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/cpu.h>
34 #include <linux/vs_cvirt.h>
35 #include <linux/vserver/sched.h>
37 #include <asm/uaccess.h>
38 #include <asm/unistd.h>
39 #include <asm/div64.h>
40 #include <asm/timex.h>
43 #ifdef CONFIG_TIME_INTERPOLATION
44 static void time_interpolator_update(long delta_nsec);
46 #define time_interpolator_update(x)
50 * per-CPU timer vector definitions:
54 #define TVN_SIZE (1 << TVN_BITS)
55 #define TVR_SIZE (1 << TVR_BITS)
56 #define TVN_MASK (TVN_SIZE - 1)
57 #define TVR_MASK (TVR_SIZE - 1)
59 typedef struct tvec_s {
60 struct list_head vec[TVN_SIZE];
63 typedef struct tvec_root_s {
64 struct list_head vec[TVR_SIZE];
67 struct tvec_t_base_s {
69 unsigned long timer_jiffies;
70 struct timer_list *running_timer;
76 } ____cacheline_aligned_in_smp;
78 typedef struct tvec_t_base_s tvec_base_t;
80 static inline void set_running_timer(tvec_base_t *base,
81 struct timer_list *timer)
84 base->running_timer = timer;
88 /* Fake initialization */
89 static DEFINE_PER_CPU(tvec_base_t, tvec_bases) = { SPIN_LOCK_UNLOCKED };
91 static void check_timer_failed(struct timer_list *timer)
93 static int whine_count;
94 if (whine_count < 16) {
96 printk("Uninitialised timer!\n");
97 printk("This is just a warning. Your computer is OK\n");
98 printk("function=0x%p, data=0x%lx\n",
99 timer->function, timer->data);
105 spin_lock_init(&timer->lock);
106 timer->magic = TIMER_MAGIC;
109 static inline void check_timer(struct timer_list *timer)
111 if (timer->magic != TIMER_MAGIC)
112 check_timer_failed(timer);
116 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
118 unsigned long expires = timer->expires;
119 unsigned long idx = expires - base->timer_jiffies;
120 struct list_head *vec;
122 if (idx < TVR_SIZE) {
123 int i = expires & TVR_MASK;
124 vec = base->tv1.vec + i;
125 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
126 int i = (expires >> TVR_BITS) & TVN_MASK;
127 vec = base->tv2.vec + i;
128 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
129 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
130 vec = base->tv3.vec + i;
131 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
132 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
133 vec = base->tv4.vec + i;
134 } else if ((signed long) idx < 0) {
136 * Can happen if you add a timer with expires == jiffies,
137 * or you set a timer to go off in the past
139 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
142 /* If the timeout is larger than 0xffffffff on 64-bit
143 * architectures then we use the maximum timeout:
145 if (idx > 0xffffffffUL) {
147 expires = idx + base->timer_jiffies;
149 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
150 vec = base->tv5.vec + i;
155 list_add_tail(&timer->entry, vec);
158 int __mod_timer(struct timer_list *timer, unsigned long expires)
160 tvec_base_t *old_base, *new_base;
164 BUG_ON(!timer->function);
168 spin_lock_irqsave(&timer->lock, flags);
169 new_base = &__get_cpu_var(tvec_bases);
171 old_base = timer->base;
174 * Prevent deadlocks via ordering by old_base < new_base.
176 if (old_base && (new_base != old_base)) {
177 if (old_base < new_base) {
178 spin_lock(&new_base->lock);
179 spin_lock(&old_base->lock);
181 spin_lock(&old_base->lock);
182 spin_lock(&new_base->lock);
185 * The timer base might have been cancelled while we were
186 * trying to take the lock(s):
188 if (timer->base != old_base) {
189 spin_unlock(&new_base->lock);
190 spin_unlock(&old_base->lock);
194 spin_lock(&new_base->lock);
195 if (timer->base != old_base) {
196 spin_unlock(&new_base->lock);
202 * Delete the previous timeout (if there was any), and install
206 list_del(&timer->entry);
209 timer->expires = expires;
210 internal_add_timer(new_base, timer);
211 timer->base = new_base;
213 if (old_base && (new_base != old_base))
214 spin_unlock(&old_base->lock);
215 spin_unlock(&new_base->lock);
216 spin_unlock_irqrestore(&timer->lock, flags);
221 EXPORT_SYMBOL(__mod_timer);
224 * add_timer_on - start a timer on a particular CPU
225 * @timer: the timer to be added
226 * @cpu: the CPU to start it on
228 * This is not very scalable on SMP. Double adds are not possible.
