4 * Kernel internal timers, kernel timekeeping, basic process system calls
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
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:
53 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
54 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
55 #define TVN_SIZE (1 << TVN_BITS)
56 #define TVR_SIZE (1 << TVR_BITS)
57 #define TVN_MASK (TVN_SIZE - 1)
58 #define TVR_MASK (TVR_SIZE - 1)
62 struct timer_list *running_timer;
65 typedef struct tvec_s {
66 struct list_head vec[TVN_SIZE];
69 typedef struct tvec_root_s {
70 struct list_head vec[TVR_SIZE];
73 struct tvec_t_base_s {
74 struct timer_base_s t_base;
75 unsigned long timer_jiffies;
81 } ____cacheline_aligned_in_smp;
83 typedef struct tvec_t_base_s tvec_base_t;
84 static DEFINE_PER_CPU(tvec_base_t, tvec_bases);
86 static inline void set_running_timer(tvec_base_t *base,
87 struct timer_list *timer)
90 base->t_base.running_timer = timer;
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 timer->magic = TIMER_MAGIC;
111 static inline void check_timer(struct timer_list *timer)
113 if (timer->magic != TIMER_MAGIC)
114 check_timer_failed(timer);
118 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
120 unsigned long expires = timer->expires;
121 unsigned long idx = expires - base->timer_jiffies;
122 struct list_head *vec;
124 if (idx < TVR_SIZE) {
125 int i = expires & TVR_MASK;
126 vec = base->tv1.vec + i;
127 } else if (idx < 1 << (TVR_BITS + TVN_BITS)) {
128 int i = (expires >> TVR_BITS) & TVN_MASK;
129 vec = base->tv2.vec + i;
130 } else if (idx < 1 << (TVR_BITS + 2 * TVN_BITS)) {
131 int i = (expires >> (TVR_BITS + TVN_BITS)) & TVN_MASK;
132 vec = base->tv3.vec + i;
133 } else if (idx < 1 << (TVR_BITS + 3 * TVN_BITS)) {
134 int i = (expires >> (TVR_BITS + 2 * TVN_BITS)) & TVN_MASK;
135 vec = base->tv4.vec + i;
136 } else if ((signed long) idx < 0) {
138 * Can happen if you add a timer with expires == jiffies,
139 * or you set a timer to go off in the past
141 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
144 /* If the timeout is larger than 0xffffffff on 64-bit
145 * architectures then we use the maximum timeout:
147 if (idx > 0xffffffffUL) {
149 expires = idx + base->timer_jiffies;
151 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
152 vec = base->tv5.vec + i;
157 list_add_tail(&timer->entry, vec);
160 typedef struct timer_base_s timer_base_t;
162 * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
163 * at compile time, and we need timer->base to lock the timer.
165 timer_base_t __init_timer_base
166 ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED };
167 EXPORT_SYMBOL(__init_timer_base);
170 * init_timer - initialize a timer.
171 * @timer: the timer to be initialized
173 * init_timer() must be done to a timer prior calling *any* of the
174 * other timer functions.
176 void fastcall init_timer(struct timer_list *timer)
178 timer->entry.next = NULL;
179 timer->base = &per_cpu(tvec_bases, raw_smp_processor_id()).t_base;
180 timer->magic = TIMER_MAGIC;
182 EXPORT_SYMBOL(init_timer);
184 static inline void detach_timer(struct timer_list *timer,
187 struct list_head *entry = &timer->entry;
189 __list_del(entry->prev, entry->next);
192 entry->prev = LIST_POISON2;
196 * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
197 * means that all timers which are tied to this base via timer->base are
198 * locked, and the base itself is locked too.
200 * So __run_timers/migrate_timers can safely modify all timers which could
201 * be found on ->tvX lists.
