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_GPL(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);
313 smp_wmb(); /* the list del must have taken effect before timer->base
314 * change is visible to other CPUs, or a concurrent mod_timer
315 * would cause a race with list_add
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;
442 spin_lock_irq(&base->lock);
443 while (time_after_eq(jiffies, base->timer_jiffies)) {
444 struct list_head work_list = LIST_HEAD_INIT(work_list);
445 struct list_head *head = &work_list;
446 int index = base->timer_jiffies & TVR_MASK;
452 (!cascade(base, &base->tv2, INDEX(0))) &&
453 (!cascade(base, &base->tv3, INDEX(1))) &&
454 !cascade(base, &base->tv4, INDEX(2)))
455 cascade(base, &base->tv5, INDEX(3));
456 ++base->timer_jiffies;
457 list_splice_init(base->tv1.vec + index, &work_list);
459 if (!list_empty(head)) {
460 void (*fn)(unsigned long);
463 timer = list_entry(head->next,struct timer_list,entry);
464 fn = timer->function;
467 list_del(&timer->entry);
468 set_running_timer(base, timer);
469 smp_wmb(); /* the list del must have taken effect before timer->base
470 * change is visible to other CPUs, or a concurrent mod_timer
471 * would cause a race with list_add
474 spin_unlock_irq(&base->lock);
476 spin_lock_irq(&base->lock);
480 set_running_timer(base, NULL);
481 spin_unlock_irq(&base->lock);
484 #ifdef CONFIG_NO_IDLE_HZ
486 * Find out when the next timer event is due to happen. This
487 * is used on S/390 to stop all activity when a cpus is idle.
488 * This functions needs to be called disabled.
490 unsigned long next_timer_interrupt(void)
493 struct list_head *list;
494 struct timer_list *nte;
495 unsigned long expires;
499 base = &__get_cpu_var(tvec_bases);
500 spin_lock(&base->lock);
501 expires = base->timer_jiffies + (LONG_MAX >> 1);
504 /* Look for timer events in tv1. */
505 j = base->timer_jiffies & TVR_MASK;
507 list_for_each_entry(nte, base->tv1.vec + j, entry) {
508 expires = nte->expires;
509 if (j < (base->timer_jiffies & TVR_MASK))
510 list = base->tv2.vec + (INDEX(0));
513 j = (j + 1) & TVR_MASK;
514 } while (j != (base->timer_jiffies & TVR_MASK));
517 varray[0] = &base->tv2;
518 varray[1] = &base->tv3;
519 varray[2] = &base->tv4;
520 varray[3] = &base->tv5;
521 for (i = 0; i < 4; i++) {
524 if (list_empty(varray[i]->vec + j)) {
525 j = (j + 1) & TVN_MASK;
528 list_for_each_entry(nte, varray[i]->vec + j, entry)
529 if (time_before(nte->expires, expires))
530 expires = nte->expires;
531 if (j < (INDEX(i)) && i < 3)
532 list = varray[i + 1]->vec + (INDEX(i + 1));
534 } while (j != (INDEX(i)));
539 * The search wrapped. We need to look at the next list
540 * from next tv element that would cascade into tv element
541 * where we found the timer element.
543 list_for_each_entry(nte, list, entry) {
544 if (time_before(nte->expires, expires))
545 expires = nte->expires;
548 spin_unlock(&base->lock);
553 /******************************************************************/
556 * Timekeeping variables
558 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
559 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
563 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
564 * for sub jiffie times) to get to monotonic time. Monotonic is pegged at zero
565 * at zero at system boot time, so wall_to_monotonic will be negative,
566 * however, we will ALWAYS keep the tv_nsec part positive so we can use
567 * the usual normalization.