230 void add_timer_on(struct timer_list *timer, int cpu)
232 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
235 BUG_ON(timer_pending(timer) || !timer->function);
239 spin_lock_irqsave(&base->lock, flags);
240 internal_add_timer(base, timer);
242 spin_unlock_irqrestore(&base->lock, flags);
245 EXPORT_SYMBOL(add_timer_on);
248 * mod_timer - modify a timer's timeout
249 * @timer: the timer to be modified
251 * mod_timer is a more efficient way to update the expire field of an
252 * active timer (if the timer is inactive it will be activated)
254 * mod_timer(timer, expires) is equivalent to:
256 * del_timer(timer); timer->expires = expires; add_timer(timer);
258 * Note that if there are multiple unserialized concurrent users of the
259 * same timer, then mod_timer() is the only safe way to modify the timeout,
260 * since add_timer() cannot modify an already running timer.
262 * The function returns whether it has modified a pending timer or not.
263 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
264 * active timer returns 1.)
266 int mod_timer(struct timer_list *timer, unsigned long expires)
268 BUG_ON(!timer->function);
273 * This is a common optimization triggered by the
274 * networking code - if the timer is re-modified
275 * to be the same thing then just return:
277 if (timer->expires == expires && timer_pending(timer))
280 return __mod_timer(timer, expires);
283 EXPORT_SYMBOL(mod_timer);
286 * del_timer - deactive a timer.
287 * @timer: the timer to be deactivated
289 * del_timer() deactivates a timer - this works on both active and inactive
292 * The function returns whether it has deactivated a pending timer or not.
293 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
294 * active timer returns 1.)
296 int del_timer(struct timer_list *timer)
307 spin_lock_irqsave(&base->lock, flags);
308 if (base != timer->base) {
309 spin_unlock_irqrestore(&base->lock, flags);
312 list_del(&timer->entry);
314 spin_unlock_irqrestore(&base->lock, flags);
319 EXPORT_SYMBOL(del_timer);
323 * del_timer_sync - deactivate a timer and wait for the handler to finish.
324 * @timer: the timer to be deactivated
326 * This function only differs from del_timer() on SMP: besides deactivating
327 * the timer it also makes sure the handler has finished executing on other
330 * Synchronization rules: callers must prevent restarting of the timer,
331 * otherwise this function is meaningless. It must not be called from
332 * interrupt contexts. The caller must not hold locks which would prevent
333 * completion of the timer's handler. Upon exit the timer is not queued and
334 * the handler is not running on any CPU.
336 * The function returns whether it has deactivated a pending timer or not.
338 * del_timer_sync() is slow and complicated because it copes with timer
339 * handlers which re-arm the timer (periodic timers). If the timer handler
340 * is known to not do this (a single shot timer) then use
341 * del_singleshot_timer_sync() instead.
343 int del_timer_sync(struct timer_list *timer)
351 ret += del_timer(timer);
353 for_each_online_cpu(i) {
354 base = &per_cpu(tvec_bases, i);
355 if (base->running_timer == timer) {
356 while (base->running_timer == timer) {
358 preempt_check_resched();
364 if (timer_pending(timer))
369 EXPORT_SYMBOL(del_timer_sync);
372 * del_singleshot_timer_sync - deactivate a non-recursive timer
373 * @timer: the timer to be deactivated
375 * This function is an optimization of del_timer_sync for the case where the
376 * caller can guarantee the timer does not reschedule itself in its timer
379 * Synchronization rules: callers must prevent restarting of the timer,
380 * otherwise this function is meaningless. It must not be called from
381 * interrupt contexts. The caller must not hold locks which wold prevent
382 * completion of the timer's handler. Upon exit the timer is not queued and
383 * the handler is not running on any CPU.
385 * The function returns whether it has deactivated a pending timer or not.
387 int del_singleshot_timer_sync(struct timer_list *timer)
389 int ret = del_timer(timer);
392 ret = del_timer_sync(timer);
398 EXPORT_SYMBOL(del_singleshot_timer_sync);
401 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
403 /* cascade all the timers from tv up one level */
404 struct list_head *head, *curr;
406 head = tv->vec + index;
409 * We are removing _all_ timers from the list, so we don't have to
410 * detach them individually, just clear the list afterwards.
412 while (curr != head) {
413 struct timer_list *tmp;
415 tmp = list_entry(curr, struct timer_list, entry);
416 BUG_ON(tmp->base != base);
418 internal_add_timer(base, tmp);
420 INIT_LIST_HEAD(head);
426 * __run_timers - run all expired timers (if any) on this CPU.
427 * @base: the timer vector to be processed.
429 * This function cascades all vectors and executes all expired timer
432 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
434 static inline void __run_timers(tvec_base_t *base)
436 struct timer_list *timer;
438 spin_lock_irq(&base->lock);
439 while (time_after_eq(jiffies, base->timer_jiffies)) {
440 struct list_head work_list = LIST_HEAD_INIT(work_list);
441 struct list_head *head = &work_list;
442 int index = base->timer_jiffies & TVR_MASK;
448 (!cascade(base, &base->tv2, INDEX(0))) &&
449 (!cascade(base, &base->tv3, INDEX(1))) &&
450 !cascade(base, &base->tv4, INDEX(2)))
451 cascade(base, &base->tv5, INDEX(3));
452 ++base->timer_jiffies;
453 list_splice_init(base->tv1.vec + index, &work_list);
455 if (!list_empty(head)) {
456 void (*fn)(unsigned long);
459 timer = list_entry(head->next,struct timer_list,entry);
460 fn = timer->function;
463 list_del(&timer->entry);
464 set_running_timer(base, timer);
467 spin_unlock_irq(&base->lock);
469 spin_lock_irq(&base->lock);
473 set_running_timer(base, NULL);
474 spin_unlock_irq(&base->lock);
477 #ifdef CONFIG_NO_IDLE_HZ
479 * Find out when the next timer event is due to happen. This
480 * is used on S/390 to stop all activity when a cpus is idle.
