4 * Kernel internal timers, kernel timekeeping, basic process system calls
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
37 #include <linux/vs_cvirt.h>
38 #include <linux/vserver/sched.h>
40 #include <asm/uaccess.h>
41 #include <asm/unistd.h>
42 #include <asm/div64.h>
43 #include <asm/timex.h>
46 #ifdef CONFIG_TIME_INTERPOLATION
47 static void time_interpolator_update(long delta_nsec);
49 #define time_interpolator_update(x)
52 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
54 EXPORT_SYMBOL(jiffies_64);
57 * per-CPU timer vector definitions:
59 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
60 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
61 #define TVN_SIZE (1 << TVN_BITS)
62 #define TVR_SIZE (1 << TVR_BITS)
63 #define TVN_MASK (TVN_SIZE - 1)
64 #define TVR_MASK (TVR_SIZE - 1)
66 typedef struct tvec_s {
67 struct list_head vec[TVN_SIZE];
70 typedef struct tvec_root_s {
71 struct list_head vec[TVR_SIZE];
74 struct tvec_t_base_s {
76 struct timer_list *running_timer;
77 unsigned long timer_jiffies;
83 } ____cacheline_aligned_in_smp;
85 typedef struct tvec_t_base_s tvec_base_t;
87 tvec_base_t boot_tvec_bases;
88 EXPORT_SYMBOL(boot_tvec_bases);
89 static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = { &boot_tvec_bases };
91 static inline void set_running_timer(tvec_base_t *base,
92 struct timer_list *timer)
95 base->running_timer = timer;
99 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
101 unsigned long expires = timer->expires;
102 unsigned long idx = expires - base->timer_jiffies;
103 struct list_head *vec;
105 if (idx < TVR_SIZE) {
106 int i = expires & TVR_MASK;
107 vec = base->tv1.vec + i;
108 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
109 int i = (expires >> TVR_BITS) & TVN_MASK;
110 vec = base->tv2.vec + i;
111 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
112 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
113 vec = base->tv3.vec + i;
114 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
115 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
116 vec = base->tv4.vec + i;
117 } else if ((signed long) idx < 0) {
119 * Can happen if you add a timer with expires == jiffies,
120 * or you set a timer to go off in the past
122 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
125 /* If the timeout is larger than 0xffffffff on 64-bit
126 * architectures then we use the maximum timeout:
128 if (idx > 0xffffffffUL) {
130 expires = idx + base->timer_jiffies;
132 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
133 vec = base->tv5.vec + i;
138 list_add_tail(&timer->entry, vec);
142 * init_timer - initialize a timer.
143 * @timer: the timer to be initialized
145 * init_timer() must be done to a timer prior calling *any* of the
146 * other timer functions.
148 void fastcall init_timer(struct timer_list *timer)
150 timer->entry.next = NULL;
151 timer->base = per_cpu(tvec_bases, raw_smp_processor_id());
153 EXPORT_SYMBOL(init_timer);
155 static inline void detach_timer(struct timer_list *timer,
158 struct list_head *entry = &timer->entry;
160 __list_del(entry->prev, entry->next);
163 entry->prev = LIST_POISON2;
167 * We are using hashed locking: holding per_cpu(tvec_bases).lock
168 * means that all timers which are tied to this base via timer->base are
169 * locked, and the base itself is locked too.
171 * So __run_timers/migrate_timers can safely modify all timers which could
172 * be found on ->tvX lists.
174 * When the timer's base is locked, and the timer removed from list, it is
175 * possible to set timer->base = NULL and drop the lock: the timer remains
178 static tvec_base_t *lock_timer_base(struct timer_list *timer,
179 unsigned long *flags)
185 if (likely(base != NULL)) {
186 spin_lock_irqsave(&base->lock, *flags);
187 if (likely(base == timer->base))
189 /* The timer has migrated to another CPU */
190 spin_unlock_irqrestore(&base->lock, *flags);
196 int __mod_timer(struct timer_list *timer, unsigned long expires)
198 tvec_base_t *base, *new_base;
202 BUG_ON(!timer->function);
204 base = lock_timer_base(timer, &flags);
206 if (timer_pending(timer)) {
207 detach_timer(timer, 0);
211 new_base = __get_cpu_var(tvec_bases);
213 if (base != new_base) {
215 * We are trying to schedule the timer on the local CPU.
216 * However we can't change timer's base while it is running,
217 * otherwise del_timer_sync() can't detect that the timer's
218 * handler yet has not finished. This also guarantees that
219 * the timer is serialized wrt itself.