569 struct timespec xtime __attribute__ ((aligned (16)));
570 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
572 EXPORT_SYMBOL(xtime);
574 /* Don't completely fail for HZ > 500. */
575 int tickadj = 500/HZ ? : 1; /* microsecs */
579 * phase-lock loop variables
581 /* TIME_ERROR prevents overwriting the CMOS clock */
582 int time_state = TIME_OK; /* clock synchronization status */
583 int time_status = STA_UNSYNC; /* clock status bits */
584 long time_offset; /* time adjustment (us) */
585 long time_constant = 2; /* pll time constant */
586 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
587 long time_precision = 1; /* clock precision (us) */
588 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
589 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
590 long time_phase; /* phase offset (scaled us) */
591 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
592 /* frequency offset (scaled ppm)*/
593 long time_adj; /* tick adjust (scaled 1 / HZ) */
594 long time_reftime; /* time at last adjustment (s) */
596 long time_next_adjust;
599 * this routine handles the overflow of the microsecond field
601 * The tricky bits of code to handle the accurate clock support
602 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
603 * They were originally developed for SUN and DEC kernels.
604 * All the kudos should go to Dave for this stuff.
607 static void second_overflow(void)
611 /* Bump the maxerror field */
612 time_maxerror += time_tolerance >> SHIFT_USEC;
613 if ( time_maxerror > NTP_PHASE_LIMIT ) {
614 time_maxerror = NTP_PHASE_LIMIT;
615 time_status |= STA_UNSYNC;
619 * Leap second processing. If in leap-insert state at
620 * the end of the day, the system clock is set back one
621 * second; if in leap-delete state, the system clock is
622 * set ahead one second. The microtime() routine or
623 * external clock driver will insure that reported time
624 * is always monotonic. The ugly divides should be
627 switch (time_state) {
630 if (time_status & STA_INS)
631 time_state = TIME_INS;
632 else if (time_status & STA_DEL)
633 time_state = TIME_DEL;
637 if (xtime.tv_sec % 86400 == 0) {
639 wall_to_monotonic.tv_sec++;
640 /* The timer interpolator will make time change gradually instead
641 * of an immediate jump by one second.
643 time_interpolator_update(-NSEC_PER_SEC);
644 time_state = TIME_OOP;
646 printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
651 if ((xtime.tv_sec + 1) % 86400 == 0) {
653 wall_to_monotonic.tv_sec--;
654 /* Use of time interpolator for a gradual change of time */
655 time_interpolator_update(NSEC_PER_SEC);
656 time_state = TIME_WAIT;
658 printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
663 time_state = TIME_WAIT;
667 if (!(time_status & (STA_INS | STA_DEL)))
668 time_state = TIME_OK;
672 * Compute the phase adjustment for the next second. In
673 * PLL mode, the offset is reduced by a fixed factor
674 * times the time constant. In FLL mode the offset is
675 * used directly. In either mode, the maximum phase
676 * adjustment for each second is clamped so as to spread
677 * the adjustment over not more than the number of
678 * seconds between updates.
680 if (time_offset < 0) {
681 ltemp = -time_offset;
682 if (!(time_status & STA_FLL))
683 ltemp >>= SHIFT_KG + time_constant;
684 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
685 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
686 time_offset += ltemp;
687 time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
690 if (!(time_status & STA_FLL))
691 ltemp >>= SHIFT_KG + time_constant;
692 if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
693 ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
694 time_offset -= ltemp;
695 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
699 * Compute the frequency estimate and additional phase
700 * adjustment due to frequency error for the next
701 * second. When the PPS signal is engaged, gnaw on the
702 * watchdog counter and update the frequency computed by
703 * the pll and the PPS signal.
