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>
36 #include <linux/syscalls.h>
37 #include <linux/delay.h>
38 #include <linux/diskdump.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)
53 * per-CPU timer vector definitions:
57 #define TVN_SIZE (1 << TVN_BITS)
58 #define TVR_SIZE (1 << TVR_BITS)
59 #define TVN_MASK (TVN_SIZE - 1)
60 #define TVR_MASK (TVR_SIZE - 1)
62 typedef struct tvec_s {
63 struct list_head vec[TVN_SIZE];
66 typedef struct tvec_root_s {
67 struct list_head vec[TVR_SIZE];
70 struct tvec_t_base_s {
72 unsigned long timer_jiffies;
73 struct timer_list *running_timer;
79 } ____cacheline_aligned_in_smp;
81 typedef struct tvec_t_base_s tvec_base_t;
83 static inline void set_running_timer(tvec_base_t *base,
84 struct timer_list *timer)
87 base->running_timer = timer;
91 /* Fake initialization */
92 static DEFINE_PER_CPU(tvec_base_t, tvec_bases) = { SPIN_LOCK_UNLOCKED };
94 static void check_timer_failed(struct timer_list *timer)
96 static int whine_count;
97 if (whine_count < 16) {
99 printk("Uninitialised timer!\n");
100 printk("This is just a warning. Your computer is OK\n");
101 printk("function=0x%p, data=0x%lx\n",
102 timer->function, timer->data);
108 spin_lock_init(&timer->lock);
109 timer->magic = TIMER_MAGIC;
112 static inline void check_timer(struct timer_list *timer)
114 if (timer->magic != TIMER_MAGIC)
115 check_timer_failed(timer);
119 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
121 unsigned long expires = timer->expires;
122 unsigned long idx = expires - base->timer_jiffies;
123 struct list_head *vec;
125 if (idx < TVR_SIZE) {
126 int i = expires & TVR_MASK;
127 vec = base->tv1.vec + i;
128 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
129 int i = (expires >> TVR_BITS) & TVN_MASK;
130 vec = base->tv2.vec + i;
131 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
132 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
133 vec = base->tv3.vec + i;
134 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
135 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
136 vec = base->tv4.vec + i;
137 } else if ((signed long) idx < 0) {
139 * Can happen if you add a timer with expires == jiffies,
140 * or you set a timer to go off in the past
142 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
145 /* If the timeout is larger than 0xffffffff on 64-bit
146 * architectures then we use the maximum timeout:
148 if (idx > 0xffffffffUL) {
150 expires = idx + base->timer_jiffies;
152 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
153 vec = base->tv5.vec + i;
158 list_add_tail(&timer->entry, vec);
161 int __mod_timer(struct timer_list *timer, unsigned long expires)
163 tvec_base_t *old_base, *new_base;
167 BUG_ON(!timer->function);
171 spin_lock_irqsave(&timer->lock, flags);
172 new_base = &__get_cpu_var(tvec_bases);
174 old_base = timer->base;
177 * Prevent deadlocks via ordering by old_base < new_base.
179 if (old_base && (new_base != old_base)) {
180 if (old_base < new_base) {
181 spin_lock(&new_base->lock);
182 spin_lock(&old_base->lock);
184 spin_lock(&old_base->lock);
185 spin_lock(&new_base->lock);
188 * The timer base might have been cancelled while we were
189 * trying to take the lock(s):
191 if (timer->base != old_base) {
192 spin_unlock(&new_base->lock);
193 spin_unlock(&old_base->lock);
197 spin_lock(&new_base->lock);
198 if (timer->base != old_base) {
199 spin_unlock(&new_base->lock);
205 * Delete the previous timeout (if there was any), and install
209 list_del(&timer->entry);
212 timer->expires = expires;
213 internal_add_timer(new_base, timer);
214 timer->base = new_base;
216 if (old_base && (new_base != old_base))
217 spin_unlock(&old_base->lock);
218 spin_unlock(&new_base->lock);
219 spin_unlock_irqrestore(&timer->lock, flags);
224 EXPORT_SYMBOL(__mod_timer);
227 * add_timer_on - start a timer on a particular CPU
228 * @timer: the timer to be added
229 * @cpu: the CPU to start it on
231 * This is not very scalable on SMP. Double adds are not possible.