221 if (likely(base->running_timer != timer)) {
222 /* See the comment in lock_timer_base() */
224 spin_unlock(&base->lock);
226 spin_lock(&base->lock);
231 timer->expires = expires;
232 internal_add_timer(base, timer);
233 spin_unlock_irqrestore(&base->lock, flags);
238 EXPORT_SYMBOL(__mod_timer);
241 * add_timer_on - start a timer on a particular CPU
242 * @timer: the timer to be added
243 * @cpu: the CPU to start it on
245 * This is not very scalable on SMP. Double adds are not possible.
247 void add_timer_on(struct timer_list *timer, int cpu)
249 tvec_base_t *base = per_cpu(tvec_bases, cpu);
252 BUG_ON(timer_pending(timer) || !timer->function);
253 spin_lock_irqsave(&base->lock, flags);
255 internal_add_timer(base, timer);
256 spin_unlock_irqrestore(&base->lock, flags);
261 * mod_timer - modify a timer's timeout
262 * @timer: the timer to be modified
264 * mod_timer is a more efficient way to update the expire field of an
265 * active timer (if the timer is inactive it will be activated)
267 * mod_timer(timer, expires) is equivalent to:
269 * del_timer(timer); timer->expires = expires; add_timer(timer);
271 * Note that if there are multiple unserialized concurrent users of the
272 * same timer, then mod_timer() is the only safe way to modify the timeout,
273 * since add_timer() cannot modify an already running timer.
275 * The function returns whether it has modified a pending timer or not.
276 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
277 * active timer returns 1.)
279 int mod_timer(struct timer_list *timer, unsigned long expires)
281 BUG_ON(!timer->function);
284 * This is a common optimization triggered by the
285 * networking code - if the timer is re-modified
286 * to be the same thing then just return:
288 if (timer->expires == expires && timer_pending(timer))
291 return __mod_timer(timer, expires);
294 EXPORT_SYMBOL(mod_timer);
297 * del_timer - deactive a timer.
298 * @timer: the timer to be deactivated
300 * del_timer() deactivates a timer - this works on both active and inactive
303 * The function returns whether it has deactivated a pending timer or not.
304 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
305 * active timer returns 1.)
307 int del_timer(struct timer_list *timer)
313 if (timer_pending(timer)) {
314 base = lock_timer_base(timer, &flags);
315 if (timer_pending(timer)) {
316 detach_timer(timer, 1);
319 spin_unlock_irqrestore(&base->lock, flags);
325 EXPORT_SYMBOL(del_timer);
329 * This function tries to deactivate a timer. Upon successful (ret >= 0)
330 * exit the timer is not queued and the handler is not running on any CPU.
332 * It must not be called from interrupt contexts.
334 int try_to_del_timer_sync(struct timer_list *timer)
340 base = lock_timer_base(timer, &flags);
342 if (base->running_timer == timer)
346 if (timer_pending(timer)) {
347 detach_timer(timer, 1);
351 spin_unlock_irqrestore(&base->lock, flags);
357 * del_timer_sync - deactivate a timer and wait for the handler to finish.
358 * @timer: the timer to be deactivated
360 * This function only differs from del_timer() on SMP: besides deactivating
361 * the timer it also makes sure the handler has finished executing on other
364 * Synchronization rules: callers must prevent restarting of the timer,
365 * otherwise this function is meaningless. It must not be called from
366 * interrupt contexts. The caller must not hold locks which would prevent
367 * completion of the timer's handler. The timer's handler must not call
368 * add_timer_on(). Upon exit the timer is not queued and the handler is
369 * not running on any CPU.
371 * The function returns whether it has deactivated a pending timer or not.
373 int del_timer_sync(struct timer_list *timer)
376 int ret = try_to_del_timer_sync(timer);
382 EXPORT_SYMBOL(del_timer_sync);
385 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
387 /* cascade all the timers from tv up one level */
388 struct list_head *head, *curr;
390 head = tv->vec + index;
393 * We are removing _all_ timers from the list, so we don't have to
394 * detach them individually, just clear the list afterwards.
396 while (curr != head) {
397 struct timer_list *tmp;
399 tmp = list_entry(curr, struct timer_list, entry);
400 BUG_ON(tmp->base != base);
402 internal_add_timer(base, tmp);
404 INIT_LIST_HEAD(head);
410 * __run_timers - run all expired timers (if any) on this CPU.
411 * @base: the timer vector to be processed.