706 if (pps_valid == PPS_VALID) { /* PPS signal lost */
707 pps_jitter = MAXTIME;
708 pps_stabil = MAXFREQ;
709 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
710 STA_PPSWANDER | STA_PPSERROR);
712 ltemp = time_freq + pps_freq;
714 time_adj -= -ltemp >>
715 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
718 (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
721 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
722 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
725 time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
727 time_adj += (time_adj >> 2) + (time_adj >> 5);
730 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
731 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
734 time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
736 time_adj += (time_adj >> 6) + (time_adj >> 7);
740 /* in the NTP reference this is called "hardclock()" */
741 static void update_wall_time_one_tick(void)
743 long time_adjust_step, delta_nsec;
745 if ( (time_adjust_step = time_adjust) != 0 ) {
746 /* We are doing an adjtime thing.
748 * Prepare time_adjust_step to be within bounds.
749 * Note that a positive time_adjust means we want the clock
752 * Limit the amount of the step to be in the range
753 * -tickadj .. +tickadj
755 if (time_adjust > tickadj)
756 time_adjust_step = tickadj;
757 else if (time_adjust < -tickadj)
758 time_adjust_step = -tickadj;
760 /* Reduce by this step the amount of time left */
761 time_adjust -= time_adjust_step;
763 delta_nsec = tick_nsec + time_adjust_step * 1000;
765 * Advance the phase, once it gets to one microsecond, then
766 * advance the tick more.
768 time_phase += time_adj;
769 if (time_phase <= -FINENSEC) {
770 long ltemp = -time_phase >> (SHIFT_SCALE - 10);
771 time_phase += ltemp << (SHIFT_SCALE - 10);
774 else if (time_phase >= FINENSEC) {
775 long ltemp = time_phase >> (SHIFT_SCALE - 10);
776 time_phase -= ltemp << (SHIFT_SCALE - 10);
779 xtime.tv_nsec += delta_nsec;
780 time_interpolator_update(delta_nsec);
782 /* Changes by adjtime() do not take effect till next tick. */
783 if (time_next_adjust != 0) {
784 time_adjust = time_next_adjust;
785 time_next_adjust = 0;
790 * Using a loop looks inefficient, but "ticks" is
791 * usually just one (we shouldn't be losing ticks,
792 * we're doing this this way mainly for interrupt
793 * latency reasons, not because we think we'll
794 * have lots of lost timer ticks
796 static void update_wall_time(unsigned long ticks)
800 update_wall_time_one_tick();
801 if (xtime.tv_nsec >= 1000000000) {
802 xtime.tv_nsec -= 1000000000;
809 static inline void do_process_times(struct task_struct *p,
810 unsigned long user, unsigned long system)
814 psecs = (p->utime += user);
815 psecs += (p->stime += system);
816 if (p->signal && !unlikely(p->state & (EXIT_DEAD|EXIT_ZOMBIE)) &&
817 (psecs / HZ >= p->rlim[RLIMIT_CPU].rlim_cur)) {
818 /* Send SIGXCPU every second.. */
820 send_sig(SIGXCPU, p, 1);
821 /* and SIGKILL when we go over max.. */
822 if (psecs / HZ >= p->rlim[RLIMIT_CPU].rlim_max)
823 send_sig(SIGKILL, p, 1);
827 static inline void do_it_virt(struct task_struct * p, unsigned long ticks)
829 unsigned long it_virt = p->it_virt_value;
834 it_virt = p->it_virt_incr;
835 send_sig(SIGVTALRM, p, 1);
837 p->it_virt_value = it_virt;
841 static inline void do_it_prof(struct task_struct *p)
843 unsigned long it_prof = p->it_prof_value;
846 if (--it_prof == 0) {
847 it_prof = p->it_prof_incr;
848 send_sig(SIGPROF, p, 1);
850 p->it_prof_value = it_prof;
854 static void update_one_process(struct task_struct *p, unsigned long user,
855 unsigned long system, int cpu)
857 do_process_times(p, user, system);
863 * Called from the timer interrupt handler to charge one tick to the current
864 * process. user_tick is 1 if the tick is user time, 0 for system.
