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/diskdump.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)
54 * per-CPU timer vector definitions:
57 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
58 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
59 #define TVN_SIZE (1 << TVN_BITS)
60 #define TVR_SIZE (1 << TVR_BITS)
61 #define TVN_MASK (TVN_SIZE - 1)
62 #define TVR_MASK (TVR_SIZE - 1)
64 typedef struct tvec_s {
65 struct list_head vec[TVN_SIZE];
68 typedef struct tvec_root_s {
69 struct list_head vec[TVR_SIZE];
72 struct tvec_t_base_s {
74 unsigned long timer_jiffies;
75 struct timer_list *running_timer;
81 } ____cacheline_aligned_in_smp;
83 typedef struct tvec_t_base_s tvec_base_t;
85 static inline void set_running_timer(tvec_base_t *base,
86 struct timer_list *timer)
89 base->running_timer = timer;
93 /* Fake initialization */
94 static DEFINE_PER_CPU(tvec_base_t, tvec_bases) = { SPIN_LOCK_UNLOCKED };
96 static void check_timer_failed(struct timer_list *timer)
98 static int whine_count;
99 if (whine_count < 16) {
101 printk("Uninitialised timer!\n");
102 printk("This is just a warning. Your computer is OK\n");
103 printk("function=0x%p, data=0x%lx\n",
104 timer->function, timer->data);
110 spin_lock_init(&timer->lock);
111 timer->magic = TIMER_MAGIC;
114 static inline void check_timer(struct timer_list *timer)
116 if (timer->magic != TIMER_MAGIC)
117 check_timer_failed(timer);
121 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
123 unsigned long expires = timer->expires;
124 unsigned long idx = expires - base->timer_jiffies;
125 struct list_head *vec;
127 if (idx < TVR_SIZE) {
128 int i = expires & TVR_MASK;
129 vec = base->tv1.vec + i;
130 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
131 int i = (expires >> TVR_BITS) & TVN_MASK;
132 vec = base->tv2.vec + i;
133 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
134 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
135 vec = base->tv3.vec + i;
136 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
137 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
138 vec = base->tv4.vec + i;
139 } else if ((signed long) idx < 0) {
141 * Can happen if you add a timer with expires == jiffies,
142 * or you set a timer to go off in the past
144 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
147 /* If the timeout is larger than 0xffffffff on 64-bit
148 * architectures then we use the maximum timeout:
150 if (idx > 0xffffffffUL) {
152 expires = idx + base->timer_jiffies;
154 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
155 vec = base->tv5.vec + i;
160 list_add_tail(&timer->entry, vec);
163 int __mod_timer(struct timer_list *timer, unsigned long expires)
165 tvec_base_t *old_base, *new_base;
169 BUG_ON(!timer->function);
173 spin_lock_irqsave(&timer->lock, flags);
174 new_base = &__get_cpu_var(tvec_bases);
176 old_base = timer->base;
179 * Prevent deadlocks via ordering by old_base < new_base.
181 if (old_base && (new_base != old_base)) {
182 if (old_base < new_base) {
183 spin_lock(&new_base->lock);
184 spin_lock(&old_base->lock);
186 spin_lock(&old_base->lock);
187 spin_lock(&new_base->lock);
190 * The timer base might have been cancelled while we were
191 * trying to take the lock(s):
193 if (timer->base != old_base) {
194 spin_unlock(&new_base->lock);
195 spin_unlock(&old_base->lock);
199 spin_lock(&new_base->lock);
200 if (timer->base != old_base) {
201 spin_unlock(&new_base->lock);
207 * Delete the previous timeout (if there was any), and install
211 list_del(&timer->entry);
214 timer->expires = expires;
215 internal_add_timer(new_base, timer);
216 timer->base = new_base;
218 if (old_base && (new_base != old_base))
219 spin_unlock(&old_base->lock);
220 spin_unlock(&new_base->lock);
221 spin_unlock_irqrestore(&timer->lock, flags);
226 EXPORT_SYMBOL(__mod_timer);
229 * add_timer_on - start a timer on a particular CPU
230 * @timer: the timer to be added
231 * @cpu: the CPU to start it on
233 * This is not very scalable on SMP. Double adds are not possible.
235 void add_timer_on(struct timer_list *timer, int cpu)
237 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
240 BUG_ON(timer_pending(timer) || !timer->function);
244 spin_lock_irqsave(&base->lock, flags);
245 internal_add_timer(base, timer);
247 spin_unlock_irqrestore(&base->lock, flags);
252 * mod_timer - modify a timer's timeout
253 * @timer: the timer to be modified
255 * mod_timer is a more efficient way to update the expire field of an
256 * active timer (if the timer is inactive it will be activated)
258 * mod_timer(timer, expires) is equivalent to:
260 * del_timer(timer); timer->expires = expires; add_timer(timer);
262 * Note that if there are multiple unserialized concurrent users of the
263 * same timer, then mod_timer() is the only safe way to modify the timeout,
264 * since add_timer() cannot modify an already running timer.