413 * This function cascades all vectors and executes all expired timer
416 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
418 static inline void __run_timers(tvec_base_t *base)
420 struct timer_list *timer;
422 spin_lock_irq(&base->lock);
423 while (time_after_eq(jiffies, base->timer_jiffies)) {
424 struct list_head work_list = LIST_HEAD_INIT(work_list);
425 struct list_head *head = &work_list;
426 int index = base->timer_jiffies & TVR_MASK;
432 (!cascade(base, &base->tv2, INDEX(0))) &&
433 (!cascade(base, &base->tv3, INDEX(1))) &&
434 !cascade(base, &base->tv4, INDEX(2)))
435 cascade(base, &base->tv5, INDEX(3));
436 ++base->timer_jiffies;
437 list_splice_init(base->tv1.vec + index, &work_list);
438 while (!list_empty(head)) {
439 void (*fn)(unsigned long);
442 timer = list_entry(head->next,struct timer_list,entry);
443 fn = timer->function;
446 set_running_timer(base, timer);
447 detach_timer(timer, 1);
448 spin_unlock_irq(&base->lock);
450 int preempt_count = preempt_count();
452 if (preempt_count != preempt_count()) {
453 printk(KERN_WARNING "huh, entered %p "
454 "with preempt_count %08x, exited"
461 spin_lock_irq(&base->lock);
464 set_running_timer(base, NULL);
465 spin_unlock_irq(&base->lock);
468 #ifdef CONFIG_NO_IDLE_HZ
470 * Find out when the next timer event is due to happen. This
471 * is used on S/390 to stop all activity when a cpus is idle.
472 * This functions needs to be called disabled.
474 unsigned long next_timer_interrupt(void)
477 struct list_head *list;
478 struct timer_list *nte;
479 unsigned long expires;
480 unsigned long hr_expires = MAX_JIFFY_OFFSET;
485 hr_delta = hrtimer_get_next_event();
486 if (hr_delta.tv64 != KTIME_MAX) {
487 struct timespec tsdelta;
488 tsdelta = ktime_to_timespec(hr_delta);
489 hr_expires = timespec_to_jiffies(&tsdelta);
491 return hr_expires + jiffies;
493 hr_expires += jiffies;
495 base = __get_cpu_var(tvec_bases);
496 spin_lock(&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->lock);
547 * It can happen that other CPUs service timer IRQs and increment
548 * jiffies, but we have not yet got a local timer tick to process
549 * the timer wheels. In that case, the expiry time can be before
550 * jiffies, but since the high-resolution timer here is relative to
551 * jiffies, the default expression when high-resolution timers are
554 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
556 * would falsely evaluate to true. If that is the case, just
557 * return jiffies so that we can immediately fire the local timer
559 if (time_before(expires, jiffies))
563 * It can happen that other CPUs service timer IRQs and increment
564 * jiffies, but we have not yet got a local timer tick to process
565 * the timer wheels. In that case, the expiry time can be before
566 * jiffies, but since the high-resolution timer here is relative to
567 * jiffies, the default expression when high-resolution timers are
570 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
572 * would falsely evaluate to true. If that is the case, just
573 * return jiffies so that we can immediately fire the local timer
575 if (time_before(expires, jiffies))
578 if (time_before(hr_expires, expires))
585 /******************************************************************/
588 * Timekeeping variables
590 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
591 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
595 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
596 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
597 * at zero at system boot time, so wall_to_monotonic will be negative,
598 * however, we will ALWAYS keep the tv_nsec part positive so we can use
599 * the usual normalization.
601 struct timespec xtime __attribute__ ((aligned (16)));
602 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
604 EXPORT_SYMBOL(xtime);
606 /* Don't completely fail for HZ > 500. */
607 int tickadj = 500/HZ ? : 1; /* microsecs */
611 * phase-lock loop variables
613 /* TIME_ERROR prevents overwriting the CMOS clock */
614 int time_state = TIME_OK; /* clock synchronization status */
615 int time_status = STA_UNSYNC; /* clock status bits */
616 long time_offset; /* time adjustment (us) */
617 long time_constant = 2; /* pll time constant */
618 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
619 long time_precision = 1; /* clock precision (us) */
620 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
621 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
622 static long time_phase; /* phase offset (scaled us) */
623 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
624 /* frequency offset (scaled ppm)*/
625 static long time_adj; /* tick adjust (scaled 1 / HZ) */
626 long time_reftime; /* time at last adjustment (s) */
628 long time_next_adjust;
631 * this routine handles the overflow of the microsecond field
633 * The tricky bits of code to handle the accurate clock support
634 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
635 * They were originally developed for SUN and DEC kernels.
636 * All the kudos should go to Dave for this stuff.