233 void add_timer_on(struct timer_list *timer, int cpu)
235 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
238 BUG_ON(timer_pending(timer) || !timer->function);
242 spin_lock_irqsave(&base->lock, flags);
243 internal_add_timer(base, timer);
245 spin_unlock_irqrestore(&base->lock, flags);
250 * mod_timer - modify a timer's timeout
251 * @timer: the timer to be modified
253 * mod_timer is a more efficient way to update the expire field of an
254 * active timer (if the timer is inactive it will be activated)
256 * mod_timer(timer, expires) is equivalent to:
258 * del_timer(timer); timer->expires = expires; add_timer(timer);
260 * Note that if there are multiple unserialized concurrent users of the
261 * same timer, then mod_timer() is the only safe way to modify the timeout,
262 * since add_timer() cannot modify an already running timer.
264 * The function returns whether it has modified a pending timer or not.
265 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
266 * active timer returns 1.)
268 int mod_timer(struct timer_list *timer, unsigned long expires)
270 BUG_ON(!timer->function);
275 * This is a common optimization triggered by the
276 * networking code - if the timer is re-modified
277 * to be the same thing then just return:
279 if (timer->expires == expires && timer_pending(timer))
282 return __mod_timer(timer, expires);
285 EXPORT_SYMBOL(mod_timer);
288 * del_timer - deactive a timer.
289 * @timer: the timer to be deactivated
291 * del_timer() deactivates a timer - this works on both active and inactive
294 * The function returns whether it has deactivated a pending timer or not.
295 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
296 * active timer returns 1.)
298 int del_timer(struct timer_list *timer)
309 spin_lock_irqsave(&base->lock, flags);
310 if (base != timer->base) {
311 spin_unlock_irqrestore(&base->lock, flags);
314 list_del(&timer->entry);
315 /* Need to make sure that anybody who sees a NULL base also sees the list ops */
318 spin_unlock_irqrestore(&base->lock, flags);
323 EXPORT_SYMBOL(del_timer);
327 * del_timer_sync - deactivate a timer and wait for the handler to finish.
328 * @timer: the timer to be deactivated
330 * This function only differs from del_timer() on SMP: besides deactivating
331 * the timer it also makes sure the handler has finished executing on other
334 * Synchronization rules: callers must prevent restarting of the timer,
335 * otherwise this function is meaningless. It must not be called from
336 * interrupt contexts. The caller must not hold locks which would prevent
337 * completion of the timer's handler. Upon exit the timer is not queued and
338 * the handler is not running on any CPU.
340 * The function returns whether it has deactivated a pending timer or not.
342 * del_timer_sync() is slow and complicated because it copes with timer
343 * handlers which re-arm the timer (periodic timers). If the timer handler
344 * is known to not do this (a single shot timer) then use
345 * del_singleshot_timer_sync() instead.
347 int del_timer_sync(struct timer_list *timer)
355 ret += del_timer(timer);
357 for_each_online_cpu(i) {
358 base = &per_cpu(tvec_bases, i);
359 if (base->running_timer == timer) {
360 while (base->running_timer == timer) {
362 preempt_check_resched();
368 if (timer_pending(timer))
373 EXPORT_SYMBOL(del_timer_sync);
376 * del_singleshot_timer_sync - deactivate a non-recursive timer
377 * @timer: the timer to be deactivated
379 * This function is an optimization of del_timer_sync for the case where the
380 * caller can guarantee the timer does not reschedule itself in its timer
383 * Synchronization rules: callers must prevent restarting of the timer,
384 * otherwise this function is meaningless. It must not be called from
385 * interrupt contexts. The caller must not hold locks which wold prevent
386 * completion of the timer's handler. Upon exit the timer is not queued and
387 * the handler is not running on any CPU.
389 * The function returns whether it has deactivated a pending timer or not.
391 int del_singleshot_timer_sync(struct timer_list *timer)
393 int ret = del_timer(timer);
396 ret = del_timer_sync(timer);
402 EXPORT_SYMBOL(del_singleshot_timer_sync);
405 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
407 /* cascade all the timers from tv up one level */
408 struct list_head *head, *curr;
410 head = tv->vec + index;
413 * We are removing _all_ timers from the list, so we don't have to
414 * detach them individually, just clear the list afterwards.
416 while (curr != head) {
417 struct timer_list *tmp;
419 tmp = list_entry(curr, struct timer_list, entry);
420 BUG_ON(tmp->base != base);
422 internal_add_timer(base, tmp);
424 INIT_LIST_HEAD(head);
430 * __run_timers - run all expired timers (if any) on this CPU.
431 * @base: the timer vector to be processed.
