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_base.h>
38 #include <linux/vs_cvirt.h>
39 #include <linux/vserver/sched.h>
41 #include <asm/uaccess.h>
42 #include <asm/unistd.h>
43 #include <asm/div64.h>
44 #include <asm/timex.h>
47 #ifdef CONFIG_TIME_INTERPOLATION
48 static void time_interpolator_update(long delta_nsec);
50 #define time_interpolator_update(x)
53 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
55 EXPORT_SYMBOL(jiffies_64);
58 * per-CPU timer vector definitions:
60 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
61 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
62 #define TVN_SIZE (1 << TVN_BITS)
63 #define TVR_SIZE (1 << TVR_BITS)
64 #define TVN_MASK (TVN_SIZE - 1)
65 #define TVR_MASK (TVR_SIZE - 1)
67 typedef struct tvec_s {
68 struct list_head vec[TVN_SIZE];
71 typedef struct tvec_root_s {
72 struct list_head vec[TVR_SIZE];
75 struct tvec_t_base_s {
77 struct timer_list *running_timer;
78 unsigned long timer_jiffies;
84 } ____cacheline_aligned_in_smp;
86 typedef struct tvec_t_base_s tvec_base_t;
88 tvec_base_t boot_tvec_bases;
89 EXPORT_SYMBOL(boot_tvec_bases);
90 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = { &boot_tvec_bases };
92 static inline void set_running_timer(tvec_base_t *base,
93 struct timer_list *timer)
96 base->running_timer = timer;
100 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
102 unsigned long expires = timer->expires;
103 unsigned long idx = expires - base->timer_jiffies;
104 struct list_head *vec;
106 if (idx < TVR_SIZE) {
107 int i = expires & TVR_MASK;
108 vec = base->tv1.vec + i;
109 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
110 int i = (expires >> TVR_BITS) & TVN_MASK;
111 vec = base->tv2.vec + i;
112 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
113 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
114 vec = base->tv3.vec + i;
115 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
116 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
117 vec = base->tv4.vec + i;
118 } else if ((signed long) idx < 0) {
120 * Can happen if you add a timer with expires == jiffies,
121 * or you set a timer to go off in the past
123 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
126 /* If the timeout is larger than 0xffffffff on 64-bit
127 * architectures then we use the maximum timeout:
129 if (idx > 0xffffffffUL) {
131 expires = idx + base->timer_jiffies;
133 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
134 vec = base->tv5.vec + i;
139 list_add_tail(&timer->entry, vec);
143 * init_timer - initialize a timer.
144 * @timer: the timer to be initialized
146 * init_timer() must be done to a timer prior calling *any* of the
147 * other timer functions.
149 void fastcall init_timer(struct timer_list *timer)
151 timer->entry.next = NULL;
152 timer->base = per_cpu(tvec_bases, raw_smp_processor_id());
154 EXPORT_SYMBOL(init_timer);
156 static inline void detach_timer(struct timer_list *timer,
159 struct list_head *entry = &timer->entry;
161 __list_del(entry->prev, entry->next);
164 entry->prev = LIST_POISON2;
168 * We are using hashed locking: holding per_cpu(tvec_bases).lock
169 * means that all timers which are tied to this base via timer->base are
170 * locked, and the base itself is locked too.
172 * So __run_timers/migrate_timers can safely modify all timers which could
173 * be found on ->tvX lists.
175 * When the timer's base is locked, and the timer removed from list, it is
176 * possible to set timer->base = NULL and drop the lock: the timer remains
179 static tvec_base_t *lock_timer_base(struct timer_list *timer,
180 unsigned long *flags)
186 if (likely(base != NULL)) {
187 spin_lock_irqsave(&base->lock, *flags);
188 if (likely(base == timer->base))
190 /* The timer has migrated to another CPU */
191 spin_unlock_irqrestore(&base->lock, *flags);
197 int __mod_timer(struct timer_list *timer, unsigned long expires)
199 tvec_base_t *base, *new_base;
203 BUG_ON(!timer->function);
205 base = lock_timer_base(timer, &flags);
207 if (timer_pending(timer)) {
208 detach_timer(timer, 0);
212 new_base = __get_cpu_var(tvec_bases);
214 if (base != new_base) {
216 * We are trying to schedule the timer on the local CPU.
217 * However we can't change timer's base while it is running,
218 * otherwise del_timer_sync() can't detect that the timer's
219 * handler yet has not finished. This also guarantees that
220 * the timer is serialized wrt itself.