866 void update_process_times(int user_tick)
868 struct task_struct *p = current;
869 int cpu = smp_processor_id(), system = user_tick ^ 1;
871 update_one_process(p, user_tick, system, cpu);
873 scheduler_tick(user_tick, system);
877 * Nr of active tasks - counted in fixed-point numbers
879 static unsigned long count_active_tasks(void)
881 return (nr_running() + nr_uninterruptible()) * FIXED_1;
885 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
886 * imply that avenrun[] is the standard name for this kind of thing.
887 * Nothing else seems to be standardized: the fractional size etc
888 * all seem to differ on different machines.
890 * Requires xtime_lock to access.
892 unsigned long avenrun[3];
895 * calc_load - given tick count, update the avenrun load estimates.
896 * This is called while holding a write_lock on xtime_lock.
898 static inline void calc_load(unsigned long ticks)
900 unsigned long active_tasks; /* fixed-point */
901 static int count = LOAD_FREQ;
906 active_tasks = count_active_tasks();
907 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
908 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
909 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
913 /* jiffies at the most recent update of wall time */
914 unsigned long wall_jiffies = INITIAL_JIFFIES;
917 * This read-write spinlock protects us from races in SMP while
918 * playing with xtime and avenrun.
920 #ifndef ARCH_HAVE_XTIME_LOCK
921 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
923 EXPORT_SYMBOL(xtime_lock);
927 * This function runs timers and the timer-tq in bottom half context.
929 static void run_timer_softirq(struct softirq_action *h)
931 tvec_base_t *base = &__get_cpu_var(tvec_bases);
933 if (time_after_eq(jiffies, base->timer_jiffies))
938 * Called by the local, per-CPU timer interrupt on SMP.
940 void run_local_timers(void)
942 raise_softirq(TIMER_SOFTIRQ);
946 * Called by the timer interrupt. xtime_lock must already be taken
949 static inline void update_times(void)
953 ticks = jiffies - wall_jiffies;
955 wall_jiffies += ticks;
956 update_wall_time(ticks);
962 * The 64-bit jiffies value is not atomic - you MUST NOT read it
963 * without sampling the sequence number in xtime_lock.
964 * jiffies is defined in the linker script...
967 void do_timer(struct pt_regs *regs)
971 /* SMP process accounting uses the local APIC timer */
973 update_process_times(user_mode(regs));
978 #ifdef __ARCH_WANT_SYS_ALARM
981 * For backwards compatibility? This can be done in libc so Alpha
982 * and all newer ports shouldn't need it.
984 asmlinkage unsigned long sys_alarm(unsigned int seconds)
986 struct itimerval it_new, it_old;
987 unsigned int oldalarm;
989 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
990 it_new.it_value.tv_sec = seconds;
991 it_new.it_value.tv_usec = 0;
992 do_setitimer(ITIMER_REAL, &it_new, &it_old);
993 oldalarm = it_old.it_value.tv_sec;
994 /* ehhh.. We can't return 0 if we have an alarm pending.. */
995 /* And we'd better return too much than too little anyway */
996 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
1006 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1007 * should be moved into arch/i386 instead?
1011 * sys_getpid - return the thread group id of the current process
1013 * Note, despite the name, this returns the tgid not the pid. The tgid and
1014 * the pid are identical unless CLONE_THREAD was specified on clone() in
1015 * which case the tgid is the same in all threads of the same group.
1017 * This is SMP safe as current->tgid does not change.
1019 asmlinkage long sys_getpid(void)
1021 return vx_map_tgid(current->tgid);
1025 * Accessing ->group_leader->real_parent is not SMP-safe, it could
1026 * change from under us. However, rather than getting any lock
1027 * we can use an optimistic algorithm: get the parent
1028 * pid, and go back and check that the parent is still
1029 * the same. If it has changed (which is extremely unlikely
1030 * indeed), we just try again..
1032 * NOTE! This depends on the fact that even if we _do_
1033 * get an old value of "parent", we can happily dereference
1034 * the pointer (it was and remains a dereferencable kernel pointer
1035 * no matter what): we just can't necessarily trust the result
1036 * until we know that the parent pointer is valid.