266 * The function returns whether it has modified a pending timer or not.
267 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
268 * active timer returns 1.)
270 int mod_timer(struct timer_list *timer, unsigned long expires)
272 BUG_ON(!timer->function);
277 * This is a common optimization triggered by the
278 * networking code - if the timer is re-modified
279 * to be the same thing then just return:
281 if (timer->expires == expires && timer_pending(timer))
284 return __mod_timer(timer, expires);
287 EXPORT_SYMBOL(mod_timer);
290 * del_timer - deactive a timer.
291 * @timer: the timer to be deactivated
293 * del_timer() deactivates a timer - this works on both active and inactive
296 * The function returns whether it has deactivated a pending timer or not.
297 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
298 * active timer returns 1.)
300 int del_timer(struct timer_list *timer)
311 spin_lock_irqsave(&base->lock, flags);
312 if (base != timer->base) {
313 spin_unlock_irqrestore(&base->lock, flags);
316 list_del(&timer->entry);
317 /* Need to make sure that anybody who sees a NULL base also sees the list ops */
320 spin_unlock_irqrestore(&base->lock, flags);
325 EXPORT_SYMBOL(del_timer);
329 * del_timer_sync - deactivate a timer and wait for the handler to finish.
330 * @timer: the timer to be deactivated
332 * This function only differs from del_timer() on SMP: besides deactivating
333 * the timer it also makes sure the handler has finished executing on other
336 * Synchronization rules: callers must prevent restarting of the timer,
337 * otherwise this function is meaningless. It must not be called from
338 * interrupt contexts. The caller must not hold locks which would prevent
339 * completion of the timer's handler. Upon exit the timer is not queued and
340 * the handler is not running on any CPU.
342 * The function returns whether it has deactivated a pending timer or not.
344 * del_timer_sync() is slow and complicated because it copes with timer
345 * handlers which re-arm the timer (periodic timers). If the timer handler
346 * is known to not do this (a single shot timer) then use
347 * del_singleshot_timer_sync() instead.
349 int del_timer_sync(struct timer_list *timer)
357 ret += del_timer(timer);
359 for_each_online_cpu(i) {
360 base = &per_cpu(tvec_bases, i);
361 if (base->running_timer == timer) {
362 while (base->running_timer == timer) {
364 preempt_check_resched();
370 if (timer_pending(timer))
375 EXPORT_SYMBOL(del_timer_sync);
378 * del_singleshot_timer_sync - deactivate a non-recursive timer
379 * @timer: the timer to be deactivated
381 * This function is an optimization of del_timer_sync for the case where the
382 * caller can guarantee the timer does not reschedule itself in its timer
385 * Synchronization rules: callers must prevent restarting of the timer,
386 * otherwise this function is meaningless. It must not be called from
387 * interrupt contexts. The caller must not hold locks which wold prevent
388 * completion of the timer's handler. Upon exit the timer is not queued and
389 * the handler is not running on any CPU.
391 * The function returns whether it has deactivated a pending timer or not.
393 int del_singleshot_timer_sync(struct timer_list *timer)
395 int ret = del_timer(timer);
398 ret = del_timer_sync(timer);
404 EXPORT_SYMBOL(del_singleshot_timer_sync);
407 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
409 /* cascade all the timers from tv up one level */
410 struct list_head *head, *curr;
412 head = tv->vec + index;
415 * We are removing _all_ timers from the list, so we don't have to
416 * detach them individually, just clear the list afterwards.
418 while (curr != head) {
419 struct timer_list *tmp;
421 tmp = list_entry(curr, struct timer_list, entry);
422 BUG_ON(tmp->base != base);
424 internal_add_timer(base, tmp);
426 INIT_LIST_HEAD(head);
432 * __run_timers - run all expired timers (if any) on this CPU.
433 * @base: the timer vector to be processed.
435 * This function cascades all vectors and executes all expired timer
438 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
440 static inline void __run_timers(tvec_base_t *base)
442 struct timer_list *timer;
445 spin_lock_irqsave(&base->lock, flags);
446 while (time_after_eq(jiffies, base->timer_jiffies)) {
447 struct list_head work_list = LIST_HEAD_INIT(work_list);
448 struct list_head *head = &work_list;
449 int index = base->timer_jiffies & TVR_MASK;
455 (!cascade(base, &base->tv2, INDEX(0))) &&
456 (!cascade(base, &base->tv3, INDEX(1))) &&
457 !cascade(base, &base->tv4, INDEX(2)))
458 cascade(base, &base->tv5, INDEX(3));
459 ++base->timer_jiffies;
460 list_splice_init(base->tv1.vec + index, &work_list);
462 if (!list_empty(head)) {
463 void (*fn)(unsigned long);
466 timer = list_entry(head->next,struct timer_list,entry);
467 fn = timer->function;
470 list_del(&timer->entry);
471 set_running_timer(base, timer);
474 spin_unlock_irqrestore(&base->lock, flags);
476 u32 preempt_count = preempt_count();
478 if (preempt_count != preempt_count()) {
479 printk("huh, entered %p with %08x, exited with %08x?\n", fn, preempt_count, preempt_count());
483 spin_lock_irq(&base->lock);
487 set_running_timer(base, NULL);
488 spin_unlock_irqrestore(&base->lock, flags);
491 #ifdef CONFIG_NO_IDLE_HZ
493 * Find out when the next timer event is due to happen. This
494 * is used on S/390 to stop all activity when a cpus is idle.