222 if (likely(base->running_timer != timer)) {
223 /* See the comment in lock_timer_base() */
225 spin_unlock(&base->lock);
227 spin_lock(&base->lock);
232 timer->expires = expires;
233 internal_add_timer(base, timer);
234 spin_unlock_irqrestore(&base->lock, flags);
239 EXPORT_SYMBOL(__mod_timer);
242 * add_timer_on - start a timer on a particular CPU
243 * @timer: the timer to be added
244 * @cpu: the CPU to start it on
246 * This is not very scalable on SMP. Double adds are not possible.
248 void add_timer_on(struct timer_list *timer, int cpu)
250 tvec_base_t *base = per_cpu(tvec_bases, cpu);
253 BUG_ON(timer_pending(timer) || !timer->function);
254 spin_lock_irqsave(&base->lock, flags);
256 internal_add_timer(base, timer);
257 spin_unlock_irqrestore(&base->lock, flags);
262 * mod_timer - modify a timer's timeout
263 * @timer: the timer to be modified
265 * mod_timer is a more efficient way to update the expire field of an
266 * active timer (if the timer is inactive it will be activated)
268 * mod_timer(timer, expires) is equivalent to:
270 * del_timer(timer); timer->expires = expires; add_timer(timer);
272 * Note that if there are multiple unserialized concurrent users of the
273 * same timer, then mod_timer() is the only safe way to modify the timeout,
274 * since add_timer() cannot modify an already running timer.
276 * The function returns whether it has modified a pending timer or not.
277 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
278 * active timer returns 1.)
280 int mod_timer(struct timer_list *timer, unsigned long expires)
282 BUG_ON(!timer->function);
285 * This is a common optimization triggered by the
286 * networking code - if the timer is re-modified
287 * to be the same thing then just return:
289 if (timer->expires == expires && timer_pending(timer))
292 return __mod_timer(timer, expires);
295 EXPORT_SYMBOL(mod_timer);
298 * del_timer - deactive a timer.
299 * @timer: the timer to be deactivated
301 * del_timer() deactivates a timer - this works on both active and inactive
304 * The function returns whether it has deactivated a pending timer or not.
305 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
306 * active timer returns 1.)
308 int del_timer(struct timer_list *timer)
314 if (timer_pending(timer)) {
315 base = lock_timer_base(timer, &flags);
316 if (timer_pending(timer)) {
317 detach_timer(timer, 1);
320 spin_unlock_irqrestore(&base->lock, flags);
326 EXPORT_SYMBOL(del_timer);
330 * This function tries to deactivate a timer. Upon successful (ret >= 0)
331 * exit the timer is not queued and the handler is not running on any CPU.
333 * It must not be called from interrupt contexts.
335 int try_to_del_timer_sync(struct timer_list *timer)
341 base = lock_timer_base(timer, &flags);
343 if (base->running_timer == timer)
347 if (timer_pending(timer)) {
348 detach_timer(timer, 1);
352 spin_unlock_irqrestore(&base->lock, flags);
358 * del_timer_sync - deactivate a timer and wait for the handler to finish.
359 * @timer: the timer to be deactivated
361 * This function only differs from del_timer() on SMP: besides deactivating
362 * the timer it also makes sure the handler has finished executing on other
365 * Synchronization rules: callers must prevent restarting of the timer,
366 * otherwise this function is meaningless. It must not be called from
367 * interrupt contexts. The caller must not hold locks which would prevent
368 * completion of the timer's handler. The timer's handler must not call
369 * add_timer_on(). Upon exit the timer is not queued and the handler is
370 * not running on any CPU.
372 * The function returns whether it has deactivated a pending timer or not.
374 int del_timer_sync(struct timer_list *timer)
377 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 list_head *head, *curr;
391 head = tv->vec + index;
394 * We are removing _all_ timers from the list, so we don't have to
395 * detach them individually, just clear the list afterwards.
397 while (curr != head) {
398 struct timer_list *tmp;
400 tmp = list_entry(curr, struct timer_list, entry);
401 BUG_ON(tmp->base != base);
403 internal_add_timer(base, tmp);
405 INIT_LIST_HEAD(head);
411 * __run_timers - run all expired timers (if any) on this CPU.
412 * @base: the timer vector to be processed.
