;
1281 case CPU_UP_PREPARE:
1282 init_timers_cpu(cpu);
1284 #ifdef CONFIG_HOTPLUG_CPU
1286 migrate_timers(cpu);
1295 static struct notifier_block __devinitdata timers_nb = {
1296 .notifier_call = timer_cpu_notify,
1300 void __init init_timers(void)
1302 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1303 (void *)(long)smp_processor_id());
1304 register_cpu_notifier(&timers_nb);
1305 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1308 #ifdef CONFIG_TIME_INTERPOLATION
1310 struct time_interpolator *time_interpolator;
1311 static struct time_interpolator *time_interpolator_list;
1312 static DEFINE_SPINLOCK(time_interpolator_lock);
1314 static inline u64 time_interpolator_get_cycles(unsigned int src)
1316 unsigned long (*x)(void);
1320 case TIME_SOURCE_FUNCTION:
1321 x = time_interpolator->addr;
1324 case TIME_SOURCE_MMIO64 :
1325 return readq((void __iomem *) time_interpolator->addr);
1327 case TIME_SOURCE_MMIO32 :
1328 return readl((void __iomem *) time_interpolator->addr);
1330 default: return get_cycles();
1334 static inline u64 time_interpolator_get_counter(int writelock)
1336 unsigned int src = time_interpolator->source;
1338 if (time_interpolator->jitter)
1344 lcycle = time_interpolator->last_cycle;
1345 now = time_interpolator_get_cycles(src);
1346 if (lcycle && time_after(lcycle, now))
1349 /* When holding the xtime write lock, there's no need
1350 * to add the overhead of the cmpxchg. Readers are
1351 * force to retry until the write lock is released.
1354 time_interpolator->last_cycle = now;
1357 /* Keep track of the last timer value returned. The use of cmpxchg here
1358 * will cause contention in an SMP environment.
1360 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1364 return time_interpolator_get_cycles(src);
1367 void time_interpolator_reset(void)
1369 time_interpolator->offset = 0;
1370 time_interpolator->last_counter = time_interpolator_get_counter(1);
1373 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1375 unsigned long time_interpolator_get_offset(void)
1377 /* If we do not have a time interpolator set up then just return zero */
1378 if (!time_interpolator)
1381 return time_interpolator->offset +
1382 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1385 #define INTERPOLATOR_ADJUST 65536
1386 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1388 static void time_interpolator_update(long delta_nsec)
1391 unsigned long offset;
1393 /* If there is no time interpolator set up then do nothing */
1394 if (!time_interpolator)
1398 * The interpolator compensates for late ticks by accumulating the late
1399 * time in time_interpolator->offset. A tick earlier than expected will
1400 * lead to a reset of the offset and a corresponding jump of the clock
1401 * forward. Again this only works if the interpolator clock is running
1402 * slightly slower than the regular clock and the tuning logic insures
1406 counter = time_interpolator_get_counter(1);
1407 offset = time_interpolator->offset +
1408 GET_TI_NSECS(counter, time_interpolator);
1410 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1411 time_interpolator->offset = offset - delta_nsec;
1413 time_interpolator->skips++;
1414 time_interpolator->ns_skipped += delta_nsec - offset;
1415 time_interpolator->offset = 0;
1417 time_interpolator->last_counter = counter;
1419 /* Tuning logic for time interpolator invoked every minute or so.
1420 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1421 * Increase interpolator clock speed if we skip too much time.
1423 if (jiffies % INTERPOLATOR_ADJUST == 0)
1425 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1426 time_interpolator->nsec_per_cyc--;
1427 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1428 time_interpolator->nsec_per_cyc++;
1429 time_interpolator->skips = 0;
1430 time_interpolator->ns_skipped = 0;
1435 is_better_time_interpolator(struct time_interpolator *new)
1437 if (!time_interpolator)
1439 return new->frequency > 2*time_interpolator->frequency ||
1440 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1444 register_time_interpolator(struct time_interpolator *ti)
1446 unsigned long flags;
1449 if (ti->frequency == 0 || ti->mask == 0)
1452 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1453 spin_lock(&time_interpolator_lock);
1454 write_seqlock_irqsave(&xtime_lock, flags);
1455 if (is_better_time_interpolator(ti)) {
1456 time_interpolator = ti;
1457 time_interpolator_reset();
1459 write_sequnlock_irqrestore(&xtime_lock, flags);
1461 ti->next = time_interpolator_list;
1462 time_interpolator_list = ti;
1463 spin_unlock(&time_interpolator_lock);
1467 unregister_time_interpolator(struct time_interpolator *ti)
1469 struct time_interpolator *curr, **prev;
1470 unsigned long flags;
1472 spin_lock(&time_interpolator_lock);
1473 prev = &time_interpolator_list;
1474 for (curr = *prev; curr; curr = curr->next) {
1482 write_seqlock_irqsave(&xtime_lock, flags);
1483 if (ti == time_interpolator) {
1484 /* we lost the best time-interpolator: */
1485 time_interpolator = NULL;
1486 /* find the next-best interpolator */
1487 for (curr = time_interpolator_list; curr; curr = curr->next)
1488 if (is_better_time_interpolator(curr))
1489 time_interpolator = curr;
1490 time_interpolator_reset();
1492 write_sequnlock_irqrestore(&xtime_lock, flags);
1493 spin_unlock(&time_interpolator_lock);
1495 #endif /* CONFIG_TIME_INTERPOLATION */
1498 * msleep - sleep safely even with waitqueue interruptions
1499 * @msecs: Time in milliseconds to sleep for
1501 void msleep(unsigned int msecs)
1503 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1506 timeout = schedule_timeout_uninterruptible(timeout);
1509 EXPORT_SYMBOL(msleep);
1512 * msleep_interruptible - sleep waiting for signals
1513 * @msecs: Time in milliseconds to sleep for
1515 unsigned long msleep_interruptible(unsigned int msecs)
1517 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1519 while (timeout && !signal_pending(current))
1520 timeout = schedule_timeout_interruptible(timeout);
1521 return jiffies_to_msecs(timeout);
1524 EXPORT_SYMBOL(msleep_interruptible);