1038 * NOTE2: ->group_leader never changes from under us.
1040 asmlinkage long sys_getppid(void)
1043 struct task_struct *me = current;
1044 struct task_struct *parent;
1046 parent = me->group_leader->real_parent;
1051 struct task_struct *old = parent;
1054 * Make sure we read the pid before re-reading the
1058 parent = me->group_leader->real_parent;
1065 return vx_map_pid(pid);
1068 asmlinkage long sys_getuid(void)
1070 /* Only we change this so SMP safe */
1071 return current->uid;
1074 asmlinkage long sys_geteuid(void)
1076 /* Only we change this so SMP safe */
1077 return current->euid;
1080 asmlinkage long sys_getgid(void)
1082 /* Only we change this so SMP safe */
1083 return current->gid;
1086 asmlinkage long sys_getegid(void)
1088 /* Only we change this so SMP safe */
1089 return current->egid;
1094 static void process_timeout(unsigned long __data)
1096 wake_up_process((task_t *)__data);
1100 * schedule_timeout - sleep until timeout
1101 * @timeout: timeout value in jiffies
1103 * Make the current task sleep until @timeout jiffies have
1104 * elapsed. The routine will return immediately unless
1105 * the current task state has been set (see set_current_state()).
1107 * You can set the task state as follows -
1109 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1110 * pass before the routine returns. The routine will return 0
1112 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1113 * delivered to the current task. In this case the remaining time
1114 * in jiffies will be returned, or 0 if the timer expired in time
1116 * The current task state is guaranteed to be TASK_RUNNING when this
1119 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1120 * the CPU away without a bound on the timeout. In this case the return
1121 * value will be %MAX_SCHEDULE_TIMEOUT.
1123 * In all cases the return value is guaranteed to be non-negative.
1125 fastcall signed long __sched schedule_timeout(signed long timeout)
1127 struct timer_list timer;
1128 unsigned long expire;
1132 case MAX_SCHEDULE_TIMEOUT:
1134 * These two special cases are useful to be comfortable
1135 * in the caller. Nothing more. We could take
1136 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1137 * but I' d like to return a valid offset (>=0) to allow
1138 * the caller to do everything it want with the retval.
1144 * Another bit of PARANOID. Note that the retval will be
1145 * 0 since no piece of kernel is supposed to do a check
1146 * for a negative retval of schedule_timeout() (since it
1147 * should never happens anyway). You just have the printk()
1148 * that will tell you if something is gone wrong and where.
1152 printk(KERN_ERR "schedule_timeout: wrong timeout "
1153 "value %lx from %p\n", timeout,
1154 __builtin_return_address(0));
1155 current->state = TASK_RUNNING;
1160 expire = timeout + jiffies;
1163 timer.expires = expire;
1164 timer.data = (unsigned long) current;
1165 timer.function = process_timeout;
1169 del_singleshot_timer_sync(&timer);
1171 timeout = expire - jiffies;
1174 return timeout < 0 ? 0 : timeout;
1177 EXPORT_SYMBOL(schedule_timeout);
1179 /* Thread ID - the internal kernel "pid" */
1180 asmlinkage long sys_gettid(void)
1182 return current->pid;
1185 static long __sched nanosleep_restart(struct restart_block *restart)
1187 unsigned long expire = restart->arg0, now = jiffies;
1188 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1191 /* Did it expire while we handled signals? */
1192 if (!time_after(expire, now))
1195 current->state = TASK_INTERRUPTIBLE;
1196 expire = schedule_timeout(expire - now);
1201 jiffies_to_timespec(expire, &t);
1203 ret = -ERESTART_RESTARTBLOCK;
1204 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1206 /* The 'restart' block is already filled in */
1211 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1214 unsigned long expire;