639 static void second_overflow(void)
643 /* Bump the maxerror field */
644 time_maxerror += time_tolerance >> SHIFT_USEC;
645 if (time_maxerror > NTP_PHASE_LIMIT) {
646 time_maxerror = NTP_PHASE_LIMIT;
647 time_status |= STA_UNSYNC;
651 * Leap second processing. If in leap-insert state at the end of the
652 * day, the system clock is set back one second; if in leap-delete
653 * state, the system clock is set ahead one second. The microtime()
654 * routine or external clock driver will insure that reported time is
655 * always monotonic. The ugly divides should be replaced.
657 switch (time_state) {
659 if (time_status & STA_INS)
660 time_state = TIME_INS;
661 else if (time_status & STA_DEL)
662 time_state = TIME_DEL;
665 if (xtime.tv_sec % 86400 == 0) {
667 wall_to_monotonic.tv_sec++;
669 * The timer interpolator will make time change
670 * gradually instead of an immediate jump by one second
672 time_interpolator_update(-NSEC_PER_SEC);
673 time_state = TIME_OOP;
675 printk(KERN_NOTICE "Clock: inserting leap second "
680 if ((xtime.tv_sec + 1) % 86400 == 0) {
682 wall_to_monotonic.tv_sec--;
684 * Use of time interpolator for a gradual change of
687 time_interpolator_update(NSEC_PER_SEC);
688 time_state = TIME_WAIT;
690 printk(KERN_NOTICE "Clock: deleting leap second "
695 time_state = TIME_WAIT;
698 if (!(time_status & (STA_INS | STA_DEL)))
699 time_state = TIME_OK;
703 * Compute the phase adjustment for the next second. In PLL mode, the
704 * offset is reduced by a fixed factor times the time constant. In FLL
705 * mode the offset is used directly. In either mode, the maximum phase
706 * adjustment for each second is clamped so as to spread the adjustment
707 * over not more than the number of seconds between updates.
710 if (!(time_status & STA_FLL))
711 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
712 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
713 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
714 time_offset -= ltemp;
715 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
718 * Compute the frequency estimate and additional phase adjustment due
719 * to frequency error for the next second.
722 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
726 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
727 * get 128.125; => only 0.125% error (p. 14)
729 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
733 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
734 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
736 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
740 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
741 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
743 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
748 * Returns how many microseconds we need to add to xtime this tick
749 * in doing an adjustment requested with adjtime.
751 static long adjtime_adjustment(void)
753 long time_adjust_step;
755 time_adjust_step = time_adjust;
756 if (time_adjust_step) {
758 * We are doing an adjtime thing. Prepare time_adjust_step to
759 * be within bounds. Note that a positive time_adjust means we
760 * want the clock to run faster.
762 * Limit the amount of the step to be in the range
763 * -tickadj .. +tickadj
765 time_adjust_step = min(time_adjust_step, (long)tickadj);
766 time_adjust_step = max(time_adjust_step, (long)-tickadj);
768 return time_adjust_step;
771 /* in the NTP reference this is called "hardclock()" */
772 static void update_wall_time_one_tick(void)
774 long time_adjust_step, delta_nsec;
776 time_adjust_step = adjtime_adjustment();
777 if (time_adjust_step)
778 /* Reduce by this step the amount of time left */
779 time_adjust -= time_adjust_step;
780 delta_nsec = tick_nsec + time_adjust_step * 1000;
782 * Advance the phase, once it gets to one microsecond, then
783 * advance the tick more.
785 time_phase += time_adj;
786 if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) {
787 long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10));
788 time_phase -= ltemp << (SHIFT_SCALE - 10);
791 xtime.tv_nsec += delta_nsec;
792 time_interpolator_update(delta_nsec);
794 /* Changes by adjtime() do not take effect till next tick. */
795 if (time_next_adjust != 0) {
796 time_adjust = time_next_adjust;
797 time_next_adjust = 0;
802 * Return how long ticks are at the moment, that is, how much time
803 * update_wall_time_one_tick will add to xtime next time we call it
804 * (assuming no calls to do_adjtimex in the meantime).
805 * The return value is in fixed-point nanoseconds with SHIFT_SCALE-10
806 * bits to the right of the binary point.
807 * This function has no side-effects.
809 u64 current_tick_length(void)
813 delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
814 return ((u64) delta_nsec << (SHIFT_SCALE - 10)) + time_adj;
818 * Using a loop looks inefficient, but "ticks" is
819 * usually just one (we shouldn't be losing ticks,
820 * we're doing this this way mainly for interrupt
821 * latency reasons, not because we think we'll
822 * have lots of lost timer ticks
824 static void update_wall_time(unsigned long ticks)
828 update_wall_time_one_tick();
829 if (xtime.tv_nsec >= 1000000000) {
830 xtime.tv_nsec -= 1000000000;
838 * Called from the timer interrupt handler to charge one tick to the current
839 * process. user_tick is 1 if the tick is user time, 0 for system.