203 * When the timer's base is locked, and the timer removed from list, it is
204 * possible to set timer->base = NULL and drop the lock: the timer remains
207 static timer_base_t *lock_timer_base(struct timer_list *timer,
208 unsigned long *flags)
214 if (likely(base != NULL)) {
215 spin_lock_irqsave(&base->lock, *flags);
216 if (likely(base == timer->base))
218 /* The timer has migrated to another CPU */
219 spin_unlock_irqrestore(&base->lock, *flags);
225 int __mod_timer(struct timer_list *timer, unsigned long expires)
228 tvec_base_t *new_base;
232 BUG_ON(!timer->function);
235 base = lock_timer_base(timer, &flags);
237 if (timer_pending(timer)) {
238 detach_timer(timer, 0);
242 new_base = &__get_cpu_var(tvec_bases);
244 if (base != &new_base->t_base) {
246 * We are trying to schedule the timer on the local CPU.
247 * However we can't change timer's base while it is running,
248 * otherwise del_timer_sync() can't detect that the timer's
249 * handler yet has not finished. This also guarantees that
250 * the timer is serialized wrt itself.
252 if (unlikely(base->running_timer == timer)) {
253 /* The timer remains on a former base */
254 new_base = container_of(base, tvec_base_t, t_base);
256 /* See the comment in lock_timer_base() */
258 spin_unlock(&base->lock);
259 spin_lock(&new_base->t_base.lock);
260 timer->base = &new_base->t_base;
264 timer->expires = expires;
265 internal_add_timer(new_base, timer);
266 spin_unlock_irqrestore(&new_base->t_base.lock, flags);
271 EXPORT_SYMBOL(__mod_timer);
274 * add_timer_on - start a timer on a particular CPU
275 * @timer: the timer to be added
276 * @cpu: the CPU to start it on
278 * This is not very scalable on SMP. Double adds are not possible.
280 void add_timer_on(struct timer_list *timer, int cpu)
282 tvec_base_t *base = &per_cpu(tvec_bases, cpu);
285 BUG_ON(timer_pending(timer) || !timer->function);
289 spin_lock_irqsave(&base->t_base.lock, flags);
290 timer->base = &base->t_base;
291 internal_add_timer(base, timer);
292 spin_unlock_irqrestore(&base->t_base.lock, flags);
297 * mod_timer - modify a timer's timeout
298 * @timer: the timer to be modified
300 * mod_timer is a more efficient way to update the expire field of an
301 * active timer (if the timer is inactive it will be activated)
303 * mod_timer(timer, expires) is equivalent to:
305 * del_timer(timer); timer->expires = expires; add_timer(timer);
307 * Note that if there are multiple unserialized concurrent users of the
308 * same timer, then mod_timer() is the only safe way to modify the timeout,
309 * since add_timer() cannot modify an already running timer.
311 * The function returns whether it has modified a pending timer or not.
312 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
313 * active timer returns 1.)
315 int mod_timer(struct timer_list *timer, unsigned long expires)
317 BUG_ON(!timer->function);
322 * This is a common optimization triggered by the
323 * networking code - if the timer is re-modified
324 * to be the same thing then just return:
326 if (timer->expires == expires && timer_pending(timer))
329 return __mod_timer(timer, expires);
332 EXPORT_SYMBOL(mod_timer);
335 * del_timer - deactive a timer.
336 * @timer: the timer to be deactivated
338 * del_timer() deactivates a timer - this works on both active and inactive
341 * The function returns whether it has deactivated a pending timer or not.
342 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
343 * active timer returns 1.)
345 int del_timer(struct timer_list *timer)
353 if (timer_pending(timer)) {
354 base = lock_timer_base(timer, &flags);
355 if (timer_pending(timer)) {
356 detach_timer(timer, 1);
359 spin_unlock_irqrestore(&base->lock, flags);
365 EXPORT_SYMBOL(del_timer);
369 * This function tries to deactivate a timer. Upon successful (ret >= 0)
370 * exit the timer is not queued and the handler is not running on any CPU.
372 * It must not be called from interrupt contexts.
374 int try_to_del_timer_sync(struct timer_list *timer)
380 base = lock_timer_base(timer, &flags);
382 if (base->running_timer == timer)
386 if (timer_pending(timer)) {
387 detach_timer(timer, 1);
391 spin_unlock_irqrestore(&base->lock, flags);
397 * del_timer_sync - deactivate a timer and wait for the handler to finish.