414 * This function cascades all vectors and executes all expired timer
417 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
419 static inline void __run_timers(tvec_base_t *base)
421 struct timer_list *timer;
423 spin_lock_irq(&base->lock);
424 while (time_after_eq(jiffies, base->timer_jiffies)) {
425 struct list_head work_list = LIST_HEAD_INIT(work_list);
426 struct list_head *head = &work_list;
427 int index = base->timer_jiffies & TVR_MASK;
433 (!cascade(base, &base->tv2, INDEX(0))) &&
434 (!cascade(base, &base->tv3, INDEX(1))) &&
435 !cascade(base, &base->tv4, INDEX(2)))
436 cascade(base, &base->tv5, INDEX(3));
437 ++base->timer_jiffies;
438 list_splice_init(base->tv1.vec + index, &work_list);
439 while (!list_empty(head)) {
440 void (*fn)(unsigned long);
443 timer = list_entry(head->next,struct timer_list,entry);
444 fn = timer->function;
447 set_running_timer(base, timer);
448 detach_timer(timer, 1);
449 spin_unlock_irq(&base->lock);
451 int preempt_count = preempt_count();
453 if (preempt_count != preempt_count()) {
454 printk(KERN_WARNING "huh, entered %p "
455 "with preempt_count %08x, exited"
462 spin_lock_irq(&base->lock);
465 set_running_timer(base, NULL);
466 spin_unlock_irq(&base->lock);
469 #ifdef CONFIG_NO_IDLE_HZ
471 * Find out when the next timer event is due to happen. This
472 * is used on S/390 to stop all activity when a cpus is idle.
473 * This functions needs to be called disabled.
475 unsigned long next_timer_interrupt(void)
478 struct list_head *list;
479 struct timer_list *nte;
480 unsigned long expires;
481 unsigned long hr_expires = MAX_JIFFY_OFFSET;
486 hr_delta = hrtimer_get_next_event();
487 if (hr_delta.tv64 != KTIME_MAX) {
488 struct timespec tsdelta;
489 tsdelta = ktime_to_timespec(hr_delta);
490 hr_expires = timespec_to_jiffies(&tsdelta);
492 return hr_expires + jiffies;
494 hr_expires += jiffies;
496 base = __get_cpu_var(tvec_bases);
497 spin_lock(&base->lock);
498 expires = base->timer_jiffies + (LONG_MAX >> 1);
501 /* Look for timer events in tv1. */
502 j = base->timer_jiffies & TVR_MASK;
504 list_for_each_entry(nte, base->tv1.vec + j, entry) {
505 expires = nte->expires;
506 if (j < (base->timer_jiffies & TVR_MASK))
507 list = base->tv2.vec + (INDEX(0));
510 j = (j + 1) & TVR_MASK;
511 } while (j != (base->timer_jiffies & TVR_MASK));
514 varray[0] = &base->tv2;
515 varray[1] = &base->tv3;
516 varray[2] = &base->tv4;
517 varray[3] = &base->tv5;
518 for (i = 0; i < 4; i++) {
521 if (list_empty(varray[i]->vec + j)) {
522 j = (j + 1) & TVN_MASK;
525 list_for_each_entry(nte, varray[i]->vec + j, entry)
526 if (time_before(nte->expires, expires))
527 expires = nte->expires;
528 if (j < (INDEX(i)) && i < 3)
529 list = varray[i + 1]->vec + (INDEX(i + 1));
531 } while (j != (INDEX(i)));
536 * The search wrapped. We need to look at the next list
537 * from next tv element that would cascade into tv element
538 * where we found the timer element.
540 list_for_each_entry(nte, list, entry) {
541 if (time_before(nte->expires, expires))
542 expires = nte->expires;
545 spin_unlock(&base->lock);
548 * It can happen that other CPUs service timer IRQs and increment
549 * jiffies, but we have not yet got a local timer tick to process
550 * the timer wheels. In that case, the expiry time can be before
551 * jiffies, but since the high-resolution timer here is relative to
552 * jiffies, the default expression when high-resolution timers are
555 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
557 * would falsely evaluate to true. If that is the case, just
558 * return jiffies so that we can immediately fire the local timer
560 if (time_before(expires, jiffies))
564 * It can happen that other CPUs service timer IRQs and increment
565 * jiffies, but we have not yet got a local timer tick to process
566 * the timer wheels. In that case, the expiry time can be before
567 * jiffies, but since the high-resolution timer here is relative to
568 * jiffies, the default expression when high-resolution timers are
571 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
573 * would falsely evaluate to true. If that is the case, just
574 * return jiffies so that we can immediately fire the local timer
576 if (time_before(expires, jiffies))
579 if (time_before(hr_expires, expires))
586 /******************************************************************/
589 * Timekeeping variables
591 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
592 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
596 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
597 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
598 * at zero at system boot time, so wall_to_monotonic will be negative,
599 * however, we will ALWAYS keep the tv_nsec part positive so we can use
600 * the usual normalization.