481 * This functions needs to be called disabled.
483 unsigned long next_timer_interrupt(void)
486 struct list_head *list;
487 struct timer_list *nte;
488 unsigned long expires;
492 base = &__get_cpu_var(tvec_bases);
493 spin_lock(&base->lock);
494 expires = base->timer_jiffies + (LONG_MAX >> 1);
497 /* Look for timer events in tv1. */
498 j = base->timer_jiffies & TVR_MASK;
500 list_for_each_entry(nte, base->tv1.vec + j, entry) {
501 expires = nte->expires;
502 if (j < (base->timer_jiffies & TVR_MASK))
503 list = base->tv2.vec + (INDEX(0));
506 j = (j + 1) & TVR_MASK;
507 } while (j != (base->timer_jiffies & TVR_MASK));
510 varray[0] = &base->tv2;
511 varray[1] = &base->tv3;
512 varray[2] = &base->tv4;
513 varray[3] = &base->tv5;
514 for (i = 0; i < 4; i++) {
517 if (list_empty(varray[i]->vec + j)) {
518 j = (j + 1) & TVN_MASK;
521 list_for_each_entry(nte, varray[i]->vec + j, entry)
522 if (time_before(nte->expires, expires))
523 expires = nte->expires;
524 if (j < (INDEX(i)) && i < 3)
525 list = varray[i + 1]->vec + (INDEX(i + 1));
527 } while (j != (INDEX(i)));
532 * The search wrapped. We need to look at the next list
533 * from next tv element that would cascade into tv element
534 * where we found the timer element.
536 list_for_each_entry(nte, list, entry) {
537 if (time_before(nte->expires, expires))
538 expires = nte->expires;
541 spin_unlock(&base->lock);
546 /******************************************************************/
549 * Timekeeping variables
551 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
552 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
556 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
557 * for sub jiffie times) to get to monotonic time. Monotonic is pegged at zero
558 * at zero at system boot time, so wall_to_monotonic will be negative,
559 * however, we will ALWAYS keep the tv_nsec part positive so we can use
560 * the usual normalization.
562 struct timespec xtime __attribute__ ((aligned (16)));
563 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
565 EXPORT_SYMBOL(xtime);
567 /* Don't completely fail for HZ > 500. */
568 int tickadj = 500/HZ ? : 1; /* microsecs */
572 * phase-lock loop variables
574 /* TIME_ERROR prevents overwriting the CMOS clock */
575 int time_state = TIME_OK; /* clock synchronization status */
576 int time_status = STA_UNSYNC; /* clock status bits */
577 long time_offset; /* time adjustment (us) */
578 long time_constant = 2; /* pll time constant */
579 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
580 long time_precision = 1; /* clock precision (us) */
581 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
582 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
583 long time_phase; /* phase offset (scaled us) */
584 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
585 /* frequency offset (scaled ppm)*/
586 long time_adj; /* tick adjust (scaled 1 / HZ) */
587 long time_reftime; /* time at last adjustment (s) */
589 long time_next_adjust;
592 * this routine handles the overflow of the microsecond field
594 * The tricky bits of code to handle the accurate clock support
595 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
596 * They were originally developed for SUN and DEC kernels.
597 * All the kudos should go to Dave for this stuff.
600 static void second_overflow(void)
604 /* Bump the maxerror field */
605 time_maxerror += time_tolerance >> SHIFT_USEC;
606 if ( time_maxerror > NTP_PHASE_LIMIT ) {
607 time_maxerror = NTP_PHASE_LIMIT;
608 time_status |= STA_UNSYNC;
612 * Leap second processing. If in leap-insert state at
613 * the end of the day, the system clock is set back one
614 * second; if in leap-delete state, the system clock is
615 * set ahead one second. The microtime() routine or
616 * external clock driver will insure that reported time
617 * is always monotonic. The ugly divides should be
620 switch (time_state) {
623 if (time_status & STA_INS)
624 time_state = TIME_INS;
625 else if (time_status & STA_DEL)
626 time_state = TIME_DEL;
630 if (xtime.tv_sec % 86400 == 0) {
632 wall_to_monotonic.tv_sec++;
633 /* The timer interpolator will make time change gradually instead
634 * of an immediate jump by one second.