398 * @timer: the timer to be deactivated
400 * This function only differs from del_timer() on SMP: besides deactivating
401 * the timer it also makes sure the handler has finished executing on other
404 * Synchronization rules: callers must prevent restarting of the timer,
405 * otherwise this function is meaningless. It must not be called from
406 * interrupt contexts. The caller must not hold locks which would prevent
407 * completion of the timer's handler. The timer's handler must not call
408 * add_timer_on(). Upon exit the timer is not queued and the handler is
409 * not running on any CPU.
411 * The function returns whether it has deactivated a pending timer or not.
413 int del_timer_sync(struct timer_list *timer)
418 int ret = try_to_del_timer_sync(timer);
424 EXPORT_SYMBOL(del_timer_sync);
427 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
429 /* cascade all the timers from tv up one level */
430 struct list_head *head, *curr;
432 head = tv->vec + index;
435 * We are removing _all_ timers from the list, so we don't have to
436 * detach them individually, just clear the list afterwards.
438 while (curr != head) {
439 struct timer_list *tmp;
441 tmp = list_entry(curr, struct timer_list, entry);
442 BUG_ON(tmp->base != &base->t_base);
444 internal_add_timer(base, tmp);
446 INIT_LIST_HEAD(head);
452 * __run_timers - run all expired timers (if any) on this CPU.
453 * @base: the timer vector to be processed.
455 * This function cascades all vectors and executes all expired timer
458 #define INDEX(N) (base->timer_jiffies >> (TVR_BITS + N * TVN_BITS)) & TVN_MASK
460 static inline void __run_timers(tvec_base_t *base)
462 struct timer_list *timer;
464 spin_lock_irq(&base->t_base.lock);
465 while (time_after_eq(jiffies, base->timer_jiffies)) {
466 struct list_head work_list = LIST_HEAD_INIT(work_list);
467 struct list_head *head = &work_list;
468 int index = base->timer_jiffies & TVR_MASK;
474 (!cascade(base, &base->tv2, INDEX(0))) &&
475 (!cascade(base, &base->tv3, INDEX(1))) &&
476 !cascade(base, &base->tv4, INDEX(2)))
477 cascade(base, &base->tv5, INDEX(3));
478 ++base->timer_jiffies;
479 list_splice_init(base->tv1.vec + index, &work_list);
480 while (!list_empty(head)) {
481 void (*fn)(unsigned long);
484 timer = list_entry(head->next,struct timer_list,entry);
485 fn = timer->function;
488 set_running_timer(base, timer);
489 detach_timer(timer, 1);
490 spin_unlock_irq(&base->t_base.lock);
492 int preempt_count = preempt_count();
494 if (preempt_count != preempt_count()) {
495 printk(KERN_WARNING "huh, entered %p "
496 "with preempt_count %08x, exited"
503 spin_lock_irq(&base->t_base.lock);
506 set_running_timer(base, NULL);
507 spin_unlock_irq(&base->t_base.lock);
510 #ifdef CONFIG_NO_IDLE_HZ
512 * Find out when the next timer event is due to happen. This
513 * is used on S/390 to stop all activity when a cpus is idle.
514 * This functions needs to be called disabled.
516 unsigned long next_timer_interrupt(void)
519 struct list_head *list;
520 struct timer_list *nte;
521 unsigned long expires;
525 base = &__get_cpu_var(tvec_bases);
526 spin_lock(&base->t_base.lock);
527 expires = base->timer_jiffies + (LONG_MAX >> 1);
530 /* Look for timer events in tv1. */
531 j = base->timer_jiffies & TVR_MASK;
533 list_for_each_entry(nte, base->tv1.vec + j, entry) {
534 expires = nte->expires;
535 if (j < (base->timer_jiffies & TVR_MASK))
536 list = base->tv2.vec + (INDEX(0));
539 j = (j + 1) & TVR_MASK;
540 } while (j != (base->timer_jiffies & TVR_MASK));
543 varray[0] = &base->tv2;
544 varray[1] = &base->tv3;
545 varray[2] = &base->tv4;
546 varray[3] = &base->tv5;
547 for (i = 0; i < 4; i++) {
550 if (list_empty(varray[i]->vec + j)) {
551 j = (j + 1) & TVN_MASK;
554 list_for_each_entry(nte, varray[i]->vec + j, entry)
555 if (time_before(nte->expires, expires))
556 expires = nte->expires;
557 if (j < (INDEX(i)) && i < 3)
558 list = varray[i + 1]->vec + (INDEX(i + 1));
560 } while (j != (INDEX(i)));
565 * The search wrapped. We need to look at the next list
566 * from next tv element that would cascade into tv element
567 * where we found the timer element.