433 * This function cascades all vectors and executes all expired timer
436 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
438 static inline void __run_timers(tvec_base_t *base)
440 struct timer_list *timer;
443 spin_lock_irqsave(&base->lock, flags);
444 while (time_after_eq(jiffies, base->timer_jiffies)) {
445 struct list_head work_list = LIST_HEAD_INIT(work_list);
446 struct list_head *head = &work_list;
447 int index = base->timer_jiffies & TVR_MASK;
453 (!cascade(base, &base->tv2, INDEX(0))) &&
454 (!cascade(base, &base->tv3, INDEX(1))) &&
455 !cascade(base, &base->tv4, INDEX(2)))
456 cascade(base, &base->tv5, INDEX(3));
457 ++base->timer_jiffies;
458 list_splice_init(base->tv1.vec + index, &work_list);
460 if (!list_empty(head)) {
461 void (*fn)(unsigned long);
464 timer = list_entry(head->next,struct timer_list,entry);
465 fn = timer->function;
468 list_del(&timer->entry);
469 set_running_timer(base, timer);
472 spin_unlock_irqrestore(&base->lock, flags);
474 spin_lock_irq(&base->lock);
478 set_running_timer(base, NULL);
479 spin_unlock_irqrestore(&base->lock, flags);
482 #ifdef CONFIG_NO_IDLE_HZ
484 * Find out when the next timer event is due to happen. This
485 * is used on S/390 to stop all activity when a cpus is idle.
486 * This functions needs to be called disabled.
488 unsigned long next_timer_interrupt(void)
491 struct list_head *list;
492 struct timer_list *nte;
493 unsigned long expires;
497 base = &__get_cpu_var(tvec_bases);
498 spin_lock(&base->lock);
499 expires = base->timer_jiffies + (LONG_MAX >> 1);
502 /* Look for timer events in tv1. */
503 j = base->timer_jiffies & TVR_MASK;
505 list_for_each_entry(nte, base->tv1.vec + j, entry) {
506 expires = nte->expires;
507 if (j < (base->timer_jiffies & TVR_MASK))
508 list = base->tv2.vec + (INDEX(0));
511 j = (j + 1) & TVR_MASK;
512 } while (j != (base->timer_jiffies & TVR_MASK));
515 varray[0] = &base->tv2;
516 varray[1] = &base->tv3;
517 varray[2] = &base->tv4;
518 varray[3] = &base->tv5;
519 for (i = 0; i < 4; i++) {
522 if (list_empty(varray[i]->vec + j)) {
523 j = (j + 1) & TVN_MASK;
526 list_for_each_entry(nte, varray[i]->vec + j, entry)
527 if (time_before(nte->expires, expires))
528 expires = nte->expires;
529 if (j < (INDEX(i)) && i < 3)
530 list = varray[i + 1]->vec + (INDEX(i + 1));
532 } while (j != (INDEX(i)));
537 * The search wrapped. We need to look at the next list
538 * from next tv element that would cascade into tv element
539 * where we found the timer element.
541 list_for_each_entry(nte, list, entry) {
542 if (time_before(nte->expires, expires))
543 expires = nte->expires;
546 spin_unlock(&base->lock);
551 /******************************************************************/
554 * Timekeeping variables
556 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
557 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
561 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
562 * for sub jiffie times) to get to monotonic time. Monotonic is pegged at zero
563 * at zero at system boot time, so wall_to_monotonic will be negative,
564 * however, we will ALWAYS keep the tv_nsec part positive so we can use
565 * the usual normalization.
567 struct timespec xtime __attribute__ ((aligned (16)));
568 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
570 EXPORT_SYMBOL(xtime);
572 /* Don't completely fail for HZ > 500. */
573 int tickadj = 500/HZ ? : 1; /* microsecs */
577 * phase-lock loop variables
579 /* TIME_ERROR prevents overwriting the CMOS clock */
580 int time_state = TIME_OK; /* clock synchronization status */
581 int time_status = STA_UNSYNC; /* clock status bits */
582 long time_offset; /* time adjustment (us) */
583 long time_constant = 2; /* pll time constant */
584 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
585 long time_precision = 1; /* clock precision (us) */
586 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
587 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
588 long time_phase; /* phase offset (scaled us) */
589 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
590 /* frequency offset (scaled ppm)*/
591 long time_adj; /* tick adjust (scaled 1 / HZ) */
592 long time_reftime; /* time at last adjustment (s) */
594 long time_next_adjust;
597 * this routine handles the overflow of the microsecond field
599 * The tricky bits of code to handle the accurate clock support
600 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
601 * They were originally developed for SUN and DEC kernels.
602 * All the kudos should go to Dave for this stuff.