602 struct timespec xtime __attribute__ ((aligned (16)));
603 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
605 EXPORT_SYMBOL(xtime);
607 /* Don't completely fail for HZ > 500. */
608 int tickadj = 500/HZ ? : 1; /* microsecs */
612 * phase-lock loop variables
614 /* TIME_ERROR prevents overwriting the CMOS clock */
615 int time_state = TIME_OK; /* clock synchronization status */
616 int time_status = STA_UNSYNC; /* clock status bits */
617 long time_offset; /* time adjustment (us) */
618 long time_constant = 2; /* pll time constant */
619 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
620 long time_precision = 1; /* clock precision (us) */
621 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
622 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
623 static long time_phase; /* phase offset (scaled us) */
624 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
625 /* frequency offset (scaled ppm)*/
626 static long time_adj; /* tick adjust (scaled 1 / HZ) */
627 long time_reftime; /* time at last adjustment (s) */
629 long time_next_adjust;
632 * this routine handles the overflow of the microsecond field
634 * The tricky bits of code to handle the accurate clock support
635 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
636 * They were originally developed for SUN and DEC kernels.
637 * All the kudos should go to Dave for this stuff.
640 static void second_overflow(void)
644 /* Bump the maxerror field */
645 time_maxerror += time_tolerance >> SHIFT_USEC;
646 if (time_maxerror > NTP_PHASE_LIMIT) {
647 time_maxerror = NTP_PHASE_LIMIT;
648 time_status |= STA_UNSYNC;
652 * Leap second processing. If in leap-insert state at the end of the
653 * day, the system clock is set back one second; if in leap-delete
654 * state, the system clock is set ahead one second. The microtime()
655 * routine or external clock driver will insure that reported time is
656 * always monotonic. The ugly divides should be replaced.
658 switch (time_state) {
660 if (time_status & STA_INS)
661 time_state = TIME_INS;
662 else if (time_status & STA_DEL)
663 time_state = TIME_DEL;
666 if (xtime.tv_sec % 86400 == 0) {
668 wall_to_monotonic.tv_sec++;
670 * The timer interpolator will make time change
671 * gradually instead of an immediate jump by one second
673 time_interpolator_update(-NSEC_PER_SEC);
674 time_state = TIME_OOP;
676 printk(KERN_NOTICE "Clock: inserting leap second "
681 if ((xtime.tv_sec + 1) % 86400 == 0) {
683 wall_to_monotonic.tv_sec--;
685 * Use of time interpolator for a gradual change of
688 time_interpolator_update(NSEC_PER_SEC);
689 time_state = TIME_WAIT;
691 printk(KERN_NOTICE "Clock: deleting leap second "
696 time_state = TIME_WAIT;
699 if (!(time_status & (STA_INS | STA_DEL)))
700 time_state = TIME_OK;
704 * Compute the phase adjustment for the next second. In PLL mode, the
705 * offset is reduced by a fixed factor times the time constant. In FLL
706 * mode the offset is used directly. In either mode, the maximum phase
707 * adjustment for each second is clamped so as to spread the adjustment
708 * over not more than the number of seconds between updates.
711 if (!(time_status & STA_FLL))
712 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
713 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
714 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
715 time_offset -= ltemp;
716 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
719 * Compute the frequency estimate and additional phase adjustment due
720 * to frequency error for the next second.
723 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
727 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
728 * get 128.125; => only 0.125% error (p. 14)
730 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
734 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
735 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
737 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
741 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
742 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
744 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
749 * Returns how many microseconds we need to add to xtime this tick
750 * in doing an adjustment requested with adjtime.
752 static long adjtime_adjustment(void)
754 long time_adjust_step;
756 time_adjust_step = time_adjust;
757 if (time_adjust_step) {
759 * We are doing an adjtime thing. Prepare time_adjust_step to
760 * be within bounds. Note that a positive time_adjust means we
761 * want the clock to run faster.
763 * Limit the amount of the step to be in the range
764 * -tickadj .. +tickadj
766 time_adjust_step = min(time_adjust_step, (long)tickadj);
767 time_adjust_step = max(time_adjust_step, (long)-tickadj);
769 return time_adjust_step;
772 /* in the NTP reference this is called "hardclock()" */
773 static void update_wall_time_one_tick(void)
775 long time_adjust_step, delta_nsec;
777 time_adjust_step = adjtime_adjustment();
778 if (time_adjust_step)
779 /* Reduce by this step the amount of time left */
780 time_adjust -= time_adjust_step;
781 delta_nsec = tick_nsec + time_adjust_step * 1000;
783 * Advance the phase, once it gets to one microsecond, then
784 * advance the tick more.
786 time_phase += time_adj;
787 if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) {
788 long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10));
789 time_phase -= ltemp << (SHIFT_SCALE - 10);
792 xtime.tv_nsec += delta_nsec;
793 time_interpolator_update(delta_nsec);
795 /* Changes by adjtime() do not take effect till next tick. */
796 if (time_next_adjust != 0) {
797 time_adjust = time_next_adjust;
798 time_next_adjust = 0;
803 * Return how long ticks are at the moment, that is, how much time
804 * update_wall_time_one_tick will add to xtime next time we call it
805 * (assuming no calls to do_adjtimex in the meantime).