636 time_interpolator_update(-NSEC_PER_SEC);
637 time_state = TIME_OOP;
639 printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
644 if ((xtime.tv_sec + 1) % 86400 == 0) {
646 wall_to_monotonic.tv_sec--;
647 /* Use of time interpolator for a gradual change of time */
648 time_interpolator_update(NSEC_PER_SEC);
649 time_state = TIME_WAIT;
651 printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
656 time_state = TIME_WAIT;
660 if (!(time_status & (STA_INS | STA_DEL)))
661 time_state = TIME_OK;
665 * Compute the phase adjustment for the next second. In
666 * PLL mode, the offset is reduced by a fixed factor
667 * times the time constant. In FLL mode the offset is
668 * used directly. In either mode, the maximum phase
669 * adjustment for each second is clamped so as to spread
670 * the adjustment over not more than the number of
671 * seconds between updates.
673 if (time_offset < 0) {
674 ltemp = -time_offset;
675 if (!(time_status & STA_FLL))
676 ltemp >>= SHIFT_KG + time_constant;
677 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
678 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
679 time_offset += ltemp;
680 time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
683 if (!(time_status & STA_FLL))
684 ltemp >>= SHIFT_KG + time_constant;
685 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
686 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
687 time_offset -= ltemp;
688 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
692 * Compute the frequency estimate and additional phase
693 * adjustment due to frequency error for the next
694 * second. When the PPS signal is engaged, gnaw on the
695 * watchdog counter and update the frequency computed by
696 * the pll and the PPS signal.
699 if (pps_valid == PPS_VALID) { /* PPS signal lost */
700 pps_jitter = MAXTIME;
701 pps_stabil = MAXFREQ;
702 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
703 STA_PPSWANDER | STA_PPSERROR);
705 ltemp = time_freq + pps_freq;
707 time_adj -= -ltemp >>
708 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
711 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
714 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
715 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
718 time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
720 time_adj += (time_adj >> 2) + (time_adj >> 5);
723 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
724 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
727 time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
729 time_adj += (time_adj >> 6) + (time_adj >> 7);
733 /* in the NTP reference this is called "hardclock()" */
734 static void update_wall_time_one_tick(void)
736 long time_adjust_step, delta_nsec;
738 if ( (time_adjust_step = time_adjust) != 0 ) {
739 /* We are doing an adjtime thing.
741 * Prepare time_adjust_step to be within bounds.
742 * Note that a positive time_adjust means we want the clock
745 * Limit the amount of the step to be in the range
746 * -tickadj .. +tickadj
748 if (time_adjust > tickadj)
749 time_adjust_step = tickadj;
750 else if (time_adjust < -tickadj)
751 time_adjust_step = -tickadj;
753 /* Reduce by this step the amount of time left */
754 time_adjust -= time_adjust_step;
756 delta_nsec = tick_nsec + time_adjust_step * 1000;
758 * Advance the phase, once it gets to one microsecond, then
759 * advance the tick more.
761 time_phase += time_adj;
762 if (time_phase <= -FINENSEC) {
763 long ltemp = -time_phase >> (SHIFT_SCALE - 10);
764 time_phase += ltemp << (SHIFT_SCALE - 10);
767 else if (time_phase >= FINENSEC) {
768 long ltemp = time_phase >> (SHIFT_SCALE - 10);
769 time_phase -= ltemp << (SHIFT_SCALE - 10);
772 xtime.tv_nsec += delta_nsec;
773 time_interpolator_update(delta_nsec);
775 /* Changes by adjtime() do not take effect till next tick. */
776 if (time_next_adjust != 0) {
777 time_adjust = time_next_adjust;
778 time_next_adjust = 0;
783 * Using a loop looks inefficient, but "ticks" is
784 * usually just one (we shouldn't be losing ticks,
785 * we're doing this this way mainly for interrupt
786 * latency reasons, not because we think we'll
787 * have lots of lost timer ticks
789 static void update_wall_time(unsigned long ticks)
793 update_wall_time_one_tick();
796 if (xtime.tv_nsec >= 1000000000) {
797 xtime.tv_nsec -= 1000000000;
803 static inline void do_process_times(struct task_struct *p,
804 unsigned long user, unsigned long system)
808 psecs = (p->utime += user);
809 psecs += (p->stime += system);
810 if (psecs / HZ >= p->rlim[RLIMIT_CPU].rlim_cur) {
811 /* Send SIGXCPU every second.. */
813 send_sig(SIGXCPU, p, 1);
814 /* and SIGKILL when we go over max.. */
815 if (psecs / HZ >= p->rlim[RLIMIT_CPU].rlim_max)
816 send_sig(SIGKILL, p, 1);
820 static inline void do_it_virt(struct task_struct * p, unsigned long ticks)
822 unsigned long it_virt = p->it_virt_value;
827 it_virt = p->it_virt_incr;
828 send_sig(SIGVTALRM, p, 1);
830 p->it_virt_value = it_virt;
834 static inline void do_it_prof(struct task_struct *p)
836 unsigned long it_prof = p->it_prof_value;
839 if (--it_prof == 0) {
840 it_prof = p->it_prof_incr;
841 send_sig(SIGPROF, p, 1);
843 p->it_prof_value = it_prof;
847 static void update_one_process(struct task_struct *p, unsigned long user,
848 unsigned long system, int cpu)
850 do_process_times(p, user, system);
856 * Called from the timer interrupt handler to charge one tick to the current
857 * process. user_tick is 1 if the tick is user time, 0 for system.