605 static void second_overflow(void)
609 /* Bump the maxerror field */
610 time_maxerror += time_tolerance >> SHIFT_USEC;
611 if ( time_maxerror > NTP_PHASE_LIMIT ) {
612 time_maxerror = NTP_PHASE_LIMIT;
613 time_status |= STA_UNSYNC;
617 * Leap second processing. If in leap-insert state at
618 * the end of the day, the system clock is set back one
619 * second; if in leap-delete state, the system clock is
620 * set ahead one second. The microtime() routine or
621 * external clock driver will insure that reported time
622 * is always monotonic. The ugly divides should be
625 switch (time_state) {
628 if (time_status & STA_INS)
629 time_state = TIME_INS;
630 else if (time_status & STA_DEL)
631 time_state = TIME_DEL;
635 if (xtime.tv_sec % 86400 == 0) {
637 wall_to_monotonic.tv_sec++;
638 /* The timer interpolator will make time change gradually instead
639 * of an immediate jump by one second.
641 time_interpolator_update(-NSEC_PER_SEC);
642 time_state = TIME_OOP;
644 printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
649 if ((xtime.tv_sec + 1) % 86400 == 0) {
651 wall_to_monotonic.tv_sec--;
652 /* Use of time interpolator for a gradual change of time */
653 time_interpolator_update(NSEC_PER_SEC);
654 time_state = TIME_WAIT;
656 printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
661 time_state = TIME_WAIT;
665 if (!(time_status & (STA_INS | STA_DEL)))
666 time_state = TIME_OK;
670 * Compute the phase adjustment for the next second. In
671 * PLL mode, the offset is reduced by a fixed factor
672 * times the time constant. In FLL mode the offset is
673 * used directly. In either mode, the maximum phase
674 * adjustment for each second is clamped so as to spread
675 * the adjustment over not more than the number of
676 * seconds between updates.
678 if (time_offset < 0) {
679 ltemp = -time_offset;
680 if (!(time_status & STA_FLL))
681 ltemp >>= SHIFT_KG + time_constant;
682 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
683 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
684 time_offset += ltemp;
685 time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
688 if (!(time_status & STA_FLL))
689 ltemp >>= SHIFT_KG + time_constant;
690 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
691 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
692 time_offset -= ltemp;
693 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
697 * Compute the frequency estimate and additional phase
698 * adjustment due to frequency error for the next
699 * second. When the PPS signal is engaged, gnaw on the
700 * watchdog counter and update the frequency computed by
701 * the pll and the PPS signal.
704 if (pps_valid == PPS_VALID) { /* PPS signal lost */
705 pps_jitter = MAXTIME;
706 pps_stabil = MAXFREQ;
707 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
708 STA_PPSWANDER | STA_PPSERROR);
710 ltemp = time_freq + pps_freq;
712 time_adj -= -ltemp >>
713 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
716 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
719 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
720 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
723 time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
725 time_adj += (time_adj >> 2) + (time_adj >> 5);
728 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
729 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
732 time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
734 time_adj += (time_adj >> 6) + (time_adj >> 7);
738 /* in the NTP reference this is called "hardclock()" */
739 static void update_wall_time_one_tick(void)
741 long time_adjust_step, delta_nsec;
743 if ( (time_adjust_step = time_adjust) != 0 ) {
744 /* We are doing an adjtime thing.
746 * Prepare time_adjust_step to be within bounds.
747 * Note that a positive time_adjust means we want the clock
750 * Limit the amount of the step to be in the range
751 * -tickadj .. +tickadj
753 if (time_adjust > tickadj)
754 time_adjust_step = tickadj;
755 else if (time_adjust < -tickadj)
756 time_adjust_step = -tickadj;
758 /* Reduce by this step the amount of time left */
759 time_adjust -= time_adjust_step;
761 delta_nsec = tick_nsec + time_adjust_step * 1000;
763 * Advance the phase, once it gets to one microsecond, then
764 * advance the tick more.