841 void update_process_times(int user_tick)
843 struct task_struct *p = current;
844 int cpu = smp_processor_id();
846 /* Note: this timer irq context must be accounted for as well. */
848 account_user_time(p, jiffies_to_cputime(1));
850 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
852 if (rcu_pending(cpu))
853 rcu_check_callbacks(cpu, user_tick);
855 run_posix_cpu_timers(p);
859 * Nr of active tasks - counted in fixed-point numbers
861 static unsigned long count_active_tasks(void)
863 return nr_active() * FIXED_1;
867 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
868 * imply that avenrun[] is the standard name for this kind of thing.
869 * Nothing else seems to be standardized: the fractional size etc
870 * all seem to differ on different machines.
872 * Requires xtime_lock to access.
874 unsigned long avenrun[3];
876 EXPORT_SYMBOL(avenrun);
879 * calc_load - given tick count, update the avenrun load estimates.
880 * This is called while holding a write_lock on xtime_lock.
882 static inline void calc_load(unsigned long ticks)
884 unsigned long active_tasks; /* fixed-point */
885 static int count = LOAD_FREQ;
890 active_tasks = count_active_tasks();
891 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
892 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
893 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
897 /* jiffies at the most recent update of wall time */
898 unsigned long wall_jiffies = INITIAL_JIFFIES;
901 * This read-write spinlock protects us from races in SMP while
902 * playing with xtime and avenrun.
904 #ifndef ARCH_HAVE_XTIME_LOCK
905 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
907 EXPORT_SYMBOL(xtime_lock);
911 * This function runs timers and the timer-tq in bottom half context.
913 static void run_timer_softirq(struct softirq_action *h)
915 tvec_base_t *base = __get_cpu_var(tvec_bases);
917 hrtimer_run_queues();
918 if (time_after_eq(jiffies, base->timer_jiffies))
923 * Called by the local, per-CPU timer interrupt on SMP.
925 void run_local_timers(void)
927 raise_softirq(TIMER_SOFTIRQ);
932 * Called by the timer interrupt. xtime_lock must already be taken
935 static inline void update_times(void)
939 ticks = jiffies - wall_jiffies;
941 wall_jiffies += ticks;
942 update_wall_time(ticks);
948 * The 64-bit jiffies value is not atomic - you MUST NOT read it
949 * without sampling the sequence number in xtime_lock.
950 * jiffies is defined in the linker script...
953 void do_timer(struct pt_regs *regs)
956 /* prevent loading jiffies before storing new jiffies_64 value. */
961 #ifdef __ARCH_WANT_SYS_ALARM
964 * For backwards compatibility? This can be done in libc so Alpha
965 * and all newer ports shouldn't need it.
967 asmlinkage unsigned long sys_alarm(unsigned int seconds)
969 return alarm_setitimer(seconds);
976 * sys_getpid - return the thread group id of the current process
978 * Note, despite the name, this returns the tgid not the pid. The tgid and
979 * the pid are identical unless CLONE_THREAD was specified on clone() in
980 * which case the tgid is the same in all threads of the same group.
982 * This is SMP safe as current->tgid does not change.
984 asmlinkage long sys_getpid(void)
986 return vx_map_tgid(current->tgid);
990 * Accessing ->real_parent is not SMP-safe, it could
991 * change from under us. However, we can use a stale
992 * value of ->real_parent under rcu_read_lock(), see
993 * release_task()->call_rcu(delayed_put_task_struct).
995 asmlinkage long sys_getppid(void)
1000 pid = rcu_dereference(current->real_parent)->tgid;
1002 return vx_map_pid(pid);
1008 * The Alpha uses getxpid, getxuid, and getxgid instead.
1011 asmlinkage long do_getxpid(long *ppid)
1013 *ppid = sys_getppid();
1014 return sys_getpid();
1019 asmlinkage long sys_getuid(void)
1021 /* Only we change this so SMP safe */
1022 return current->uid;
1025 asmlinkage long sys_geteuid(void)
1027 /* Only we change this so SMP safe */
1028 return current->euid;
1031 asmlinkage long sys_getgid(void)
1033 /* Only we change this so SMP safe */
1034 return current->gid;
1037 asmlinkage long sys_getegid(void)
1039 /* Only we change this so SMP safe */
1040 return current->egid;
1045 static void process_timeout(unsigned long __data)
1047 wake_up_process((task_t *)__data);
1051 * schedule_timeout - sleep until timeout
1052 * @timeout: timeout value in jiffies
1054 * Make the current task sleep until @timeout jiffies have
1055 * elapsed. The routine will return immediately unless
1056 * the current task state has been set (see set_current_state()).