806 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
807 * bits to the right of the binary point.
808 * This function has no side-effects.
810 u64 current_tick_length(void)
814 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
815 return ((u64) delta_nsec << (SHIFT_SCALE - 10)) + time_adj;
819 * Using a loop looks inefficient, but "ticks" is
820 * usually just one (we shouldn't be losing ticks,
821 * we're doing this this way mainly for interrupt
822 * latency reasons, not because we think we'll
823 * have lots of lost timer ticks
825 static void update_wall_time(unsigned long ticks)
829 update_wall_time_one_tick();
830 if (xtime.tv_nsec >= 1000000000) {
831 xtime.tv_nsec -= 1000000000;
839 * Called from the timer interrupt handler to charge one tick to the current
840 * process. user_tick is 1 if the tick is user time, 0 for system.
842 void update_process_times(int user_tick)
844 struct task_struct *p = current;
845 int cpu = smp_processor_id();
847 /* Note: this timer irq context must be accounted for as well. */
849 account_user_time(p, jiffies_to_cputime(1));
851 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
853 if (rcu_pending(cpu))
854 rcu_check_callbacks(cpu, user_tick);
856 run_posix_cpu_timers(p);
860 * Nr of active tasks - counted in fixed-point numbers
862 static unsigned long count_active_tasks(void)
864 return nr_active() * FIXED_1;
868 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
869 * imply that avenrun[] is the standard name for this kind of thing.
870 * Nothing else seems to be standardized: the fractional size etc
871 * all seem to differ on different machines.
873 * Requires xtime_lock to access.
875 unsigned long avenrun[3];
877 EXPORT_SYMBOL(avenrun);
880 * calc_load - given tick count, update the avenrun load estimates.
881 * This is called while holding a write_lock on xtime_lock.
883 static inline void calc_load(unsigned long ticks)
885 unsigned long active_tasks; /* fixed-point */
886 static int count = LOAD_FREQ;
891 active_tasks = count_active_tasks();
892 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
893 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
894 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
898 /* jiffies at the most recent update of wall time */
899 unsigned long wall_jiffies = INITIAL_JIFFIES;
902 * This read-write spinlock protects us from races in SMP while
903 * playing with xtime and avenrun.
905 #ifndef ARCH_HAVE_XTIME_LOCK
906 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
908 EXPORT_SYMBOL(xtime_lock);
912 * This function runs timers and the timer-tq in bottom half context.
914 static void run_timer_softirq(struct softirq_action *h)
916 tvec_base_t *base = __get_cpu_var(tvec_bases);
918 hrtimer_run_queues();
919 if (time_after_eq(jiffies, base->timer_jiffies))
924 * Called by the local, per-CPU timer interrupt on SMP.
926 void run_local_timers(void)
928 raise_softirq(TIMER_SOFTIRQ);
933 * Called by the timer interrupt. xtime_lock must already be taken
936 static inline void update_times(void)
940 ticks = jiffies - wall_jiffies;
942 wall_jiffies += ticks;
943 update_wall_time(ticks);
949 * The 64-bit jiffies value is not atomic - you MUST NOT read it
950 * without sampling the sequence number in xtime_lock.
951 * jiffies is defined in the linker script...
954 void do_timer(struct pt_regs *regs)
957 /* prevent loading jiffies before storing new jiffies_64 value. */
962 #ifdef __ARCH_WANT_SYS_ALARM
965 * For backwards compatibility? This can be done in libc so Alpha
966 * and all newer ports shouldn't need it.
968 asmlinkage unsigned long sys_alarm(unsigned int seconds)
970 return alarm_setitimer(seconds);
977 * sys_getpid - return the thread group id of the current process
979 * Note, despite the name, this returns the tgid not the pid. The tgid and
980 * the pid are identical unless CLONE_THREAD was specified on clone() in
981 * which case the tgid is the same in all threads of the same group.
983 * This is SMP safe as current->tgid does not change.
985 asmlinkage long sys_getpid(void)
987 return vx_map_tgid(current->tgid);
991 * Accessing ->real_parent is not SMP-safe, it could
992 * change from under us. However, we can use a stale
993 * value of ->real_parent under rcu_read_lock(), see
994 * release_task()->call_rcu(delayed_put_task_struct).
996 asmlinkage long sys_getppid(void)
1001 pid = rcu_dereference(current->real_parent)->tgid;
1003 return vx_map_pid(pid);
1009 * The Alpha uses getxpid, getxuid, and getxgid instead.