766 time_phase += time_adj;
767 if (time_phase <= -FINENSEC) {
768 long ltemp = -time_phase >> (SHIFT_SCALE - 10);
769 time_phase += ltemp << (SHIFT_SCALE - 10);
772 else if (time_phase >= FINENSEC) {
773 long ltemp = time_phase >> (SHIFT_SCALE - 10);
774 time_phase -= ltemp << (SHIFT_SCALE - 10);
777 xtime.tv_nsec += delta_nsec;
778 time_interpolator_update(delta_nsec);
780 /* Changes by adjtime() do not take effect till next tick. */
781 if (time_next_adjust != 0) {
782 time_adjust = time_next_adjust;
783 time_next_adjust = 0;
788 * Using a loop looks inefficient, but "ticks" is
789 * usually just one (we shouldn't be losing ticks,
790 * we're doing this this way mainly for interrupt
791 * latency reasons, not because we think we'll
792 * have lots of lost timer ticks
794 static void update_wall_time(unsigned long ticks)
798 update_wall_time_one_tick();
799 if (xtime.tv_nsec >= 1000000000) {
800 xtime.tv_nsec -= 1000000000;
807 static inline void do_process_times(struct task_struct *p,
808 unsigned long user, unsigned long system)
812 psecs = (p->utime += user);
813 psecs += (p->stime += system);
814 if (p->signal && !unlikely(p->state & (EXIT_DEAD|EXIT_ZOMBIE)) &&
815 psecs / HZ >= p->signal->rlim[RLIMIT_CPU].rlim_cur) {
816 /* Send SIGXCPU every second.. */
818 send_sig(SIGXCPU, p, 1);
819 /* and SIGKILL when we go over max.. */
820 if (psecs / HZ >= p->signal->rlim[RLIMIT_CPU].rlim_max)
821 send_sig(SIGKILL, p, 1);
825 static inline void do_it_virt(struct task_struct * p, unsigned long ticks)
827 unsigned long it_virt = p->it_virt_value;
832 it_virt = p->it_virt_incr;
833 send_sig(SIGVTALRM, p, 1);
835 p->it_virt_value = it_virt;
839 static inline void do_it_prof(struct task_struct *p)
841 unsigned long it_prof = p->it_prof_value;
844 if (--it_prof == 0) {
845 it_prof = p->it_prof_incr;
846 send_sig(SIGPROF, p, 1);
848 p->it_prof_value = it_prof;
852 static void update_one_process(struct task_struct *p, unsigned long user,
853 unsigned long system, int cpu)
855 do_process_times(p, user, system);
861 * Called from the timer interrupt handler to charge one tick to the current
862 * process. user_tick is 1 if the tick is user time, 0 for system.
864 void update_process_times(int user_tick)
866 struct task_struct *p = current;
867 int cpu = smp_processor_id(), system = user_tick ^ 1;
869 update_one_process(p, user_tick, system, cpu);
871 scheduler_tick(user_tick, system);
875 * Nr of active tasks - counted in fixed-point numbers
877 static unsigned long count_active_tasks(void)
879 return (nr_running() + nr_uninterruptible()) * FIXED_1;
883 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
884 * imply that avenrun[] is the standard name for this kind of thing.
885 * Nothing else seems to be standardized: the fractional size etc
886 * all seem to differ on different machines.
888 * Requires xtime_lock to access.
890 unsigned long avenrun[3];
893 * calc_load - given tick count, update the avenrun load estimates.
894 * This is called while holding a write_lock on xtime_lock.
896 static inline void calc_load(unsigned long ticks)
898 unsigned long active_tasks; /* fixed-point */
899 static int count = LOAD_FREQ;
904 active_tasks = count_active_tasks();
905 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
906 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
907 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
911 /* jiffies at the most recent update of wall time */
912 unsigned long wall_jiffies = INITIAL_JIFFIES;
915 * This read-write spinlock protects us from races in SMP while
916 * playing with xtime and avenrun.
918 #ifndef ARCH_HAVE_XTIME_LOCK
919 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
921 EXPORT_SYMBOL(xtime_lock);
925 * This function runs timers and the timer-tq in bottom half context.
927 static void run_timer_softirq(struct softirq_action *h)
929 tvec_base_t *base = &__get_cpu_var(tvec_bases);
931 if (time_after_eq(jiffies, base->timer_jiffies))
936 * Called by the local, per-CPU timer interrupt on SMP.
938 void run_local_timers(void)
940 raise_softirq(TIMER_SOFTIRQ);
944 * Called by the timer interrupt. xtime_lock must already be taken
947 static inline void update_times(void)
951 ticks = jiffies - wall_jiffies;
953 wall_jiffies += ticks;
954 update_wall_time(ticks);
960 * The 64-bit jiffies value is not atomic - you MUST NOT read it
961 * without sampling the sequence number in xtime_lock.
962 * jiffies is defined in the linker script...
965 void do_timer(struct pt_regs *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;
1123 if (crashdump_mode()) {
1124 diskdump_mdelay(timeout);
1125 set_current_state(TASK_RUNNING);
1131 case MAX_SCHEDULE_TIMEOUT:
1133 * These two special cases are useful to be comfortable
1134 * in the caller. Nothing more. We could take
1135 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1136 * but I' d like to return a valid offset (>=0) to allow
1137 * the caller to do everything it want with the retval.