569 list_for_each_entry(nte, list, entry) {
570 if (time_before(nte->expires, expires))
571 expires = nte->expires;
574 spin_unlock(&base->t_base.lock);
579 /******************************************************************/
582 * Timekeeping variables
584 unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
585 unsigned long tick_nsec = TICK_NSEC; /* ACTHZ period (nsec) */
589 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
590 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
591 * at zero at system boot time, so wall_to_monotonic will be negative,
592 * however, we will ALWAYS keep the tv_nsec part positive so we can use
593 * the usual normalization.
595 struct timespec xtime __attribute__ ((aligned (16)));
596 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
598 EXPORT_SYMBOL(xtime);
600 /* Don't completely fail for HZ > 500. */
601 int tickadj = 500/HZ ? : 1; /* microsecs */
605 * phase-lock loop variables
607 /* TIME_ERROR prevents overwriting the CMOS clock */
608 int time_state = TIME_OK; /* clock synchronization status */
609 int time_status = STA_UNSYNC; /* clock status bits */
610 long time_offset; /* time adjustment (us) */
611 long time_constant = 2; /* pll time constant */
612 long time_tolerance = MAXFREQ; /* frequency tolerance (ppm) */
613 long time_precision = 1; /* clock precision (us) */
614 long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
615 long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
616 static long time_phase; /* phase offset (scaled us) */
617 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
618 /* frequency offset (scaled ppm)*/
619 static long time_adj; /* tick adjust (scaled 1 / HZ) */
620 long time_reftime; /* time at last adjustment (s) */
622 long time_next_adjust;
625 * this routine handles the overflow of the microsecond field
627 * The tricky bits of code to handle the accurate clock support
628 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
629 * They were originally developed for SUN and DEC kernels.
630 * All the kudos should go to Dave for this stuff.
633 static void second_overflow(void)
637 /* Bump the maxerror field */
638 time_maxerror += time_tolerance >> SHIFT_USEC;
639 if (time_maxerror > NTP_PHASE_LIMIT) {
640 time_maxerror = NTP_PHASE_LIMIT;
641 time_status |= STA_UNSYNC;
645 * Leap second processing. If in leap-insert state at the end of the
646 * day, the system clock is set back one second; if in leap-delete
647 * state, the system clock is set ahead one second. The microtime()
648 * routine or external clock driver will insure that reported time is
649 * always monotonic. The ugly divides should be replaced.
651 switch (time_state) {
653 if (time_status & STA_INS)
654 time_state = TIME_INS;
655 else if (time_status & STA_DEL)
656 time_state = TIME_DEL;
659 if (xtime.tv_sec % 86400 == 0) {
661 wall_to_monotonic.tv_sec++;
663 * The timer interpolator will make time change
664 * gradually instead of an immediate jump by one second
666 time_interpolator_update(-NSEC_PER_SEC);
667 time_state = TIME_OOP;
669 printk(KERN_NOTICE "Clock: inserting leap second "
674 if ((xtime.tv_sec + 1) % 86400 == 0) {
676 wall_to_monotonic.tv_sec--;
678 * Use of time interpolator for a gradual change of
681 time_interpolator_update(NSEC_PER_SEC);
682 time_state = TIME_WAIT;
684 printk(KERN_NOTICE "Clock: deleting leap second "
689 time_state = TIME_WAIT;
692 if (!(time_status & (STA_INS | STA_DEL)))
693 time_state = TIME_OK;
697 * Compute the phase adjustment for the next second. In PLL mode, the
698 * offset is reduced by a fixed factor times the time constant. In FLL
699 * mode the offset is used directly. In either mode, the maximum phase
700 * adjustment for each second is clamped so as to spread the adjustment
701 * over not more than the number of seconds between updates.