1058 * You can set the task state as follows -
1060 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1061 * pass before the routine returns. The routine will return 0
1063 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1064 * delivered to the current task. In this case the remaining time
1065 * in jiffies will be returned, or 0 if the timer expired in time
1067 * The current task state is guaranteed to be TASK_RUNNING when this
1070 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1071 * the CPU away without a bound on the timeout. In this case the return
1072 * value will be %MAX_SCHEDULE_TIMEOUT.
1074 * In all cases the return value is guaranteed to be non-negative.
1076 fastcall signed long __sched schedule_timeout(signed long timeout)
1078 struct timer_list timer;
1079 unsigned long expire;
1083 case MAX_SCHEDULE_TIMEOUT:
1085 * These two special cases are useful to be comfortable
1086 * in the caller. Nothing more. We could take
1087 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1088 * but I' d like to return a valid offset (>=0) to allow
1089 * the caller to do everything it want with the retval.
1095 * Another bit of PARANOID. Note that the retval will be
1096 * 0 since no piece of kernel is supposed to do a check
1097 * for a negative retval of schedule_timeout() (since it
1098 * should never happens anyway). You just have the printk()
1099 * that will tell you if something is gone wrong and where.
1103 printk(KERN_ERR "schedule_timeout: wrong timeout "
1104 "value %lx from %p\n", timeout,
1105 __builtin_return_address(0));
1106 current->state = TASK_RUNNING;
1111 expire = timeout + jiffies;
1113 setup_timer(&timer, process_timeout, (unsigned long)current);
1114 __mod_timer(&timer, expire);
1116 del_singleshot_timer_sync(&timer);
1118 timeout = expire - jiffies;
1121 return timeout < 0 ? 0 : timeout;
1123 EXPORT_SYMBOL(schedule_timeout);
1126 * We can use __set_current_state() here because schedule_timeout() calls
1127 * schedule() unconditionally.
1129 signed long __sched schedule_timeout_interruptible(signed long timeout)
1131 __set_current_state(TASK_INTERRUPTIBLE);
1132 return schedule_timeout(timeout);
1134 EXPORT_SYMBOL(schedule_timeout_interruptible);
1136 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1138 __set_current_state(TASK_UNINTERRUPTIBLE);
1139 return schedule_timeout(timeout);
1141 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1143 /* Thread ID - the internal kernel "pid" */
1144 asmlinkage long sys_gettid(void)
1146 return current->pid;
1150 * sys_sysinfo - fill in sysinfo struct
1152 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1155 unsigned long mem_total, sav_total;
1156 unsigned int mem_unit, bitcount;
1159 memset((char *)&val, 0, sizeof(struct sysinfo));
1163 seq = read_seqbegin(&xtime_lock);
1166 * This is annoying. The below is the same thing
1167 * posix_get_clock_monotonic() does, but it wants to
1168 * take the lock which we want to cover the loads stuff
1172 getnstimeofday(&tp);
1173 tp.tv_sec += wall_to_monotonic.tv_sec;
1174 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1175 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1176 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1179 if (vx_flags(VXF_VIRT_UPTIME, 0))
1180 vx_vsi_uptime(&tp, NULL);
1181 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1183 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1184 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1185 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1187 val.procs = nr_threads;
1188 } while (read_seqretry(&xtime_lock, seq));
1194 * If the sum of all the available memory (i.e. ram + swap)
1195 * is less than can be stored in a 32 bit unsigned long then
1196 * we can be binary compatible with 2.2.x kernels. If not,
1197 * well, in that case 2.2.x was broken anyways...
1199 * -Erik Andersen <andersee@debian.org>
1202 mem_total = val.totalram + val.totalswap;
1203 if (mem_total < val.totalram || mem_total < val.totalswap)
1206 mem_unit = val.mem_unit;
1207 while (mem_unit > 1) {
1210 sav_total = mem_total;
1212 if (mem_total < sav_total)
1217 * If mem_total did not overflow, multiply all memory values by
1218 * val.mem_unit and set it to 1. This leaves things compatible
1219 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1224 val.totalram <<= bitcount;
1225 val.freeram <<= bitcount;
1226 val.sharedram <<= bitcount;
1227 val.bufferram <<= bitcount;
1228 val.totalswap <<= bitcount;
1229 val.freeswap <<= bitcount;
1230 val.totalhigh <<= bitcount;
1231 val.freehigh <<= bitcount;
1234 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1240 static int __devinit init_timers_cpu(int cpu)
1244 static char __devinitdata tvec_base_done[NR_CPUS];
1246 if (!tvec_base_done[cpu]) {
1247 static char boot_done;
1251 * The APs use this path later in boot
1253 base = kmalloc_node(sizeof(*base), GFP_KERNEL,
1257 memset(base, 0, sizeof(*base));
1258 per_cpu(tvec_bases, cpu) = base;
1261 * This is for the boot CPU - we use compile-time
1262 * static initialisation because per-cpu memory isn't
1263 * ready yet and because the memory allocators are not
1264 * initialised either.