859 void update_process_times(int user_tick)
861 struct task_struct *p = current;
862 int cpu = smp_processor_id(), system = user_tick ^ 1;
864 update_one_process(p, user_tick, system, cpu);
866 scheduler_tick(user_tick, system);
870 * Nr of active tasks - counted in fixed-point numbers
872 static unsigned long count_active_tasks(void)
874 return (nr_running() + nr_uninterruptible()) * FIXED_1;
878 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
879 * imply that avenrun[] is the standard name for this kind of thing.
880 * Nothing else seems to be standardized: the fractional size etc
881 * all seem to differ on different machines.
883 * Requires xtime_lock to access.
885 unsigned long avenrun[3];
888 * calc_load - given tick count, update the avenrun load estimates.
889 * This is called while holding a write_lock on xtime_lock.
891 static inline void calc_load(unsigned long ticks)
893 unsigned long active_tasks; /* fixed-point */
894 static int count = LOAD_FREQ;
899 active_tasks = count_active_tasks();
900 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
901 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
902 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
906 /* jiffies at the most recent update of wall time */
907 unsigned long wall_jiffies = INITIAL_JIFFIES;
910 * This read-write spinlock protects us from races in SMP while
911 * playing with xtime and avenrun.
913 #ifndef ARCH_HAVE_XTIME_LOCK
914 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
916 EXPORT_SYMBOL(xtime_lock);
920 * This function runs timers and the timer-tq in bottom half context.
922 static void run_timer_softirq(struct softirq_action *h)
924 tvec_base_t *base = &__get_cpu_var(tvec_bases);
926 if (time_after_eq(jiffies, base->timer_jiffies))
931 * Called by the local, per-CPU timer interrupt on SMP.
933 void run_local_timers(void)
935 raise_softirq(TIMER_SOFTIRQ);
939 * Called by the timer interrupt. xtime_lock must already be taken
942 static inline void update_times(void)
946 ticks = jiffies - wall_jiffies;
948 wall_jiffies += ticks;
949 update_wall_time(ticks);
955 * The 64-bit jiffies value is not atomic - you MUST NOT read it
956 * without sampling the sequence number in xtime_lock.
957 * jiffies is defined in the linker script...
960 void do_timer(struct pt_regs *regs)
964 /* SMP process accounting uses the local APIC timer */
966 update_process_times(user_mode(regs));
971 #ifdef __ARCH_WANT_SYS_ALARM
974 * For backwards compatibility? This can be done in libc so Alpha
975 * and all newer ports shouldn't need it.
977 asmlinkage unsigned long sys_alarm(unsigned int seconds)
979 struct itimerval it_new, it_old;
980 unsigned int oldalarm;
982 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
983 it_new.it_value.tv_sec = seconds;
984 it_new.it_value.tv_usec = 0;
985 do_setitimer(ITIMER_REAL, &it_new, &it_old);
986 oldalarm = it_old.it_value.tv_sec;
987 /* ehhh.. We can't return 0 if we have an alarm pending.. */
988 /* And we'd better return too much than too little anyway */
989 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
999 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1000 * should be moved into arch/i386 instead?
1004 * sys_getpid - return the thread group id of the current process
1006 * Note, despite the name, this returns the tgid not the pid. The tgid and
1007 * the pid are identical unless CLONE_THREAD was specified on clone() in
1008 * which case the tgid is the same in all threads of the same group.
1010 * This is SMP safe as current->tgid does not change.
1012 asmlinkage long sys_getpid(void)
1014 return vx_map_tgid(current->tgid);
1018 * Accessing ->group_leader->real_parent is not SMP-safe, it could
1019 * change from under us. However, rather than getting any lock
1020 * we can use an optimistic algorithm: get the parent
1021 * pid, and go back and check that the parent is still
1022 * the same. If it has changed (which is extremely unlikely
1023 * indeed), we just try again..
1025 * NOTE! This depends on the fact that even if we _do_
1026 * get an old value of "parent", we can happily dereference
1027 * the pointer (it was and remains a dereferencable kernel pointer
1028 * no matter what): we just can't necessarily trust the result
1029 * until we know that the parent pointer is valid.
1031 * NOTE2: ->group_leader never changes from under us.