495 * This functions needs to be called disabled.
497 unsigned long next_timer_interrupt(void)
500 struct list_head *list;
501 struct timer_list *nte;
502 unsigned long expires;
506 base = &__get_cpu_var(tvec_bases);
507 spin_lock(&base->lock);
508 expires = base->timer_jiffies + (LONG_MAX >> 1);
511 /* Look for timer events in tv1. */
512 j = base->timer_jiffies & TVR_MASK;
514 list_for_each_entry(nte, base->tv1.vec + j, entry) {
515 expires = nte->expires;
516 if (j < (base->timer_jiffies & TVR_MASK))
517 list = base->tv2.vec + (INDEX(0));
520 j = (j + 1) & TVR_MASK;
521 } while (j != (base->timer_jiffies & TVR_MASK));
524 varray[0] = &base->tv2;
525 varray[1] = &base->tv3;
526 varray[2] = &base->tv4;
527 varray[3] = &base->tv5;
528 for (i = 0; i < 4; i++) {
531 if (list_empty(varray[i]->vec + j)) {
532 j = (j + 1) & TVN_MASK;
535 list_for_each_entry(nte, varray[i]->vec + j, entry)
536 if (time_before(nte->expires, expires))
537 expires = nte->expires;
538 if (j < (INDEX(i)) && i < 3)
539 list = varray[i + 1]->vec + (INDEX(i + 1));
541 } while (j != (INDEX(i)));
546 * The search wrapped. We need to look at the next list
547 * from next tv element that would cascade into tv element
548 * where we found the timer element.
550 list_for_each_entry(nte, list, entry) {
551 if (time_before(nte->expires, expires))
552 expires = nte->expires;
555 spin_unlock(&base->lock);
560 /******************************************************************/
563 * Timekeeping variables
565 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
566 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
570 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
571 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
572 * at zero at system boot time, so wall_to_monotonic will be negative,
573 * however, we will ALWAYS keep the tv_nsec part positive so we can use
574 * the usual normalization.
576 struct timespec xtime __attribute__ ((aligned (16)));
577 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
579 EXPORT_SYMBOL(xtime);
581 /* Don't completely fail for HZ > 500. */
582 int tickadj = 500/HZ ? : 1; /* microsecs */
586 * phase-lock loop variables
588 /* TIME_ERROR prevents overwriting the CMOS clock */
589 int time_state = TIME_OK; /* clock synchronization status */
590 int time_status = STA_UNSYNC; /* clock status bits */
591 long time_offset; /* time adjustment (us) */
592 long time_constant = 2; /* pll time constant */
593 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
594 long time_precision = 1; /* clock precision (us) */
595 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
596 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
597 static long time_phase; /* phase offset (scaled us) */
598 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
599 /* frequency offset (scaled ppm)*/
600 static long time_adj; /* tick adjust (scaled 1 / HZ) */
601 long time_reftime; /* time at last adjustment (s) */
603 long time_next_adjust;
606 * this routine handles the overflow of the microsecond field
608 * The tricky bits of code to handle the accurate clock support
609 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
610 * They were originally developed for SUN and DEC kernels.
611 * All the kudos should go to Dave for this stuff.
614 static void second_overflow(void)
618 /* Bump the maxerror field */
619 time_maxerror += time_tolerance >> SHIFT_USEC;
620 if ( time_maxerror > NTP_PHASE_LIMIT ) {
621 time_maxerror = NTP_PHASE_LIMIT;
622 time_status |= STA_UNSYNC;
626 * Leap second processing. If in leap-insert state at
627 * the end of the day, the system clock is set back one
628 * second; if in leap-delete state, the system clock is
629 * set ahead one second. The microtime() routine or
630 * external clock driver will insure that reported time
631 * is always monotonic. The ugly divides should be
634 switch (time_state) {
637 if (time_status & STA_INS)
638 time_state = TIME_INS;
639 else if (time_status & STA_DEL)
640 time_state = TIME_DEL;
644 if (xtime.tv_sec % 86400 == 0) {
646 wall_to_monotonic.tv_sec++;
647 /* The timer interpolator will make time change gradually instead
648 * of an immediate jump by one second.