1267 base = &boot_tvec_bases;
1269 tvec_base_done[cpu] = 1;
1271 base = per_cpu(tvec_bases, cpu);
1274 spin_lock_init(&base->lock);
1275 for (j = 0; j < TVN_SIZE; j++) {
1276 INIT_LIST_HEAD(base->tv5.vec + j);
1277 INIT_LIST_HEAD(base->tv4.vec + j);
1278 INIT_LIST_HEAD(base->tv3.vec + j);
1279 INIT_LIST_HEAD(base->tv2.vec + j);
1281 for (j = 0; j < TVR_SIZE; j++)
1282 INIT_LIST_HEAD(base->tv1.vec + j);
1284 base->timer_jiffies = jiffies;
1288 #ifdef CONFIG_HOTPLUG_CPU
1289 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1291 struct timer_list *timer;
1293 while (!list_empty(head)) {
1294 timer = list_entry(head->next, struct timer_list, entry);
1295 detach_timer(timer, 0);
1296 timer->base = new_base;
1297 internal_add_timer(new_base, timer);
1301 static void __devinit migrate_timers(int cpu)
1303 tvec_base_t *old_base;
1304 tvec_base_t *new_base;
1307 BUG_ON(cpu_online(cpu));
1308 old_base = per_cpu(tvec_bases, cpu);
1309 new_base = get_cpu_var(tvec_bases);
1311 local_irq_disable();
1312 spin_lock(&new_base->lock);
1313 spin_lock(&old_base->lock);
1315 BUG_ON(old_base->running_timer);
1317 for (i = 0; i < TVR_SIZE; i++)
1318 migrate_timer_list(new_base, old_base->tv1.vec + i);
1319 for (i = 0; i < TVN_SIZE; i++) {
1320 migrate_timer_list(new_base, old_base->tv2.vec + i);
1321 migrate_timer_list(new_base, old_base->tv3.vec + i);
1322 migrate_timer_list(new_base, old_base->tv4.vec + i);
1323 migrate_timer_list(new_base, old_base->tv5.vec + i);
1326 spin_unlock(&old_base->lock);
1327 spin_unlock(&new_base->lock);
1329 put_cpu_var(tvec_bases);
1331 #endif /* CONFIG_HOTPLUG_CPU */
1333 static int timer_cpu_notify(struct notifier_block *self,
1334 unsigned long action, void *hcpu)
1336 long cpu = (long)hcpu;
1338 case CPU_UP_PREPARE:
1339 if (init_timers_cpu(cpu) < 0)
1342 #ifdef CONFIG_HOTPLUG_CPU
1344 migrate_timers(cpu);
1353 static struct notifier_block timers_nb = {
1354 .notifier_call = timer_cpu_notify,
1358 void __init init_timers(void)
1360 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1361 (void *)(long)smp_processor_id());
1362 register_cpu_notifier(&timers_nb);
1363 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1366 #ifdef CONFIG_TIME_INTERPOLATION
1368 struct time_interpolator *time_interpolator __read_mostly;
1369 static struct time_interpolator *time_interpolator_list __read_mostly;
1370 static DEFINE_SPINLOCK(time_interpolator_lock);
1372 static inline u64 time_interpolator_get_cycles(unsigned int src)
1374 unsigned long (*x)(void);
1378 case TIME_SOURCE_FUNCTION:
1379 x = time_interpolator->addr;
1382 case TIME_SOURCE_MMIO64 :
1383 return readq_relaxed((void __iomem *)time_interpolator->addr);
1385 case TIME_SOURCE_MMIO32 :
1386 return readl_relaxed((void __iomem *)time_interpolator->addr);
1388 default: return get_cycles();
1392 static inline u64 time_interpolator_get_counter(int writelock)
1394 unsigned int src = time_interpolator->source;
1396 if (time_interpolator->jitter)
1402 lcycle = time_interpolator->last_cycle;
1403 now = time_interpolator_get_cycles(src);
1404 if (lcycle && time_after(lcycle, now))
1407 /* When holding the xtime write lock, there's no need
1408 * to add the overhead of the cmpxchg. Readers are
1409 * force to retry until the write lock is released.