1143 * Another bit of PARANOID. Note that the retval will be
1144 * 0 since no piece of kernel is supposed to do a check
1145 * for a negative retval of schedule_timeout() (since it
1146 * should never happens anyway). You just have the printk()
1147 * that will tell you if something is gone wrong and where.
1151 printk(KERN_ERR "schedule_timeout: wrong timeout "
1152 "value %lx from %p\n", timeout,
1153 __builtin_return_address(0));
1154 current->state = TASK_RUNNING;
1159 expire = timeout + jiffies;
1162 timer.expires = expire;
1163 timer.data = (unsigned long) current;
1164 timer.function = process_timeout;
1168 del_singleshot_timer_sync(&timer);
1170 timeout = expire - jiffies;
1173 return timeout < 0 ? 0 : timeout;
1176 EXPORT_SYMBOL(schedule_timeout);
1178 /* Thread ID - the internal kernel "pid" */
1179 asmlinkage long sys_gettid(void)
1181 return current->pid;
1184 static long __sched nanosleep_restart(struct restart_block *restart)
1186 unsigned long expire = restart->arg0, now = jiffies;
1187 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1190 /* Did it expire while we handled signals? */
1191 if (!time_after(expire, now))
1194 current->state = TASK_INTERRUPTIBLE;
1195 expire = schedule_timeout(expire - now);
1200 jiffies_to_timespec(expire, &t);
1202 ret = -ERESTART_RESTARTBLOCK;
1203 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1205 /* The 'restart' block is already filled in */
1210 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1213 unsigned long expire;
1216 if (copy_from_user(&t, rqtp, sizeof(t)))
1219 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1222 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1223 current->state = TASK_INTERRUPTIBLE;
1224 expire = schedule_timeout(expire);
1228 struct restart_block *restart;
1229 jiffies_to_timespec(expire, &t);
1230 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1233 restart = ¤t_thread_info()->restart_block;
1234 restart->fn = nanosleep_restart;
1235 restart->arg0 = jiffies + expire;
1236 restart->arg1 = (unsigned long) rmtp;
1237 ret = -ERESTART_RESTARTBLOCK;
1243 * sys_sysinfo - fill in sysinfo struct
1245 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1248 unsigned long mem_total, sav_total;
1249 unsigned int mem_unit, bitcount;
1252 memset((char *)&val, 0, sizeof(struct sysinfo));
1256 seq = read_seqbegin(&xtime_lock);
1259 * This is annoying. The below is the same thing
1260 * posix_get_clock_monotonic() does, but it wants to
1261 * take the lock which we want to cover the loads stuff
1265 getnstimeofday(&tp);
1266 tp.tv_sec += wall_to_monotonic.tv_sec;
1267 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1268 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1269 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1272 if (vx_flags(VXF_VIRT_UPTIME, 0))
1273 vx_vsi_uptime(&tp, NULL);
1274 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1276 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1277 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1278 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1280 val.procs = nr_threads;
1281 } while (read_seqretry(&xtime_lock, seq));
1283 /* if (vx_flags(VXF_VIRT_CPU, 0))
1290 * If the sum of all the available memory (i.e. ram + swap)
1291 * is less than can be stored in a 32 bit unsigned long then
1292 * we can be binary compatible with 2.2.x kernels. If not,
1293 * well, in that case 2.2.x was broken anyways...