1217 if (copy_from_user(&t, rqtp, sizeof(t)))
1220 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1223 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1224 current->state = TASK_INTERRUPTIBLE;
1225 expire = schedule_timeout(expire);
1229 struct restart_block *restart;
1230 jiffies_to_timespec(expire, &t);
1231 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1234 restart = ¤t_thread_info()->restart_block;
1235 restart->fn = nanosleep_restart;
1236 restart->arg0 = jiffies + expire;
1237 restart->arg1 = (unsigned long) rmtp;
1238 ret = -ERESTART_RESTARTBLOCK;
1244 * sys_sysinfo - fill in sysinfo struct
1246 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1249 unsigned long mem_total, sav_total;
1250 unsigned int mem_unit, bitcount;
1253 memset((char *)&val, 0, sizeof(struct sysinfo));
1257 seq = read_seqbegin(&xtime_lock);
1260 * This is annoying. The below is the same thing
1261 * posix_get_clock_monotonic() does, but it wants to
1262 * take the lock which we want to cover the loads stuff
1266 getnstimeofday(&tp);
1267 tp.tv_sec += wall_to_monotonic.tv_sec;
1268 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1269 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1270 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1273 if (vx_flags(VXF_VIRT_UPTIME, 0))
1274 vx_vsi_uptime(&tp, NULL);
1275 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1277 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1278 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1279 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1281 val.procs = nr_threads;
1282 } while (read_seqretry(&xtime_lock, seq));
1284 /* if (vx_flags(VXF_VIRT_CPU, 0))
1291 * If the sum of all the available memory (i.e. ram + swap)
1292 * is less than can be stored in a 32 bit unsigned long then
1293 * we can be binary compatible with 2.2.x kernels. If not,
1294 * well, in that case 2.2.x was broken anyways...
1296 * -Erik Andersen <andersee@debian.org>
1299 mem_total = val.totalram + val.totalswap;
1300 if (mem_total < val.totalram || mem_total < val.totalswap)
1303 mem_unit = val.mem_unit;
1304 while (mem_unit > 1) {
1307 sav_total = mem_total;
1309 if (mem_total < sav_total)
1314 * If mem_total did not overflow, multiply all memory values by
1315 * val.mem_unit and set it to 1. This leaves things compatible
1316 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1321 val.totalram <<= bitcount;
1322 val.freeram <<= bitcount;
1323 val.sharedram <<= bitcount;
1324 val.bufferram <<= bitcount;
1325 val.totalswap <<= bitcount;
1326 val.freeswap <<= bitcount;
1327 val.totalhigh <<= bitcount;
1328 val.freehigh <<= bitcount;
1331 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1337 static void __devinit init_timers_cpu(int cpu)
1342 base = &per_cpu(tvec_bases, cpu);
1343 spin_lock_init(&base->lock);
1344 for (j = 0; j < TVN_SIZE; j++) {
1345 INIT_LIST_HEAD(base->tv5.vec + j);
1346 INIT_LIST_HEAD(base->tv4.vec + j);
1347 INIT_LIST_HEAD(base->tv3.vec + j);
1348 INIT_LIST_HEAD(base->tv2.vec + j);
1350 for (j = 0; j < TVR_SIZE; j++)
1351 INIT_LIST_HEAD(base->tv1.vec + j);
1353 base->timer_jiffies = jiffies;
1356 #ifdef CONFIG_HOTPLUG_CPU
1357 static int migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1359 struct timer_list *timer;
1361 while (!list_empty(head)) {
1362 timer = list_entry(head->next, struct timer_list, entry);
1363 /* We're locking backwards from __mod_timer order here,
1365 if (!spin_trylock(&timer->lock))
1367 list_del(&timer->entry);
1368 internal_add_timer(new_base, timer);
1369 timer->base = new_base;
1370 spin_unlock(&timer->lock);
1375 static void __devinit migrate_timers(int cpu)
1377 tvec_base_t *old_base;
1378 tvec_base_t *new_base;
1381 BUG_ON(cpu_online(cpu));
1382 old_base = &per_cpu(tvec_bases, cpu);