1033 asmlinkage long sys_getppid(void)
1036 struct task_struct *me = current;
1037 struct task_struct *parent;
1039 parent = me->group_leader->real_parent;
1044 struct task_struct *old = parent;
1047 * Make sure we read the pid before re-reading the
1051 parent = me->group_leader->real_parent;
1058 return vx_map_pid(pid);
1061 asmlinkage long sys_getuid(void)
1063 /* Only we change this so SMP safe */
1064 return current->uid;
1067 asmlinkage long sys_geteuid(void)
1069 /* Only we change this so SMP safe */
1070 return current->euid;
1073 asmlinkage long sys_getgid(void)
1075 /* Only we change this so SMP safe */
1076 return current->gid;
1079 asmlinkage long sys_getegid(void)
1081 /* Only we change this so SMP safe */
1082 return current->egid;
1087 static void process_timeout(unsigned long __data)
1089 wake_up_process((task_t *)__data);
1093 * schedule_timeout - sleep until timeout
1094 * @timeout: timeout value in jiffies
1096 * Make the current task sleep until @timeout jiffies have
1097 * elapsed. The routine will return immediately unless
1098 * the current task state has been set (see set_current_state()).
1100 * You can set the task state as follows -
1102 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1103 * pass before the routine returns. The routine will return 0
1105 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1106 * delivered to the current task. In this case the remaining time
1107 * in jiffies will be returned, or 0 if the timer expired in time
1109 * The current task state is guaranteed to be TASK_RUNNING when this
1112 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1113 * the CPU away without a bound on the timeout. In this case the return
1114 * value will be %MAX_SCHEDULE_TIMEOUT.
1116 * In all cases the return value is guaranteed to be non-negative.
1118 fastcall signed long __sched schedule_timeout(signed long timeout)
1120 struct timer_list timer;
1121 unsigned long expire;
1125 case MAX_SCHEDULE_TIMEOUT:
1127 * These two special cases are useful to be comfortable
1128 * in the caller. Nothing more. We could take
1129 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1130 * but I' d like to return a valid offset (>=0) to allow
1131 * the caller to do everything it want with the retval.
1137 * Another bit of PARANOID. Note that the retval will be
1138 * 0 since no piece of kernel is supposed to do a check
1139 * for a negative retval of schedule_timeout() (since it
1140 * should never happens anyway). You just have the printk()
1141 * that will tell you if something is gone wrong and where.
1145 printk(KERN_ERR "schedule_timeout: wrong timeout "
1146 "value %lx from %p\n", timeout,
1147 __builtin_return_address(0));
1148 current->state = TASK_RUNNING;
1153 expire = timeout + jiffies;
1156 timer.expires = expire;
1157 timer.data = (unsigned long) current;
1158 timer.function = process_timeout;
1162 del_singleshot_timer_sync(&timer);
1164 timeout = expire - jiffies;
1167 return timeout < 0 ? 0 : timeout;
1170 EXPORT_SYMBOL(schedule_timeout);
1172 /* Thread ID - the internal kernel "pid" */
1173 asmlinkage long sys_gettid(void)
1175 return current->pid;
1178 static long __sched nanosleep_restart(struct restart_block *restart)
1180 unsigned long expire = restart->arg0, now = jiffies;
1181 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1184 /* Did it expire while we handled signals? */
1185 if (!time_after(expire, now))
1188 current->state = TASK_INTERRUPTIBLE;
1189 expire = schedule_timeout(expire - now);
1194 jiffies_to_timespec(expire, &t);
1196 ret = -ERESTART_RESTARTBLOCK;
1197 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1199 /* The 'restart' block is already filled in */
1204 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1207 unsigned long expire;
1210 if (copy_from_user(&t, rqtp, sizeof(t)))
1213 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1216 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1217 current->state = TASK_INTERRUPTIBLE;
1218 expire = schedule_timeout(expire);
1222 struct restart_block *restart;
1223 jiffies_to_timespec(expire, &t);
1224 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1227 restart = ¤t_thread_info()->restart_block;
1228 restart->fn = nanosleep_restart;
1229 restart->arg0 = jiffies + expire;
1230 restart->arg1 = (unsigned long) rmtp;
1231 ret = -ERESTART_RESTARTBLOCK;
1237 * sys_sysinfo - fill in sysinfo struct
1239 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1242 unsigned long mem_total, sav_total;
1243 unsigned int mem_unit, bitcount;
1246 memset((char *)&val, 0, sizeof(struct sysinfo));
1250 seq = read_seqbegin(&xtime_lock);
1253 * This is annoying. The below is the same thing
1254 * posix_get_clock_monotonic() does, but it wants to
1255 * take the lock which we want to cover the loads stuff
1259 getnstimeofday(&tp);
1260 tp.tv_sec += wall_to_monotonic.tv_sec;
1261 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1262 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1263 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1266 if (vx_flags(VXF_VIRT_UPTIME, 0))
1267 vx_vsi_uptime(&tp, NULL);
1268 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1270 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1271 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1272 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1274 val.procs = nr_threads;
1275 } while (read_seqretry(&xtime_lock, seq));
1277 /* if (vx_flags(VXF_VIRT_CPU, 0))
1284 * If the sum of all the available memory (i.e. ram + swap)
1285 * is less than can be stored in a 32 bit unsigned long then
1286 * we can be binary compatible with 2.2.x kernels. If not,
1287 * well, in that case 2.2.x was broken anyways...