1412 time_interpolator->last_cycle = now;
1415 /* Keep track of the last timer value returned. The use of cmpxchg here
1416 * will cause contention in an SMP environment.
1418 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1422 return time_interpolator_get_cycles(src);
1425 void time_interpolator_reset(void)
1427 time_interpolator->offset = 0;
1428 time_interpolator->last_counter = time_interpolator_get_counter(1);
1431 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1433 unsigned long time_interpolator_get_offset(void)
1435 /* If we do not have a time interpolator set up then just return zero */
1436 if (!time_interpolator)
1439 return time_interpolator->offset +
1440 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1443 #define INTERPOLATOR_ADJUST 65536
1444 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1446 static void time_interpolator_update(long delta_nsec)
1449 unsigned long offset;
1451 /* If there is no time interpolator set up then do nothing */
1452 if (!time_interpolator)
1456 * The interpolator compensates for late ticks by accumulating the late
1457 * time in time_interpolator->offset. A tick earlier than expected will
1458 * lead to a reset of the offset and a corresponding jump of the clock
1459 * forward. Again this only works if the interpolator clock is running
1460 * slightly slower than the regular clock and the tuning logic insures
1464 counter = time_interpolator_get_counter(1);
1465 offset = time_interpolator->offset +
1466 GET_TI_NSECS(counter, time_interpolator);
1468 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1469 time_interpolator->offset = offset - delta_nsec;
1471 time_interpolator->skips++;
1472 time_interpolator->ns_skipped += delta_nsec - offset;
1473 time_interpolator->offset = 0;
1475 time_interpolator->last_counter = counter;
1477 /* Tuning logic for time interpolator invoked every minute or so.
1478 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1479 * Increase interpolator clock speed if we skip too much time.
1481 if (jiffies % INTERPOLATOR_ADJUST == 0)
1483 if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
1484 time_interpolator->nsec_per_cyc--;
1485 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1486 time_interpolator->nsec_per_cyc++;
1487 time_interpolator->skips = 0;
1488 time_interpolator->ns_skipped = 0;
1493 is_better_time_interpolator(struct time_interpolator *new)
1495 if (!time_interpolator)
1497 return new->frequency > 2*time_interpolator->frequency ||
1498 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1502 register_time_interpolator(struct time_interpolator *ti)
1504 unsigned long flags;
1507 BUG_ON(ti->frequency == 0 || ti->mask == 0);
1509 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1510 spin_lock(&time_interpolator_lock);
1511 write_seqlock_irqsave(&xtime_lock, flags);
1512 if (is_better_time_interpolator(ti)) {
1513 time_interpolator = ti;
1514 time_interpolator_reset();
1516 write_sequnlock_irqrestore(&xtime_lock, flags);
1518 ti->next = time_interpolator_list;
1519 time_interpolator_list = ti;
1520 spin_unlock(&time_interpolator_lock);
1524 unregister_time_interpolator(struct time_interpolator *ti)
1526 struct time_interpolator *curr, **prev;
1527 unsigned long flags;
1529 spin_lock(&time_interpolator_lock);
1530 prev = &time_interpolator_list;
1531 for (curr = *prev; curr; curr = curr->next) {
1539 write_seqlock_irqsave(&xtime_lock, flags);
1540 if (ti == time_interpolator) {
1541 /* we lost the best time-interpolator: */
1542 time_interpolator = NULL;
1543 /* find the next-best interpolator */
1544 for (curr = time_interpolator_list; curr; curr = curr->next)
1545 if (is_better_time_interpolator(curr))
1546 time_interpolator = curr;
1547 time_interpolator_reset();
1549 write_sequnlock_irqrestore(&xtime_lock, flags);
1550 spin_unlock(&time_interpolator_lock);
1552 #endif /* CONFIG_TIME_INTERPOLATION */
1555 * msleep - sleep safely even with waitqueue interruptions
1556 * @msecs: Time in milliseconds to sleep for
1558 void msleep(unsigned int msecs)
1560 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1563 timeout = schedule_timeout_uninterruptible(timeout);
1566 EXPORT_SYMBOL(msleep);
1569 * msleep_interruptible - sleep waiting for signals
1570 * @msecs: Time in milliseconds to sleep for
1572 unsigned long msleep_interruptible(unsigned int msecs)
1574 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1576 while (timeout && !signal_pending(current))
1577 timeout = schedule_timeout_interruptible(timeout);
1578 return jiffies_to_msecs(timeout);
1581 EXPORT_SYMBOL(msleep_interruptible);