704 if (!(time_status & STA_FLL))
705 ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
706 ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
707 ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
708 time_offset -= ltemp;
709 time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
712 * Compute the frequency estimate and additional phase adjustment due
713 * to frequency error for the next second. When the PPS signal is
714 * engaged, gnaw on the watchdog counter and update the frequency
715 * computed by the pll and the PPS signal.
718 if (pps_valid == PPS_VALID) { /* PPS signal lost */
719 pps_jitter = MAXTIME;
720 pps_stabil = MAXFREQ;
721 time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
722 STA_PPSWANDER | STA_PPSERROR);
724 ltemp = time_freq + pps_freq;
725 time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
729 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
730 * get 128.125; => only 0.125% error (p. 14)
732 time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
736 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
737 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
739 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
743 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
744 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
746 time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
750 /* in the NTP reference this is called "hardclock()" */
751 static void update_wall_time_one_tick(void)
753 long time_adjust_step, delta_nsec;
755 if ((time_adjust_step = time_adjust) != 0 ) {
757 * We are doing an adjtime thing. Prepare time_adjust_step to
758 * be within bounds. Note that a positive time_adjust means we
759 * want the clock to run faster.
761 * Limit the amount of the step to be in the range
762 * -tickadj .. +tickadj
764 time_adjust_step = min(time_adjust_step, (long)tickadj);
765 time_adjust_step = max(time_adjust_step, (long)-tickadj);
767 /* Reduce by this step the amount of time left */
768 time_adjust -= time_adjust_step;
770 delta_nsec = tick_nsec + time_adjust_step * 1000;
772 * Advance the phase, once it gets to one microsecond, then
773 * advance the tick more.
775 time_phase += time_adj;
776 if ((time_phase >= FINENSEC) || (time_phase <= -FINENSEC)) {
777 long ltemp = shift_right(time_phase, (SHIFT_SCALE - 10));
778 time_phase -= ltemp << (SHIFT_SCALE - 10);
781 xtime.tv_nsec += delta_nsec;
782 time_interpolator_update(delta_nsec);
784 /* Changes by adjtime() do not take effect till next tick. */
785 if (time_next_adjust != 0) {
786 time_adjust = time_next_adjust;
787 time_next_adjust = 0;
792 * Using a loop looks inefficient, but "ticks" is
793 * usually just one (we shouldn't be losing ticks,
794 * we're doing this this way mainly for interrupt
795 * latency reasons, not because we think we'll
796 * have lots of lost timer ticks
798 static void update_wall_time(unsigned long ticks)
802 update_wall_time_one_tick();
803 if (xtime.tv_nsec >= 1000000000) {
804 xtime.tv_nsec -= 1000000000;
812 * Called from the timer interrupt handler to charge one tick to the current
813 * process. user_tick is 1 if the tick is user time, 0 for system.
815 void update_process_times(int user_tick)
817 struct task_struct *p = current;
818 int cpu = smp_processor_id();
820 /* Note: this timer irq context must be accounted for as well. */
822 account_user_time(p, jiffies_to_cputime(1));
824 account_system_time(p, HARDIRQ_OFFSET, jiffies_to_cputime(1));
826 if (rcu_pending(cpu))
827 rcu_check_callbacks(cpu, user_tick);
829 run_posix_cpu_timers(p);
833 * Nr of active tasks - counted in fixed-point numbers
835 static unsigned long count_active_tasks(void)
837 return (nr_running() + nr_uninterruptible()) * FIXED_1;
841 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
842 * imply that avenrun[] is the standard name for this kind of thing.
843 * Nothing else seems to be standardized: the fractional size etc
844 * all seem to differ on different machines.
846 * Requires xtime_lock to access.