1289 * -Erik Andersen <andersee@debian.org>
1292 mem_total = val.totalram + val.totalswap;
1293 if (mem_total < val.totalram || mem_total < val.totalswap)
1296 mem_unit = val.mem_unit;
1297 while (mem_unit > 1) {
1300 sav_total = mem_total;
1302 if (mem_total < sav_total)
1307 * If mem_total did not overflow, multiply all memory values by
1308 * val.mem_unit and set it to 1. This leaves things compatible
1309 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1314 val.totalram <<= bitcount;
1315 val.freeram <<= bitcount;
1316 val.sharedram <<= bitcount;
1317 val.bufferram <<= bitcount;
1318 val.totalswap <<= bitcount;
1319 val.freeswap <<= bitcount;
1320 val.totalhigh <<= bitcount;
1321 val.freehigh <<= bitcount;
1324 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1330 static void __devinit init_timers_cpu(int cpu)
1335 base = &per_cpu(tvec_bases, cpu);
1336 spin_lock_init(&base->lock);
1337 for (j = 0; j < TVN_SIZE; j++) {
1338 INIT_LIST_HEAD(base->tv5.vec + j);
1339 INIT_LIST_HEAD(base->tv4.vec + j);
1340 INIT_LIST_HEAD(base->tv3.vec + j);
1341 INIT_LIST_HEAD(base->tv2.vec + j);
1343 for (j = 0; j < TVR_SIZE; j++)
1344 INIT_LIST_HEAD(base->tv1.vec + j);
1346 base->timer_jiffies = jiffies;
1349 #ifdef CONFIG_HOTPLUG_CPU
1350 static int migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1352 struct timer_list *timer;
1354 while (!list_empty(head)) {
1355 timer = list_entry(head->next, struct timer_list, entry);
1356 /* We're locking backwards from __mod_timer order here,
1358 if (!spin_trylock(&timer->lock))
1360 list_del(&timer->entry);
1361 internal_add_timer(new_base, timer);
1362 timer->base = new_base;
1363 spin_unlock(&timer->lock);
1368 static void __devinit migrate_timers(int cpu)
1370 tvec_base_t *old_base;
1371 tvec_base_t *new_base;
1374 BUG_ON(cpu_online(cpu));
1375 old_base = &per_cpu(tvec_bases, cpu);
1376 new_base = &get_cpu_var(tvec_bases);
1378 local_irq_disable();
1380 /* Prevent deadlocks via ordering by old_base < new_base. */
1381 if (old_base < new_base) {
1382 spin_lock(&new_base->lock);
1383 spin_lock(&old_base->lock);
1385 spin_lock(&old_base->lock);
1386 spin_lock(&new_base->lock);
1389 if (old_base->running_timer)
1391 for (i = 0; i < TVR_SIZE; i++)
1392 if (!migrate_timer_list(new_base, old_base->tv1.vec + i))
1394 for (i = 0; i < TVN_SIZE; i++)
1395 if (!migrate_timer_list(new_base, old_base->tv2.vec + i)
1396 || !migrate_timer_list(new_base, old_base->tv3.vec + i)
1397 || !migrate_timer_list(new_base, old_base->tv4.vec + i)
1398 || !migrate_timer_list(new_base, old_base->tv5.vec + i))
1400 spin_unlock(&old_base->lock);
1401 spin_unlock(&new_base->lock);
1403 put_cpu_var(tvec_bases);
1407 /* Avoid deadlock with __mod_timer, by backing off. */
1408 spin_unlock(&old_base->lock);
1409 spin_unlock(&new_base->lock);
1413 #endif /* CONFIG_HOTPLUG_CPU */
1415 static int __devinit timer_cpu_notify(struct notifier_block *self,
1416 unsigned long action, void *hcpu)
1418 long cpu = (long)hcpu;
1420 case CPU_UP_PREPARE:
1421 init_timers_cpu(cpu);
1423 #ifdef CONFIG_HOTPLUG_CPU
1425 migrate_timers(cpu);
1434 static struct notifier_block __devinitdata timers_nb = {
1435 .notifier_call = timer_cpu_notify,
1439 void __init init_timers(void)
1441 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1442 (void *)(long)smp_processor_id());
1443 register_cpu_notifier(&timers_nb);
1444 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1447 #ifdef CONFIG_TIME_INTERPOLATION
1449 struct time_interpolator *time_interpolator;
1450 static struct time_interpolator *time_interpolator_list;
1451 static spinlock_t time_interpolator_lock = SPIN_LOCK_UNLOCKED;
1453 static inline unsigned long time_interpolator_get_cycles(unsigned int src)
1455 unsigned long (*x)(void);
1459 case TIME_SOURCE_FUNCTION:
1460 x = time_interpolator->addr;
1463 case TIME_SOURCE_MMIO64 :
1464 return readq(time_interpolator->addr);
1466 case TIME_SOURCE_MMIO32 :
1467 return readl(time_interpolator->addr);
1468 default: return get_cycles();
1472 static inline unsigned long time_interpolator_get_counter(void)
1474 unsigned int src = time_interpolator->source;
1476 if (time_interpolator->jitter)
1478 unsigned long lcycle;
1482 lcycle = time_interpolator->last_cycle;
1483 now = time_interpolator_get_cycles(src);
1484 if (lcycle && time_after(lcycle, now)) return lcycle;
1485 /* Keep track of the last timer value returned. The use of cmpxchg here
1486 * will cause contention in an SMP environment.