650 time_interpolator_update(-NSEC_PER_SEC);
651 time_state = TIME_OOP;
653 printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
658 if ((xtime.tv_sec + 1) % 86400 == 0) {
660 wall_to_monotonic.tv_sec--;
661 /* Use of time interpolator for a gradual change of time */
662 time_interpolator_update(NSEC_PER_SEC);
663 time_state = TIME_WAIT;
665 printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
670 time_state = TIME_WAIT;
674 if (!(time_status & (STA_INS | STA_DEL)))
675 time_state = TIME_OK;
679 * Compute the phase adjustment for the next second. In
680 * PLL mode, the offset is reduced by a fixed factor
681 * times the time constant. In FLL mode the offset is
682 * used directly. In either mode, the maximum phase
683 * adjustment for each second is clamped so as to spread
684 * the adjustment over not more than the number of
685 * seconds between updates.
687 if (time_offset < 0) {
688 ltemp = -time_offset;
689 if (!(time_status & STA_FLL))
690 ltemp >>= SHIFT_KG + time_constant;
691 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
692 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
693 time_offset += ltemp;
694 #if SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE > 0
695 time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
697 time_adj = -ltemp >> (SHIFT_HZ + SHIFT_UPDATE - SHIFT_SCALE);
701 if (!(time_status & STA_FLL))
702 ltemp >>= SHIFT_KG + time_constant;
703 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
704 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
705 time_offset -= ltemp;
706 #if SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE > 0
707 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
709 time_adj = ltemp >> (SHIFT_HZ + SHIFT_UPDATE - SHIFT_SCALE);
714 * Compute the frequency estimate and additional phase
715 * adjustment due to frequency error for the next
716 * second. When the PPS signal is engaged, gnaw on the
717 * watchdog counter and update the frequency computed by
718 * the pll and the PPS signal.
721 if (pps_valid == PPS_VALID) { /* PPS signal lost */
722 pps_jitter = MAXTIME;
723 pps_stabil = MAXFREQ;
724 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
725 STA_PPSWANDER | STA_PPSERROR);
727 ltemp = time_freq + pps_freq;
729 time_adj -= -ltemp >>
730 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
733 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
736 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
737 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
740 time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
742 time_adj += (time_adj >> 2) + (time_adj >> 5);
745 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
746 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
749 time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
751 time_adj += (time_adj >> 6) + (time_adj >> 7);
755 /* in the NTP reference this is called "hardclock()" */
756 static void update_wall_time_one_tick(void)
758 long time_adjust_step, delta_nsec;
760 if ( (time_adjust_step = time_adjust) != 0 ) {
761 /* We are doing an adjtime thing.
763 * Prepare time_adjust_step to be within bounds.
764 * Note that a positive time_adjust means we want the clock
767 * Limit the amount of the step to be in the range
768 * -tickadj .. +tickadj
770 if (time_adjust > tickadj)
771 time_adjust_step = tickadj;
772 else if (time_adjust < -tickadj)
773 time_adjust_step = -tickadj;
775 /* Reduce by this step the amount of time left */
776 time_adjust -= time_adjust_step;
778 delta_nsec = tick_nsec + time_adjust_step * 1000;
780 * Advance the phase, once it gets to one microsecond, then
781 * advance the tick more.
783 time_phase += time_adj;
784 if (time_phase <= -FINENSEC) {
785 long ltemp = -time_phase >> (SHIFT_SCALE - 10);
786 time_phase += ltemp << (SHIFT_SCALE - 10);
789 else if (time_phase >= FINENSEC) {
790 long ltemp = time_phase >> (SHIFT_SCALE - 10);
791 time_phase -= ltemp << (SHIFT_SCALE - 10);
794 xtime.tv_nsec += delta_nsec;
795 time_interpolator_update(delta_nsec);
797 /* Changes by adjtime() do not take effect till next tick. */
798 if (time_next_adjust != 0) {
799 time_adjust = time_next_adjust;
800 time_next_adjust = 0;
805 * Using a loop looks inefficient, but "ticks" is
806 * usually just one (we shouldn't be losing ticks,
807 * we're doing this this way mainly for interrupt
808 * latency reasons, not because we think we'll
809 * have lots of lost timer ticks
811 static void update_wall_time(unsigned long ticks)
815 update_wall_time_one_tick();
816 if (xtime.tv_nsec >= 1000000000) {
817 xtime.tv_nsec -= 1000000000;
825 * Called from the timer interrupt handler to charge one tick to the current
826 * process. user_tick is 1 if the tick is user time, 0 for system.
828 void update_process_times(int user_tick)
830 struct task_struct *p = current;
831 int cpu = smp_processor_id();
833 /* Note: this timer irq context must be accounted for as well. */
835 account_user_time(p, jiffies_to_cputime(1));
837 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
839 if (rcu_pending(cpu))
840 rcu_check_callbacks(cpu, user_tick);
842 run_posix_cpu_timers(p);
846 * Nr of active tasks - counted in fixed-point numbers
848 static unsigned long count_active_tasks(void)
850 return (nr_running() + nr_uninterruptible()) * FIXED_1;
854 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
855 * imply that avenrun[] is the standard name for this kind of thing.