1383 new_base = &get_cpu_var(tvec_bases);
1385 local_irq_disable();
1387 /* Prevent deadlocks via ordering by old_base < new_base. */
1388 if (old_base < new_base) {
1389 spin_lock(&new_base->lock);
1390 spin_lock(&old_base->lock);
1392 spin_lock(&old_base->lock);
1393 spin_lock(&new_base->lock);
1396 if (old_base->running_timer)
1398 for (i = 0; i < TVR_SIZE; i++)
1399 if (!migrate_timer_list(new_base, old_base->tv1.vec + i))
1401 for (i = 0; i < TVN_SIZE; i++)
1402 if (!migrate_timer_list(new_base, old_base->tv2.vec + i)
1403 || !migrate_timer_list(new_base, old_base->tv3.vec + i)
1404 || !migrate_timer_list(new_base, old_base->tv4.vec + i)
1405 || !migrate_timer_list(new_base, old_base->tv5.vec + i))
1407 spin_unlock(&old_base->lock);
1408 spin_unlock(&new_base->lock);
1410 put_cpu_var(tvec_bases);
1414 /* Avoid deadlock with __mod_timer, by backing off. */
1415 spin_unlock(&old_base->lock);
1416 spin_unlock(&new_base->lock);
1420 #endif /* CONFIG_HOTPLUG_CPU */
1422 static int __devinit timer_cpu_notify(struct notifier_block *self,
1423 unsigned long action, void *hcpu)
1425 long cpu = (long)hcpu;
1427 case CPU_UP_PREPARE:
1428 init_timers_cpu(cpu);
1430 #ifdef CONFIG_HOTPLUG_CPU
1432 migrate_timers(cpu);
1441 static struct notifier_block __devinitdata timers_nb = {
1442 .notifier_call = timer_cpu_notify,
1446 void __init init_timers(void)
1448 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1449 (void *)(long)smp_processor_id());
1450 register_cpu_notifier(&timers_nb);
1451 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1454 #ifdef CONFIG_TIME_INTERPOLATION
1456 struct time_interpolator *time_interpolator;
1457 static struct time_interpolator *time_interpolator_list;
1458 static spinlock_t time_interpolator_lock = SPIN_LOCK_UNLOCKED;
1460 static inline u64 time_interpolator_get_cycles(unsigned int src)
1462 unsigned long (*x)(void);
1466 case TIME_SOURCE_FUNCTION:
1467 x = time_interpolator->addr;
1470 case TIME_SOURCE_MMIO64 :
1471 return readq(time_interpolator->addr);
1473 case TIME_SOURCE_MMIO32 :
1474 return readl(time_interpolator->addr);
1476 default: return get_cycles();
1480 static inline u64 time_interpolator_get_counter(void)
1482 unsigned int src = time_interpolator->source;
1484 if (time_interpolator->jitter)
1490 lcycle = time_interpolator->last_cycle;
1491 now = time_interpolator_get_cycles(src);
1492 if (lcycle && time_after(lcycle, now))
1494 /* Keep track of the last timer value returned. The use of cmpxchg here
1495 * will cause contention in an SMP environment.
1497 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1501 return time_interpolator_get_cycles(src);
1504 void time_interpolator_reset(void)
1506 time_interpolator->offset = 0;
1507 time_interpolator->last_counter = time_interpolator_get_counter();
1510 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1512 unsigned long time_interpolator_get_offset(void)
1514 /* If we do not have a time interpolator set up then just return zero */
1515 if (!time_interpolator)
1518 return time_interpolator->offset +
1519 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator);
1522 #define INTERPOLATOR_ADJUST 65536
1523 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1525 static void time_interpolator_update(long delta_nsec)
1528 unsigned long offset;
1530 /* If there is no time interpolator set up then do nothing */
1531 if (!time_interpolator)
1534 /* The interpolator compensates for late ticks by accumulating
1535 * the late time in time_interpolator->offset. A tick earlier than
1536 * expected will lead to a reset of the offset and a corresponding
1537 * jump of the clock forward. Again this only works if the
1538 * interpolator clock is running slightly slower than the regular clock
1539 * and the tuning logic insures that.