1488 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1492 return time_interpolator_get_cycles(src);
1495 void time_interpolator_reset(void)
1497 time_interpolator->offset = 0;
1498 time_interpolator->last_counter = time_interpolator_get_counter();
1501 unsigned long time_interpolator_resolution(void)
1503 if (time_interpolator->frequency < NSEC_PER_SEC)
1504 return NSEC_PER_SEC / time_interpolator->frequency;
1509 #define GET_TI_NSECS(count,i) ((((count) - i->last_counter) * i->nsec_per_cyc) >> i->shift)
1511 unsigned long time_interpolator_get_offset(void)
1513 return time_interpolator->offset +
1514 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator);
1517 static void time_interpolator_update(long delta_nsec)
1519 unsigned long counter = time_interpolator_get_counter();
1520 unsigned long offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
1522 /* The interpolator compensates for late ticks by accumulating
1523 * the late time in time_interpolator->offset. A tick earlier than
1524 * expected will lead to a reset of the offset and a corresponding
1525 * jump of the clock forward. Again this only works if the
1526 * interpolator clock is running slightly slower than the regular clock
1527 * and the tuning logic insures that.
1530 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1531 time_interpolator->offset = offset - delta_nsec;
1533 time_interpolator->skips++;
1534 time_interpolator->ns_skipped += delta_nsec - offset;
1535 time_interpolator->offset = 0;
1537 time_interpolator->last_counter = counter;
1539 /* Tuning logic for time interpolator invoked every minute or so.
1540 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1541 * Increase interpolator clock speed if we skip too much time.
1543 if (jiffies % INTERPOLATOR_ADJUST == 0)
1545 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1546 time_interpolator->nsec_per_cyc--;
1547 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1548 time_interpolator->nsec_per_cyc++;
1549 time_interpolator->skips = 0;
1550 time_interpolator->ns_skipped = 0;
1555 is_better_time_interpolator(struct time_interpolator *new)
1557 if (!time_interpolator)
1559 return new->frequency > 2*time_interpolator->frequency ||
1560 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1564 register_time_interpolator(struct time_interpolator *ti)
1566 unsigned long flags;
1568 ti->nsec_per_cyc = (NSEC_PER_SEC << ti->shift) / ti->frequency;
1569 spin_lock(&time_interpolator_lock);
1570 write_seqlock_irqsave(&xtime_lock, flags);
1571 if (is_better_time_interpolator(ti)) {
1572 time_interpolator = ti;
1573 time_interpolator_reset();
1575 write_sequnlock_irqrestore(&xtime_lock, flags);
1577 ti->next = time_interpolator_list;
1578 time_interpolator_list = ti;
1579 spin_unlock(&time_interpolator_lock);
1583 unregister_time_interpolator(struct time_interpolator *ti)
1585 struct time_interpolator *curr, **prev;
1586 unsigned long flags;
1588 spin_lock(&time_interpolator_lock);
1589 prev = &time_interpolator_list;
1590 for (curr = *prev; curr; curr = curr->next) {
1598 write_seqlock_irqsave(&xtime_lock, flags);
1599 if (ti == time_interpolator) {
1600 /* we lost the best time-interpolator: */
1601 time_interpolator = NULL;
1602 /* find the next-best interpolator */
1603 for (curr = time_interpolator_list; curr; curr = curr->next)
1604 if (is_better_time_interpolator(curr))
1605 time_interpolator = curr;
1606 time_interpolator_reset();
1608 write_sequnlock_irqrestore(&xtime_lock, flags);
1609 spin_unlock(&time_interpolator_lock);
1611 #endif /* CONFIG_TIME_INTERPOLATION */
1614 * msleep - sleep safely even with waitqueue interruptions
1615 * @msecs: Time in milliseconds to sleep for
1617 void msleep(unsigned int msecs)
1619 unsigned long timeout = msecs_to_jiffies(msecs);
1622 set_current_state(TASK_UNINTERRUPTIBLE);
1623 timeout = schedule_timeout(timeout);
1627 EXPORT_SYMBOL(msleep);
1630 * msleep_interruptible - sleep waiting for waitqueue interruptions
1631 * @msecs: Time in milliseconds to sleep for
1633 unsigned long msleep_interruptible(unsigned int msecs)
1635 unsigned long timeout = msecs_to_jiffies(msecs);
1637 while (timeout && !signal_pending(current)) {
1638 set_current_state(TASK_INTERRUPTIBLE);
1639 timeout = schedule_timeout(timeout);
1641 return jiffies_to_msecs(timeout);
1644 EXPORT_SYMBOL(msleep_interruptible);