856 * Nothing else seems to be standardized: the fractional size etc
857 * all seem to differ on different machines.
859 * Requires xtime_lock to access.
861 unsigned long avenrun[3];
863 EXPORT_SYMBOL(avenrun);
866 * calc_load - given tick count, update the avenrun load estimates.
867 * This is called while holding a write_lock on xtime_lock.
869 static inline void calc_load(unsigned long ticks)
871 unsigned long active_tasks; /* fixed-point */
872 static int count = LOAD_FREQ;
877 active_tasks = count_active_tasks();
878 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
879 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
880 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
884 /* jiffies at the most recent update of wall time */
885 unsigned long wall_jiffies = INITIAL_JIFFIES;
888 * This read-write spinlock protects us from races in SMP while
889 * playing with xtime and avenrun.
891 #ifndef ARCH_HAVE_XTIME_LOCK
892 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
894 EXPORT_SYMBOL(xtime_lock);
898 * This function runs timers and the timer-tq in bottom half context.
900 static void run_timer_softirq(struct softirq_action *h)
902 tvec_base_t *base = &__get_cpu_var(tvec_bases);
904 if (time_after_eq(jiffies, base->timer_jiffies))
909 * Called by the local, per-CPU timer interrupt on SMP.
911 void run_local_timers(void)
913 raise_softirq(TIMER_SOFTIRQ);
917 * Called by the timer interrupt. xtime_lock must already be taken
920 static inline void update_times(void)
924 ticks = jiffies - wall_jiffies;
926 wall_jiffies += ticks;
927 update_wall_time(ticks);
933 * The 64-bit jiffies value is not atomic - you MUST NOT read it
934 * without sampling the sequence number in xtime_lock.
935 * jiffies is defined in the linker script...
938 void do_timer(struct pt_regs *regs)
944 #ifdef __ARCH_WANT_SYS_ALARM
947 * For backwards compatibility? This can be done in libc so Alpha
948 * and all newer ports shouldn't need it.
950 asmlinkage unsigned long sys_alarm(unsigned int seconds)
952 struct itimerval it_new, it_old;
953 unsigned int oldalarm;
955 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
956 it_new.it_value.tv_sec = seconds;
957 it_new.it_value.tv_usec = 0;
958 do_setitimer(ITIMER_REAL, &it_new, &it_old);
959 oldalarm = it_old.it_value.tv_sec;
960 /* ehhh.. We can't return 0 if we have an alarm pending.. */
961 /* And we'd better return too much than too little anyway */
962 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
971 * sys_getpid - return the thread group id of the current process
973 * Note, despite the name, this returns the tgid not the pid. The tgid and
974 * the pid are identical unless CLONE_THREAD was specified on clone() in
975 * which case the tgid is the same in all threads of the same group.
977 * This is SMP safe as current->tgid does not change.
979 asmlinkage long sys_getpid(void)
981 return vx_map_tgid(current->tgid);
985 * Accessing ->group_leader->real_parent is not SMP-safe, it could
986 * change from under us. However, rather than getting any lock
987 * we can use an optimistic algorithm: get the parent
988 * pid, and go back and check that the parent is still
989 * the same. If it has changed (which is extremely unlikely
990 * indeed), we just try again..
992 * NOTE! This depends on the fact that even if we _do_
993 * get an old value of "parent", we can happily dereference
994 * the pointer (it was and remains a dereferencable kernel pointer
995 * no matter what): we just can't necessarily trust the result
996 * until we know that the parent pointer is valid.
998 * NOTE2: ->group_leader never changes from under us.
1000 asmlinkage long sys_getppid(void)
1003 struct task_struct *me = current;
1004 struct task_struct *parent;
1006 parent = me->group_leader->real_parent;
1011 struct task_struct *old = parent;
1014 * Make sure we read the pid before re-reading the
1018 parent = me->group_leader->real_parent;
1025 return vx_map_pid(pid);
1031 * The Alpha uses getxpid, getxuid, and getxgid instead.
1034 asmlinkage long do_getxpid(long *ppid)
1036 *ppid = sys_getppid();
1037 return sys_getpid();
1042 asmlinkage long sys_getuid(void)
1044 /* Only we change this so SMP safe */
1045 return current->uid;
1048 asmlinkage long sys_geteuid(void)
1050 /* Only we change this so SMP safe */
1051 return current->euid;
1054 asmlinkage long sys_getgid(void)
1056 /* Only we change this so SMP safe */
1057 return current->gid;
1060 asmlinkage long sys_getegid(void)
1062 /* Only we change this so SMP safe */
1063 return current->egid;
1068 static void process_timeout(unsigned long __data)
1070 wake_up_process((task_t *)__data);
1074 * schedule_timeout - sleep until timeout
1075 * @timeout: timeout value in jiffies
1077 * Make the current task sleep until @timeout jiffies have
1078 * elapsed. The routine will return immediately unless
1079 * the current task state has been set (see set_current_state()).