848 unsigned long avenrun[3];
850 EXPORT_SYMBOL(avenrun);
853 * calc_load - given tick count, update the avenrun load estimates.
854 * This is called while holding a write_lock on xtime_lock.
856 static inline void calc_load(unsigned long ticks)
858 unsigned long active_tasks; /* fixed-point */
859 static int count = LOAD_FREQ;
864 active_tasks = count_active_tasks();
865 CALC_LOAD(avenrun[0], EXP_1, active_tasks);
866 CALC_LOAD(avenrun[1], EXP_5, active_tasks);
867 CALC_LOAD(avenrun[2], EXP_15, active_tasks);
871 /* jiffies at the most recent update of wall time */
872 unsigned long wall_jiffies = INITIAL_JIFFIES;
875 * This read-write spinlock protects us from races in SMP while
876 * playing with xtime and avenrun.
878 #ifndef ARCH_HAVE_XTIME_LOCK
879 seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
881 EXPORT_SYMBOL(xtime_lock);
885 * This function runs timers and the timer-tq in bottom half context.
887 static void run_timer_softirq(struct softirq_action *h)
889 tvec_base_t *base = &__get_cpu_var(tvec_bases);
891 if (time_after_eq(jiffies, base->timer_jiffies))
896 * Called by the local, per-CPU timer interrupt on SMP.
898 void run_local_timers(void)
900 raise_softirq(TIMER_SOFTIRQ);
904 * Called by the timer interrupt. xtime_lock must already be taken
907 static inline void update_times(void)
911 ticks = jiffies - wall_jiffies;
913 wall_jiffies += ticks;
914 update_wall_time(ticks);
920 * The 64-bit jiffies value is not atomic - you MUST NOT read it
921 * without sampling the sequence number in xtime_lock.
922 * jiffies is defined in the linker script...
925 void do_timer(struct pt_regs *regs)
929 softlockup_tick(regs);
932 #ifdef __ARCH_WANT_SYS_ALARM
935 * For backwards compatibility? This can be done in libc so Alpha
936 * and all newer ports shouldn't need it.
938 asmlinkage unsigned long sys_alarm(unsigned int seconds)
940 struct itimerval it_new, it_old;
941 unsigned int oldalarm;
943 it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
944 it_new.it_value.tv_sec = seconds;
945 it_new.it_value.tv_usec = 0;
946 do_setitimer(ITIMER_REAL, &it_new, &it_old);
947 oldalarm = it_old.it_value.tv_sec;
948 /* ehhh.. We can't return 0 if we have an alarm pending.. */
949 /* And we'd better return too much than too little anyway */
950 if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
960 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
961 * should be moved into arch/i386 instead?
965 * sys_getpid - return the thread group id of the current process
967 * Note, despite the name, this returns the tgid not the pid. The tgid and
968 * the pid are identical unless CLONE_THREAD was specified on clone() in
969 * which case the tgid is the same in all threads of the same group.
971 * This is SMP safe as current->tgid does not change.
973 asmlinkage long sys_getpid(void)
975 return current->tgid;
979 * Accessing ->group_leader->real_parent is not SMP-safe, it could
980 * change from under us. However, rather than getting any lock
981 * we can use an optimistic algorithm: get the parent
982 * pid, and go back and check that the parent is still
983 * the same. If it has changed (which is extremely unlikely
984 * indeed), we just try again..
986 * NOTE! This depends on the fact that even if we _do_
987 * get an old value of "parent", we can happily dereference
988 * the pointer (it was and remains a dereferencable kernel pointer
989 * no matter what): we just can't necessarily trust the result
990 * until we know that the parent pointer is valid.
992 * NOTE2: ->group_leader never changes from under us.