1542 counter = time_interpolator_get_counter();
1543 offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
1545 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1546 time_interpolator->offset = offset - delta_nsec;
1548 time_interpolator->skips++;
1549 time_interpolator->ns_skipped += delta_nsec - offset;
1550 time_interpolator->offset = 0;
1552 time_interpolator->last_counter = counter;
1554 /* Tuning logic for time interpolator invoked every minute or so.
1555 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1556 * Increase interpolator clock speed if we skip too much time.
1558 if (jiffies % INTERPOLATOR_ADJUST == 0)
1560 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1561 time_interpolator->nsec_per_cyc--;
1562 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1563 time_interpolator->nsec_per_cyc++;
1564 time_interpolator->skips = 0;
1565 time_interpolator->ns_skipped = 0;
1570 is_better_time_interpolator(struct time_interpolator *new)
1572 if (!time_interpolator)
1574 return new->frequency > 2*time_interpolator->frequency ||
1575 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1579 register_time_interpolator(struct time_interpolator *ti)
1581 unsigned long flags;
1584 if (ti->frequency == 0 || ti->mask == 0)
1587 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1588 spin_lock(&time_interpolator_lock);
1589 write_seqlock_irqsave(&xtime_lock, flags);
1590 if (is_better_time_interpolator(ti)) {
1591 time_interpolator = ti;
1592 time_interpolator_reset();
1594 write_sequnlock_irqrestore(&xtime_lock, flags);
1596 ti->next = time_interpolator_list;
1597 time_interpolator_list = ti;
1598 spin_unlock(&time_interpolator_lock);
1602 unregister_time_interpolator(struct time_interpolator *ti)
1604 struct time_interpolator *curr, **prev;
1605 unsigned long flags;
1607 spin_lock(&time_interpolator_lock);
1608 prev = &time_interpolator_list;
1609 for (curr = *prev; curr; curr = curr->next) {
1617 write_seqlock_irqsave(&xtime_lock, flags);
1618 if (ti == time_interpolator) {
1619 /* we lost the best time-interpolator: */
1620 time_interpolator = NULL;
1621 /* find the next-best interpolator */
1622 for (curr = time_interpolator_list; curr; curr = curr->next)
1623 if (is_better_time_interpolator(curr))
1624 time_interpolator = curr;
1625 time_interpolator_reset();
1627 write_sequnlock_irqrestore(&xtime_lock, flags);
1628 spin_unlock(&time_interpolator_lock);
1630 #endif /* CONFIG_TIME_INTERPOLATION */
1633 * msleep - sleep safely even with waitqueue interruptions
1634 * @msecs: Time in milliseconds to sleep for
1636 void msleep(unsigned int msecs)
1638 unsigned long timeout = msecs_to_jiffies(msecs);
1641 set_current_state(TASK_UNINTERRUPTIBLE);
1642 timeout = schedule_timeout(timeout);
1646 EXPORT_SYMBOL(msleep);
1649 * msleep_interruptible - sleep waiting for waitqueue interruptions
1650 * @msecs: Time in milliseconds to sleep for
1652 unsigned long msleep_interruptible(unsigned int msecs)
1654 unsigned long timeout = msecs_to_jiffies(msecs);
1656 while (timeout && !signal_pending(current)) {
1657 set_current_state(TASK_INTERRUPTIBLE);
1658 timeout = schedule_timeout(timeout);
1660 return jiffies_to_msecs(timeout);
1663 EXPORT_SYMBOL(msleep_interruptible);