1295 * -Erik Andersen <andersee@debian.org>
1298 mem_total = val.totalram + val.totalswap;
1299 if (mem_total < val.totalram || mem_total < val.totalswap)
1302 mem_unit = val.mem_unit;
1303 while (mem_unit > 1) {
1306 sav_total = mem_total;
1308 if (mem_total < sav_total)
1313 * If mem_total did not overflow, multiply all memory values by
1314 * val.mem_unit and set it to 1. This leaves things compatible
1315 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1320 val.totalram <<= bitcount;
1321 val.freeram <<= bitcount;
1322 val.sharedram <<= bitcount;
1323 val.bufferram <<= bitcount;
1324 val.totalswap <<= bitcount;
1325 val.freeswap <<= bitcount;
1326 val.totalhigh <<= bitcount;
1327 val.freehigh <<= bitcount;
1330 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1336 static void /* __devinit */ init_timers_cpu(int cpu)
1341 base = &per_cpu(tvec_bases, cpu);
1342 spin_lock_init(&base->lock);
1343 for (j = 0; j < TVN_SIZE; j++) {
1344 INIT_LIST_HEAD(base->tv5.vec + j);
1345 INIT_LIST_HEAD(base->tv4.vec + j);
1346 INIT_LIST_HEAD(base->tv3.vec + j);
1347 INIT_LIST_HEAD(base->tv2.vec + j);
1349 for (j = 0; j < TVR_SIZE; j++)
1350 INIT_LIST_HEAD(base->tv1.vec + j);
1352 base->timer_jiffies = jiffies;
1355 static tvec_base_t saved_tvec_base;
1357 void dump_clear_timers(void)
1359 tvec_base_t *base = &per_cpu(tvec_bases, smp_processor_id());
1361 memcpy(&saved_tvec_base, base, sizeof(saved_tvec_base));
1362 init_timers_cpu(smp_processor_id());
1365 EXPORT_SYMBOL_GPL(dump_clear_timers);
1367 void dump_run_timers(void)
1369 tvec_base_t *base = &__get_cpu_var(tvec_bases);
1374 EXPORT_SYMBOL_GPL(dump_run_timers);
1376 #ifdef CONFIG_HOTPLUG_CPU
1377 static int migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1379 struct timer_list *timer;
1381 while (!list_empty(head)) {
1382 timer = list_entry(head->next, struct timer_list, entry);
1383 /* We're locking backwards from __mod_timer order here,
1385 if (!spin_trylock(&timer->lock))
1387 list_del(&timer->entry);
1388 internal_add_timer(new_base, timer);
1389 timer->base = new_base;
1390 spin_unlock(&timer->lock);
1395 static void __devinit migrate_timers(int cpu)
1397 tvec_base_t *old_base;
1398 tvec_base_t *new_base;
1401 BUG_ON(cpu_online(cpu));
1402 old_base = &per_cpu(tvec_bases, cpu);
1403 new_base = &get_cpu_var(tvec_bases);
1405 local_irq_disable();
1407 /* Prevent deadlocks via ordering by old_base < new_base. */
1408 if (old_base < new_base) {
1409 spin_lock(&new_base->lock);
1410 spin_lock(&old_base->lock);
1412 spin_lock(&old_base->lock);
1413 spin_lock(&new_base->lock);
1416 if (old_base->running_timer)
1418 for (i = 0; i < TVR_SIZE; i++)
1419 if (!migrate_timer_list(new_base, old_base->tv1.vec + i))
1421 for (i = 0; i < TVN_SIZE; i++)
1422 if (!migrate_timer_list(new_base, old_base->tv2.vec + i)
1423 || !migrate_timer_list(new_base, old_base->tv3.vec + i)
1424 || !migrate_timer_list(new_base, old_base->tv4.vec + i)
1425 || !migrate_timer_list(new_base, old_base->tv5.vec + i))
1427 spin_unlock(&old_base->lock);
1428 spin_unlock(&new_base->lock);
1430 put_cpu_var(tvec_bases);
1434 /* Avoid deadlock with __mod_timer, by backing off. */
1435 spin_unlock(&old_base->lock);
1436 spin_unlock(&new_base->lock);
1440 #endif /* CONFIG_HOTPLUG_CPU */
1442 static int __devinit timer_cpu_notify(struct notifier_block *self,
1443 unsigned long action, void *hcpu)
1445 long cpu = (long)hcpu;
1447 case CPU_UP_PREPARE:
1448 init_timers_cpu(cpu);
1450 #ifdef CONFIG_HOTPLUG_CPU
1452 migrate_timers(cpu);
1461 static struct notifier_block __devinitdata timers_nb = {
1462 .notifier_call = timer_cpu_notify,
1466 void __init init_timers(void)
1468 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1469 (void *)(long)smp_processor_id());
1470 register_cpu_notifier(&timers_nb);
1471 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1474 #ifdef CONFIG_TIME_INTERPOLATION
1476 struct time_interpolator *time_interpolator;
1477 static struct time_interpolator *time_interpolator_list;
1478 static spinlock_t time_interpolator_lock = SPIN_LOCK_UNLOCKED;
1480 static inline u64 time_interpolator_get_cycles(unsigned int src)
1482 unsigned long (*x)(void);
1486 case TIME_SOURCE_FUNCTION:
1487 x = time_interpolator->addr;
1490 case TIME_SOURCE_MMIO64 :
1491 return readq(time_interpolator->addr);
1493 case TIME_SOURCE_MMIO32 :
1494 return readl(time_interpolator->addr);
1496 default: return get_cycles();
1500 static inline u64 time_interpolator_get_counter(void)
1502 unsigned int src = time_interpolator->source;
1504 if (time_interpolator->jitter)
1510 lcycle = time_interpolator->last_cycle;
1511 now = time_interpolator_get_cycles(src);
1512 if (lcycle && time_after(lcycle, now))
1514 /* Keep track of the last timer value returned. The use of cmpxchg here
1515 * will cause contention in an SMP environment.