994 asmlinkage long sys_getppid(void)
997 struct task_struct *me = current;
998 struct task_struct *parent;
1000 parent = me->group_leader->real_parent;
1003 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1005 struct task_struct *old = parent;
1008 * Make sure we read the pid before re-reading the
1012 parent = me->group_leader->real_parent;
1022 asmlinkage long sys_getuid(void)
1024 /* Only we change this so SMP safe */
1025 return current->uid;
1028 asmlinkage long sys_geteuid(void)
1030 /* Only we change this so SMP safe */
1031 return current->euid;
1034 asmlinkage long sys_getgid(void)
1036 /* Only we change this so SMP safe */
1037 return current->gid;
1040 asmlinkage long sys_getegid(void)
1042 /* Only we change this so SMP safe */
1043 return current->egid;
1048 static void process_timeout(unsigned long __data)
1050 wake_up_process((task_t *)__data);
1054 * schedule_timeout - sleep until timeout
1055 * @timeout: timeout value in jiffies
1057 * Make the current task sleep until @timeout jiffies have
1058 * elapsed. The routine will return immediately unless
1059 * the current task state has been set (see set_current_state()).
1061 * You can set the task state as follows -
1063 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1064 * pass before the routine returns. The routine will return 0
1066 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1067 * delivered to the current task. In this case the remaining time
1068 * in jiffies will be returned, or 0 if the timer expired in time
1070 * The current task state is guaranteed to be TASK_RUNNING when this
1073 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1074 * the CPU away without a bound on the timeout. In this case the return
1075 * value will be %MAX_SCHEDULE_TIMEOUT.
1077 * In all cases the return value is guaranteed to be non-negative.
1079 fastcall signed long __sched schedule_timeout(signed long timeout)
1081 struct timer_list timer;
1082 unsigned long expire;
1086 case MAX_SCHEDULE_TIMEOUT:
1088 * These two special cases are useful to be comfortable
1089 * in the caller. Nothing more. We could take
1090 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1091 * but I' d like to return a valid offset (>=0) to allow
1092 * the caller to do everything it want with the retval.
1098 * Another bit of PARANOID. Note that the retval will be
1099 * 0 since no piece of kernel is supposed to do a check
1100 * for a negative retval of schedule_timeout() (since it
1101 * should never happens anyway). You just have the printk()
1102 * that will tell you if something is gone wrong and where.
1106 printk(KERN_ERR "schedule_timeout: wrong timeout "
1107 "value %lx from %p\n", timeout,
1108 __builtin_return_address(0));
1109 current->state = TASK_RUNNING;
1114 expire = timeout + jiffies;
1116 setup_timer(&timer, process_timeout, (unsigned long)current);
1117 __mod_timer(&timer, expire);
1119 del_singleshot_timer_sync(&timer);
1121 timeout = expire - jiffies;
1124 return timeout < 0 ? 0 : timeout;
1126 EXPORT_SYMBOL(schedule_timeout);
1129 * We can use __set_current_state() here because schedule_timeout() calls
1130 * schedule() unconditionally.
1132 signed long __sched schedule_timeout_interruptible(signed long timeout)
1134 __set_current_state(TASK_INTERRUPTIBLE);
1135 return schedule_timeout(timeout);
1137 EXPORT_SYMBOL(schedule_timeout_interruptible);
1139 signed long __sched schedule_timeout_uninterruptible(signed long timeout)
1141 __set_current_state(TASK_UNINTERRUPTIBLE);
1142 return schedule_timeout(timeout);
1144 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
1146 /* Thread ID - the internal kernel "pid" */
1147 asmlinkage long sys_gettid(void)
1149 return current->pid;
1152 static long __sched nanosleep_restart(struct restart_block *restart)
1154 unsigned long expire = restart->arg0, now = jiffies;
1155 struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
1158 /* Did it expire while we handled signals? */
1159 if (!time_after(expire, now))
1162 expire = schedule_timeout_interruptible(expire - now);
1167 jiffies_to_timespec(expire, &t);
1169 ret = -ERESTART_RESTARTBLOCK;