1081 * You can set the task state as follows -
1083 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1084 * pass before the routine returns. The routine will return 0
1086 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1087 * delivered to the current task. In this case the remaining time
1088 * in jiffies will be returned, or 0 if the timer expired in time
1090 * The current task state is guaranteed to be TASK_RUNNING when this
1093 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1094 * the CPU away without a bound on the timeout. In this case the return
1095 * value will be %MAX_SCHEDULE_TIMEOUT.
1097 * In all cases the return value is guaranteed to be non-negative.
1099 fastcall signed long __sched schedule_timeout(signed long timeout)
1101 struct timer_list timer;
1102 unsigned long expire;
1104 if (crashdump_mode()) {
1105 diskdump_mdelay(timeout);
1106 set_current_state(TASK_RUNNING);
1112 case MAX_SCHEDULE_TIMEOUT:
1114 * These two special cases are useful to be comfortable
1115 * in the caller. Nothing more. We could take
1116 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1117 * but I' d like to return a valid offset (>=0) to allow
1118 * the caller to do everything it want with the retval.
1124 * Another bit of PARANOID. Note that the retval will be
1125 * 0 since no piece of kernel is supposed to do a check
1126 * for a negative retval of schedule_timeout() (since it
1127 * should never happens anyway). You just have the printk()
1128 * that will tell you if something is gone wrong and where.
1132 printk(KERN_ERR "schedule_timeout: wrong timeout "
1133 "value %lx from %p\n", timeout,
1134 __builtin_return_address(0));
1135 current->state = TASK_RUNNING;
1140 expire = timeout + jiffies;
1143 timer.expires = expire;
1144 timer.data = (unsigned long) current;
1145 timer.function = process_timeout;
1149 del_singleshot_timer_sync(&timer);
1151 timeout = expire - jiffies;
1154 return timeout < 0 ? 0 : timeout;
1157 EXPORT_SYMBOL(schedule_timeout);
1159 /* Thread ID - the internal kernel "pid" */
1160 asmlinkage long sys_gettid(void)
1162 return current->pid;
1165 static long __sched nanosleep_restart(struct restart_block *restart)
1167 unsigned long expire = restart->arg0, now = jiffies;
1168 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1171 /* Did it expire while we handled signals? */
1172 if (!time_after(expire, now))
1175 current->state = TASK_INTERRUPTIBLE;
1176 expire = schedule_timeout(expire - now);
1181 jiffies_to_timespec(expire, &t);
1183 ret = -ERESTART_RESTARTBLOCK;
1184 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1186 /* The 'restart' block is already filled in */
1191 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1194 unsigned long expire;
1197 if (copy_from_user(&t, rqtp, sizeof(t)))
1200 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1203 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1204 current->state = TASK_INTERRUPTIBLE;
1205 expire = schedule_timeout(expire);
1209 struct restart_block *restart;
1210 jiffies_to_timespec(expire, &t);
1211 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1214 restart = ¤t_thread_info()->restart_block;
1215 restart->fn = nanosleep_restart;
1216 restart->arg0 = jiffies + expire;
1217 restart->arg1 = (unsigned long) rmtp;
1218 ret = -ERESTART_RESTARTBLOCK;
1224 * sys_sysinfo - fill in sysinfo struct
1226 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1229 unsigned long mem_total, sav_total;
1230 unsigned int mem_unit, bitcount;
1233 memset((char *)&val, 0, sizeof(struct sysinfo));
1237 seq = read_seqbegin(&xtime_lock);
1240 * This is annoying. The below is the same thing
1241 * posix_get_clock_monotonic() does, but it wants to
1242 * take the lock which we want to cover the loads stuff
1246 getnstimeofday(&tp);
1247 tp.tv_sec += wall_to_monotonic.tv_sec;
1248 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1249 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1250 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1253 if (vx_flags(VXF_VIRT_UPTIME, 0))
1254 vx_vsi_uptime(&tp, NULL);
1255 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1257 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1258 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1259 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1261 val.procs = nr_threads;
1262 } while (read_seqretry(&xtime_lock, seq));
1268 * If the sum of all the available memory (i.e. ram + swap)
1269 * is less than can be stored in a 32 bit unsigned long then
1270 * we can be binary compatible with 2.2.x kernels. If not,
1271 * well, in that case 2.2.x was broken anyways...