1012 asmlinkage long do_getxpid(long *ppid)
1014 *ppid = sys_getppid();
1015 return sys_getpid();
1020 asmlinkage long sys_getuid(void)
1022 /* Only we change this so SMP safe */
1023 return current->uid;
1026 asmlinkage long sys_geteuid(void)
1028 /* Only we change this so SMP safe */
1029 return current->euid;
1032 asmlinkage long sys_getgid(void)
1034 /* Only we change this so SMP safe */
1035 return current->gid;
1038 asmlinkage long sys_getegid(void)
1040 /* Only we change this so SMP safe */
1041 return current->egid;
1046 static void process_timeout(unsigned long __data)
1048 wake_up_process((task_t *)__data);
1052 * schedule_timeout - sleep until timeout
1053 * @timeout: timeout value in jiffies
1055 * Make the current task sleep until @timeout jiffies have
1056 * elapsed. The routine will return immediately unless
1057 * the current task state has been set (see set_current_state()).
1059 * You can set the task state as follows -
1061 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1062 * pass before the routine returns. The routine will return 0
1064 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1065 * delivered to the current task. In this case the remaining time
1066 * in jiffies will be returned, or 0 if the timer expired in time
1068 * The current task state is guaranteed to be TASK_RUNNING when this
1071 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1072 * the CPU away without a bound on the timeout. In this case the return
1073 * value will be %MAX_SCHEDULE_TIMEOUT.
1075 * In all cases the return value is guaranteed to be non-negative.
1077 fastcall signed long __sched schedule_timeout(signed long timeout)
1079 struct timer_list timer;
1080 unsigned long expire;
1084 case MAX_SCHEDULE_TIMEOUT:
1086 * These two special cases are useful to be comfortable
1087 * in the caller. Nothing more. We could take
1088 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1089 * but I' d like to return a valid offset (>=0) to allow
1090 * the caller to do everything it want with the retval.
1096 * Another bit of PARANOID. Note that the retval will be
1097 * 0 since no piece of kernel is supposed to do a check
1098 * for a negative retval of schedule_timeout() (since it
1099 * should never happens anyway). You just have the printk()
1100 * that will tell you if something is gone wrong and where.
1104 printk(KERN_ERR "schedule_timeout: wrong timeout "
1105 "value %lx from %p\n", timeout,
1106 __builtin_return_address(0));
1107 current->state = TASK_RUNNING;
1112 expire = timeout + jiffies;
1114 setup_timer(&timer, process_timeout, (unsigned long)current);
1115 __mod_timer(&timer, expire);
1117 del_singleshot_timer_sync(&timer);
1119 timeout = expire - jiffies;
1122 return timeout < 0 ? 0 : timeout;
1124 EXPORT_SYMBOL(schedule_timeout);
1127 * We can use __set_current_state() here because schedule_timeout() calls
1128 * schedule() unconditionally.
1130 signed long __sched schedule_timeout_interruptible(signed long timeout)
1132 __set_current_state(TASK_INTERRUPTIBLE);
1133 return schedule_timeout(timeout);
1135 EXPORT_SYMBOL(schedule_timeout_interruptible);
1137 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1139 __set_current_state(TASK_UNINTERRUPTIBLE);
1140 return schedule_timeout(timeout);
1142 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1144 /* Thread ID - the internal kernel "pid" */
1145 asmlinkage long sys_gettid(void)
1147 return current->pid;
1151 * sys_sysinfo - fill in sysinfo struct
1153 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1156 unsigned long mem_total, sav_total;
1157 unsigned int mem_unit, bitcount;
1160 memset((char *)&val, 0, sizeof(struct sysinfo));
1164 seq = read_seqbegin(&xtime_lock);
1167 * This is annoying. The below is the same thing
1168 * posix_get_clock_monotonic() does, but it wants to
1169 * take the lock which we want to cover the loads stuff
1173 getnstimeofday(&tp);
1174 tp.tv_sec += wall_to_monotonic.tv_sec;
1175 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1176 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1177 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1180 if (vx_flags(VXF_VIRT_UPTIME, 0))
1181 vx_vsi_uptime(&tp, NULL);
1182 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1184 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1185 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1186 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1188 val.procs = nr_threads;
1189 } while (read_seqretry(&xtime_lock, seq));
1195 * If the sum of all the available memory (i.e. ram + swap)
1196 * is less than can be stored in a 32 bit unsigned long then
1197 * we can be binary compatible with 2.2.x kernels. If not,
1198 * well, in that case 2.2.x was broken anyways...