1517 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1521 return time_interpolator_get_cycles(src);
1524 void time_interpolator_reset(void)
1526 time_interpolator->offset = 0;
1527 time_interpolator->last_counter = time_interpolator_get_counter();
1530 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1532 unsigned long time_interpolator_get_offset(void)
1534 /* If we do not have a time interpolator set up then just return zero */
1535 if (!time_interpolator)
1538 return time_interpolator->offset +
1539 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator);
1542 #define INTERPOLATOR_ADJUST 65536
1543 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1545 static void time_interpolator_update(long delta_nsec)
1548 unsigned long offset;
1550 /* If there is no time interpolator set up then do nothing */
1551 if (!time_interpolator)
1554 /* The interpolator compensates for late ticks by accumulating
1555 * the late time in time_interpolator->offset. A tick earlier than
1556 * expected will lead to a reset of the offset and a corresponding
1557 * jump of the clock forward. Again this only works if the
1558 * interpolator clock is running slightly slower than the regular clock
1559 * and the tuning logic insures that.
1562 counter = time_interpolator_get_counter();
1563 offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
1565 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1566 time_interpolator->offset = offset - delta_nsec;
1568 time_interpolator->skips++;
1569 time_interpolator->ns_skipped += delta_nsec - offset;
1570 time_interpolator->offset = 0;
1572 time_interpolator->last_counter = counter;
1574 /* Tuning logic for time interpolator invoked every minute or so.
1575 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1576 * Increase interpolator clock speed if we skip too much time.
1578 if (jiffies % INTERPOLATOR_ADJUST == 0)
1580 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1581 time_interpolator->nsec_per_cyc--;
1582 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1583 time_interpolator->nsec_per_cyc++;
1584 time_interpolator->skips = 0;
1585 time_interpolator->ns_skipped = 0;
1590 is_better_time_interpolator(struct time_interpolator *new)
1592 if (!time_interpolator)
1594 return new->frequency > 2*time_interpolator->frequency ||
1595 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1599 register_time_interpolator(struct time_interpolator *ti)
1601 unsigned long flags;
1604 if (ti->frequency == 0 || ti->mask == 0)
1607 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1608 spin_lock(&time_interpolator_lock);
1609 write_seqlock_irqsave(&xtime_lock, flags);
1610 if (is_better_time_interpolator(ti)) {
1611 time_interpolator = ti;
1612 time_interpolator_reset();
1614 write_sequnlock_irqrestore(&xtime_lock, flags);
1616 ti->next = time_interpolator_list;
1617 time_interpolator_list = ti;
1618 spin_unlock(&time_interpolator_lock);
1622 unregister_time_interpolator(struct time_interpolator *ti)
1624 struct time_interpolator *curr, **prev;
1625 unsigned long flags;
1627 spin_lock(&time_interpolator_lock);
1628 prev = &time_interpolator_list;
1629 for (curr = *prev; curr; curr = curr->next) {
1637 write_seqlock_irqsave(&xtime_lock, flags);
1638 if (ti == time_interpolator) {
1639 /* we lost the best time-interpolator: */
1640 time_interpolator = NULL;
1641 /* find the next-best interpolator */
1642 for (curr = time_interpolator_list; curr; curr = curr->next)
1643 if (is_better_time_interpolator(curr))
1644 time_interpolator = curr;
1645 time_interpolator_reset();
1647 write_sequnlock_irqrestore(&xtime_lock, flags);
1648 spin_unlock(&time_interpolator_lock);
1650 #endif /* CONFIG_TIME_INTERPOLATION */
1653 * msleep - sleep safely even with waitqueue interruptions
1654 * @msecs: Time in milliseconds to sleep for
1656 void msleep(unsigned int msecs)
1658 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1660 if (unlikely(crashdump_mode())) {
1661 while (msecs--) udelay(1000);
1666 set_current_state(TASK_UNINTERRUPTIBLE);
1667 timeout = schedule_timeout(timeout);
1671 EXPORT_SYMBOL(msleep);
1674 * msleep_interruptible - sleep waiting for waitqueue interruptions
1675 * @msecs: Time in milliseconds to sleep for
1677 unsigned long msleep_interruptible(unsigned int msecs)
1679 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1681 while (timeout && !signal_pending(current)) {
1682 set_current_state(TASK_INTERRUPTIBLE);
1683 timeout = schedule_timeout(timeout);
1685 return jiffies_to_msecs(timeout);
1688 EXPORT_SYMBOL(msleep_interruptible);