1273 * -Erik Andersen <andersee@debian.org>
1276 mem_total = val.totalram + val.totalswap;
1277 if (mem_total < val.totalram || mem_total < val.totalswap)
1280 mem_unit = val.mem_unit;
1281 while (mem_unit > 1) {
1284 sav_total = mem_total;
1286 if (mem_total < sav_total)
1291 * If mem_total did not overflow, multiply all memory values by
1292 * val.mem_unit and set it to 1. This leaves things compatible
1293 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1298 val.totalram <<= bitcount;
1299 val.freeram <<= bitcount;
1300 val.sharedram <<= bitcount;
1301 val.bufferram <<= bitcount;
1302 val.totalswap <<= bitcount;
1303 val.freeswap <<= bitcount;
1304 val.totalhigh <<= bitcount;
1305 val.freehigh <<= bitcount;
1308 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1314 static void /* __devinit */ init_timers_cpu(int cpu)
1319 base = &per_cpu(tvec_bases, cpu);
1320 spin_lock_init(&base->lock);
1321 for (j = 0; j < TVN_SIZE; j++) {
1322 INIT_LIST_HEAD(base->tv5.vec + j);
1323 INIT_LIST_HEAD(base->tv4.vec + j);
1324 INIT_LIST_HEAD(base->tv3.vec + j);
1325 INIT_LIST_HEAD(base->tv2.vec + j);
1327 for (j = 0; j < TVR_SIZE; j++)
1328 INIT_LIST_HEAD(base->tv1.vec + j);
1330 base->timer_jiffies = jiffies;
1333 static tvec_base_t saved_tvec_base;
1335 void dump_clear_timers(void)
1337 tvec_base_t *base = &per_cpu(tvec_bases, smp_processor_id());
1339 memcpy(&saved_tvec_base, base, sizeof(saved_tvec_base));
1340 init_timers_cpu(smp_processor_id());
1343 EXPORT_SYMBOL_GPL(dump_clear_timers);
1345 void dump_run_timers(void)
1347 tvec_base_t *base = &__get_cpu_var(tvec_bases);
1352 EXPORT_SYMBOL_GPL(dump_run_timers);
1354 #ifdef CONFIG_HOTPLUG_CPU
1355 static int migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1357 struct timer_list *timer;
1359 while (!list_empty(head)) {
1360 timer = list_entry(head->next, struct timer_list, entry);
1361 /* We're locking backwards from __mod_timer order here,
1363 if (!spin_trylock(&timer->lock))
1365 list_del(&timer->entry);
1366 internal_add_timer(new_base, timer);
1367 timer->base = new_base;
1368 spin_unlock(&timer->lock);
1373 static void __devinit migrate_timers(int cpu)
1375 tvec_base_t *old_base;
1376 tvec_base_t *new_base;
1379 BUG_ON(cpu_online(cpu));
1380 old_base = &per_cpu(tvec_bases, cpu);
1381 new_base = &get_cpu_var(tvec_bases);
1383 local_irq_disable();
1385 /* Prevent deadlocks via ordering by old_base < new_base. */
1386 if (old_base < new_base) {
1387 spin_lock(&new_base->lock);
1388 spin_lock(&old_base->lock);
1390 spin_lock(&old_base->lock);
1391 spin_lock(&new_base->lock);
1394 if (old_base->running_timer)
1396 for (i = 0; i < TVR_SIZE; i++)
1397 if (!migrate_timer_list(new_base, old_base->tv1.vec + i))
1399 for (i = 0; i < TVN_SIZE; i++)
1400 if (!migrate_timer_list(new_base, old_base->tv2.vec + i)
1401 || !migrate_timer_list(new_base, old_base->tv3.vec + i)
1402 || !migrate_timer_list(new_base, old_base->tv4.vec + i)
1403 || !migrate_timer_list(new_base, old_base->tv5.vec + i))
1405 spin_unlock(&old_base->lock);
1406 spin_unlock(&new_base->lock);
1408 put_cpu_var(tvec_bases);
1412 /* Avoid deadlock with __mod_timer, by backing off. */
1413 spin_unlock(&old_base->lock);
1414 spin_unlock(&new_base->lock);
1418 #endif /* CONFIG_HOTPLUG_CPU */
1420 static int __devinit timer_cpu_notify(struct notifier_block *self,
1421 unsigned long action, void *hcpu)
1423 long cpu = (long)hcpu;
1425 case CPU_UP_PREPARE:
1426 init_timers_cpu(cpu);
1428 #ifdef CONFIG_HOTPLUG_CPU
1430 migrate_timers(cpu);
1439 static struct notifier_block __devinitdata timers_nb = {
1440 .notifier_call = timer_cpu_notify,
1444 void __init init_timers(void)
1446 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1447 (void *)(long)smp_processor_id());
1448 register_cpu_notifier(&timers_nb);
1449 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1452 #ifdef CONFIG_TIME_INTERPOLATION
1454 struct time_interpolator *time_interpolator;
1455 static struct time_interpolator *time_interpolator_list;
1456 static DEFINE_SPINLOCK(time_interpolator_lock);
1458 static inline u64 time_interpolator_get_cycles(unsigned int src)
1460 unsigned long (*x)(void);
1464 case TIME_SOURCE_FUNCTION:
1465 x = time_interpolator->addr;
1468 case TIME_SOURCE_MMIO64 :
1469 return readq((void __iomem *) time_interpolator->addr);
1471 case TIME_SOURCE_MMIO32 :
1472 return readl((void __iomem *) time_interpolator->addr);
1474 default: return get_cycles();
1478 static inline u64 time_interpolator_get_counter(void)
1480 unsigned int src = time_interpolator->source;
1482 if (time_interpolator->jitter)
1488 lcycle = time_interpolator->last_cycle;
1489 now = time_interpolator_get_cycles(src);
1490 if (lcycle && time_after(lcycle, now))
1492 /* Keep track of the last timer value returned. The use of cmpxchg here
1493 * will cause contention in an SMP environment.