1200 * -Erik Andersen <andersee@debian.org>
1203 mem_total = val.totalram + val.totalswap;
1204 if (mem_total < val.totalram || mem_total < val.totalswap)
1207 mem_unit = val.mem_unit;
1208 while (mem_unit > 1) {
1211 sav_total = mem_total;
1213 if (mem_total < sav_total)
1218 * If mem_total did not overflow, multiply all memory values by
1219 * val.mem_unit and set it to 1. This leaves things compatible
1220 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1225 val.totalram <<= bitcount;
1226 val.freeram <<= bitcount;
1227 val.sharedram <<= bitcount;
1228 val.bufferram <<= bitcount;
1229 val.totalswap <<= bitcount;
1230 val.freeswap <<= bitcount;
1231 val.totalhigh <<= bitcount;
1232 val.freehigh <<= bitcount;
1235 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1241 static int __devinit init_timers_cpu(int cpu)
1245 static char __devinitdata tvec_base_done[NR_CPUS];
1247 if (!tvec_base_done[cpu]) {
1248 static char boot_done;
1252 * The APs use this path later in boot
1254 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1258 memset(base, 0, sizeof(*base));
1259 per_cpu(tvec_bases, cpu) = base;
1262 * This is for the boot CPU - we use compile-time
1263 * static initialisation because per-cpu memory isn't
1264 * ready yet and because the memory allocators are not
1265 * initialised either.
1268 base = &boot_tvec_bases;
1270 tvec_base_done[cpu] = 1;
1272 base = per_cpu(tvec_bases, cpu);
1275 spin_lock_init(&base->lock);
1276 for (j = 0; j < TVN_SIZE; j++) {
1277 INIT_LIST_HEAD(base->tv5.vec + j);
1278 INIT_LIST_HEAD(base->tv4.vec + j);
1279 INIT_LIST_HEAD(base->tv3.vec + j);
1280 INIT_LIST_HEAD(base->tv2.vec + j);
1282 for (j = 0; j < TVR_SIZE; j++)
1283 INIT_LIST_HEAD(base->tv1.vec + j);
1285 base->timer_jiffies = jiffies;
1289 #ifdef CONFIG_HOTPLUG_CPU
1290 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1292 struct timer_list *timer;
1294 while (!list_empty(head)) {
1295 timer = list_entry(head->next, struct timer_list, entry);
1296 detach_timer(timer, 0);
1297 timer->base = new_base;
1298 internal_add_timer(new_base, timer);
1302 static void __devinit migrate_timers(int cpu)
1304 tvec_base_t *old_base;
1305 tvec_base_t *new_base;
1308 BUG_ON(cpu_online(cpu));
1309 old_base = per_cpu(tvec_bases, cpu);
1310 new_base = get_cpu_var(tvec_bases);
1312 local_irq_disable();
1313 spin_lock(&new_base->lock);
1314 spin_lock(&old_base->lock);
1316 BUG_ON(old_base->running_timer);
1318 for (i = 0; i < TVR_SIZE; i++)
1319 migrate_timer_list(new_base, old_base->tv1.vec + i);
1320 for (i = 0; i < TVN_SIZE; i++) {
1321 migrate_timer_list(new_base, old_base->tv2.vec + i);
1322 migrate_timer_list(new_base, old_base->tv3.vec + i);
1323 migrate_timer_list(new_base, old_base->tv4.vec + i);
1324 migrate_timer_list(new_base, old_base->tv5.vec + i);
1327 spin_unlock(&old_base->lock);
1328 spin_unlock(&new_base->lock);
1330 put_cpu_var(tvec_bases);
1332 #endif /* CONFIG_HOTPLUG_CPU */
1334 static int timer_cpu_notify(struct notifier_block *self,
1335 unsigned long action, void *hcpu)
1337 long cpu = (long)hcpu;
1339 case CPU_UP_PREPARE:
1340 if (init_timers_cpu(cpu) < 0)
1343 #ifdef CONFIG_HOTPLUG_CPU
1345 migrate_timers(cpu);
1354 static struct notifier_block timers_nb = {
1355 .notifier_call = timer_cpu_notify,
1359 void __init init_timers(void)
1361 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1362 (void *)(long)smp_processor_id());
1363 register_cpu_notifier(&timers_nb);
1364 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1367 #ifdef CONFIG_TIME_INTERPOLATION
1369 struct time_interpolator *time_interpolator __read_mostly;
1370 static struct time_interpolator *time_interpolator_list __read_mostly;
1371 static DEFINE_SPINLOCK(time_interpolator_lock);
1373 static inline u64 time_interpolator_get_cycles(unsigned int src)
1375 unsigned long (*x)(void);
1379 case TIME_SOURCE_FUNCTION:
1380 x = time_interpolator->addr;
1383 case TIME_SOURCE_MMIO64 :
1384 return readq_relaxed((void __iomem *)time_interpolator->addr);
1386 case TIME_SOURCE_MMIO32 :
1387 return readl_relaxed((void __iomem *)time_interpolator->addr);
1389 default: return get_cycles();
1393 static inline u64 time_interpolator_get_counter(int writelock)
1395 unsigned int src = time_interpolator->source;
1397 if (time_interpolator->jitter)
1403 lcycle = time_interpolator->last_cycle;
1404 now = time_interpolator_get_cycles(src);
1405 if (lcycle && time_after(lcycle, now))
1408 /* When holding the xtime write lock, there's no need
1409 * to add the overhead of the cmpxchg. Readers are
1410 * force to retry until the write lock is released.
1413 time_interpolator->last_cycle = now;
1416 /* Keep track of the last timer value returned. The use of cmpxchg here
1417 * will cause contention in an SMP environment.
1419 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1423 return time_interpolator_get_cycles(src);
1426 void time_interpolator_reset(void)
1428 time_interpolator->offset = 0;
1429 time_interpolator->last_counter = time_interpolator_get_counter(1);
1432 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1434 unsigned long time_interpolator_get_offset(void)
1436 /* If we do not have a time interpolator set up then just return zero */
1437 if (!time_interpolator)
1440 return time_interpolator->offset +
1441 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1444 #define INTERPOLATOR_ADJUST 65536
1445 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1447 static void time_interpolator_update(long delta_nsec)
1450 unsigned long offset;
1452 /* If there is no time interpolator set up then do nothing */
1453 if (!time_interpolator)
1457 * The interpolator compensates for late ticks by accumulating the late
1458 * time in time_interpolator->offset. A tick earlier than expected will
1459 * lead to a reset of the offset and a corresponding jump of the clock
1460 * forward. Again this only works if the interpolator clock is running
1461 * slightly slower than the regular clock and the tuning logic insures
1465 counter = time_interpolator_get_counter(1);
1466 offset = time_interpolator->offset +
1467 GET_TI_NSECS(counter, time_interpolator);
1469 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1470 time_interpolator->offset = offset - delta_nsec;
1472 time_interpolator->skips++;
1473 time_interpolator->ns_skipped += delta_nsec - offset;
1474 time_interpolator->offset = 0;
1476 time_interpolator->last_counter = counter;
1478 /* Tuning logic for time interpolator invoked every minute or so.
1479 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1480 * Increase interpolator clock speed if we skip too much time.
1482 if (jiffies % INTERPOLATOR_ADJUST == 0)
1484 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1485 time_interpolator->nsec_per_cyc--;
1486 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1487 time_interpolator->nsec_per_cyc++;
1488 time_interpolator->skips = 0;
1489 time_interpolator->ns_skipped = 0;
1494 is_better_time_interpolator(struct time_interpolator *new)
1496 if (!time_interpolator)
1498 return new->frequency > 2*time_interpolator->frequency ||
1499 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1503 register_time_interpolator(struct time_interpolator *ti)
1505 unsigned long flags;
1508 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1510 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1511 spin_lock(&time_interpolator_lock);
1512 write_seqlock_irqsave(&xtime_lock, flags);
1513 if (is_better_time_interpolator(ti)) {
1514 time_interpolator = ti;
1515 time_interpolator_reset();
1517 write_sequnlock_irqrestore(&xtime_lock, flags);
1519 ti->next = time_interpolator_list;
1520 time_interpolator_list = ti;
1521 spin_unlock(&time_interpolator_lock);
1525 unregister_time_interpolator(struct time_interpolator *ti)
1527 struct time_interpolator *curr, **prev;
1528 unsigned long flags;
1530 spin_lock(&time_interpolator_lock);
1531 prev = &time_interpolator_list;
1532 for (curr = *prev; curr; curr = curr->next) {
1540 write_seqlock_irqsave(&xtime_lock, flags);
1541 if (ti == time_interpolator) {
1542 /* we lost the best time-interpolator: */
1543 time_interpolator = NULL;
1544 /* find the next-best interpolator */
1545 for (curr = time_interpolator_list; curr; curr = curr->next)
1546 if (is_better_time_interpolator(curr))
1547 time_interpolator = curr;
1548 time_interpolator_reset();
1550 write_sequnlock_irqrestore(&xtime_lock, flags);
1551 spin_unlock(&time_interpolator_lock);
1553 #endif /* CONFIG_TIME_INTERPOLATION */
1556 * msleep - sleep safely even with waitqueue interruptions
1557 * @msecs: Time in milliseconds to sleep for
1559 void msleep(unsigned int msecs)
1561 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1564 timeout = schedule_timeout_uninterruptible(timeout);
1567 EXPORT_SYMBOL(msleep);
1570 * msleep_interruptible - sleep waiting for signals
1571 * @msecs: Time in milliseconds to sleep for
1573 unsigned long msleep_interruptible(unsigned int msecs)
1575 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1577 while (timeout && !signal_pending(current))
1578 timeout = schedule_timeout_interruptible(timeout);
1579 return jiffies_to_msecs(timeout);
1582 EXPORT_SYMBOL(msleep_interruptible);