1495 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1499 return time_interpolator_get_cycles(src);
1502 void time_interpolator_reset(void)
1504 time_interpolator->offset = 0;
1505 time_interpolator->last_counter = time_interpolator_get_counter();
1508 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1510 unsigned long time_interpolator_get_offset(void)
1512 /* If we do not have a time interpolator set up then just return zero */
1513 if (!time_interpolator)
1516 return time_interpolator->offset +
1517 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator);
1520 #define INTERPOLATOR_ADJUST 65536
1521 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1523 static void time_interpolator_update(long delta_nsec)
1526 unsigned long offset;
1528 /* If there is no time interpolator set up then do nothing */
1529 if (!time_interpolator)
1532 /* The interpolator compensates for late ticks by accumulating
1533 * the late time in time_interpolator->offset. A tick earlier than
1534 * expected will lead to a reset of the offset and a corresponding
1535 * jump of the clock forward. Again this only works if the
1536 * interpolator clock is running slightly slower than the regular clock
1537 * and the tuning logic insures that.
1540 counter = time_interpolator_get_counter();
1541 offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
1543 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1544 time_interpolator->offset = offset - delta_nsec;
1546 time_interpolator->skips++;
1547 time_interpolator->ns_skipped += delta_nsec - offset;
1548 time_interpolator->offset = 0;
1550 time_interpolator->last_counter = counter;
1552 /* Tuning logic for time interpolator invoked every minute or so.
1553 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1554 * Increase interpolator clock speed if we skip too much time.
1556 if (jiffies % INTERPOLATOR_ADJUST == 0)
1558 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1559 time_interpolator->nsec_per_cyc--;
1560 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1561 time_interpolator->nsec_per_cyc++;
1562 time_interpolator->skips = 0;
1563 time_interpolator->ns_skipped = 0;
1568 is_better_time_interpolator(struct time_interpolator *new)
1570 if (!time_interpolator)
1572 return new->frequency > 2*time_interpolator->frequency ||
1573 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1577 register_time_interpolator(struct time_interpolator *ti)
1579 unsigned long flags;
1582 if (ti->frequency == 0 || ti->mask == 0)
1585 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1586 spin_lock(&time_interpolator_lock);
1587 write_seqlock_irqsave(&xtime_lock, flags);
1588 if (is_better_time_interpolator(ti)) {
1589 time_interpolator = ti;
1590 time_interpolator_reset();
1592 write_sequnlock_irqrestore(&xtime_lock, flags);
1594 ti->next = time_interpolator_list;
1595 time_interpolator_list = ti;
1596 spin_unlock(&time_interpolator_lock);
1600 unregister_time_interpolator(struct time_interpolator *ti)
1602 struct time_interpolator *curr, **prev;
1603 unsigned long flags;
1605 spin_lock(&time_interpolator_lock);
1606 prev = &time_interpolator_list;
1607 for (curr = *prev; curr; curr = curr->next) {
1615 write_seqlock_irqsave(&xtime_lock, flags);
1616 if (ti == time_interpolator) {
1617 /* we lost the best time-interpolator: */
1618 time_interpolator = NULL;
1619 /* find the next-best interpolator */
1620 for (curr = time_interpolator_list; curr; curr = curr->next)
1621 if (is_better_time_interpolator(curr))
1622 time_interpolator = curr;
1623 time_interpolator_reset();
1625 write_sequnlock_irqrestore(&xtime_lock, flags);
1626 spin_unlock(&time_interpolator_lock);
1628 #endif /* CONFIG_TIME_INTERPOLATION */
1631 * msleep - sleep safely even with waitqueue interruptions
1632 * @msecs: Time in milliseconds to sleep for
1634 void msleep(unsigned int msecs)
1636 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1638 if (unlikely(crashdump_mode())) {
1639 while (msecs--) udelay(1000);
1644 set_current_state(TASK_UNINTERRUPTIBLE);
1645 timeout = schedule_timeout(timeout);
1649 EXPORT_SYMBOL(msleep);
1652 * msleep_interruptible - sleep waiting for waitqueue interruptions
1653 * @msecs: Time in milliseconds to sleep for
1655 unsigned long msleep_interruptible(unsigned int msecs)
1657 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1659 while (timeout && !signal_pending(current)) {
1660 set_current_state(TASK_INTERRUPTIBLE);
1661 timeout = schedule_timeout(timeout);
1663 return jiffies_to_msecs(timeout);
1666 EXPORT_SYMBOL(msleep_interruptible);