1170 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1172 /* The 'restart' block is already filled in */
1177 asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
1180 unsigned long expire;
1183 if (copy_from_user(&t, rqtp, sizeof(t)))
1186 if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
1189 expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
1190 expire = schedule_timeout_interruptible(expire);
1194 struct restart_block *restart;
1195 jiffies_to_timespec(expire, &t);
1196 if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
1199 restart = ¤t_thread_info()->restart_block;
1200 restart->fn = nanosleep_restart;
1201 restart->arg0 = jiffies + expire;
1202 restart->arg1 = (unsigned long) rmtp;
1203 ret = -ERESTART_RESTARTBLOCK;
1209 * sys_sysinfo - fill in sysinfo struct
1211 asmlinkage long sys_sysinfo(struct sysinfo __user *info)
1214 unsigned long mem_total, sav_total;
1215 unsigned int mem_unit, bitcount;
1218 memset((char *)&val, 0, sizeof(struct sysinfo));
1222 seq = read_seqbegin(&xtime_lock);
1225 * This is annoying. The below is the same thing
1226 * posix_get_clock_monotonic() does, but it wants to
1227 * take the lock which we want to cover the loads stuff
1231 getnstimeofday(&tp);
1232 tp.tv_sec += wall_to_monotonic.tv_sec;
1233 tp.tv_nsec += wall_to_monotonic.tv_nsec;
1234 if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
1235 tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
1238 val.uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
1240 val.loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
1241 val.loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
1242 val.loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
1244 val.procs = nr_threads;
1245 } while (read_seqretry(&xtime_lock, seq));
1251 * If the sum of all the available memory (i.e. ram + swap)
1252 * is less than can be stored in a 32 bit unsigned long then
1253 * we can be binary compatible with 2.2.x kernels. If not,
1254 * well, in that case 2.2.x was broken anyways...
1256 * -Erik Andersen <andersee@debian.org>
1259 mem_total = val.totalram + val.totalswap;
1260 if (mem_total < val.totalram || mem_total < val.totalswap)
1263 mem_unit = val.mem_unit;
1264 while (mem_unit > 1) {
1267 sav_total = mem_total;
1269 if (mem_total < sav_total)
1274 * If mem_total did not overflow, multiply all memory values by
1275 * val.mem_unit and set it to 1. This leaves things compatible
1276 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1281 val.totalram <<= bitcount;
1282 val.freeram <<= bitcount;
1283 val.sharedram <<= bitcount;
1284 val.bufferram <<= bitcount;
1285 val.totalswap <<= bitcount;
1286 val.freeswap <<= bitcount;
1287 val.totalhigh <<= bitcount;
1288 val.freehigh <<= bitcount;
1291 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
1297 static void __devinit init_timers_cpu(int cpu)
1302 base = &per_cpu(tvec_bases, cpu);
1303 spin_lock_init(&base->t_base.lock);
1304 for (j = 0; j < TVN_SIZE; j++) {
1305 INIT_LIST_HEAD(base->tv5.vec + j);
1306 INIT_LIST_HEAD(base->tv4.vec + j);
1307 INIT_LIST_HEAD(base->tv3.vec + j);
1308 INIT_LIST_HEAD(base->tv2.vec + j);
1310 for (j = 0; j < TVR_SIZE; j++)
1311 INIT_LIST_HEAD(base->tv1.vec + j);
1313 base->timer_jiffies = jiffies;
1316 #ifdef CONFIG_HOTPLUG_CPU
1317 static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
1319 struct timer_list *timer;
1321 while (!list_empty(head)) {
1322 timer = list_entry(head->next, struct timer_list, entry);
1323 detach_timer(timer, 0);
1324 timer->base = &new_base->t_base;
1325 internal_add_timer(new_base, timer);
1329 static void __devinit migrate_timers(int cpu)
1331 tvec_base_t *old_base;
1332 tvec_base_t *new_base;
1335 BUG_ON(cpu_online(cpu));
1336 old_base = &per_cpu(tvec_bases, cpu);
1337 new_base = &get_cpu_var(tvec_bases);
1339 local_irq_disable();
1340 spin_lock(&new_base->t_base.lock);
1341 spin_lock(&old_base->t_base.lock);
1343 if (old_base->t_base.running_timer)
1345 for (i = 0; i < TVR_SIZE; i++)
1346 migrate_timer_list(new_base, old_base->tv1.vec + i);
1347 for (i = 0; i < TVN_SIZE; i++) {
1348 migrate_timer_list(new_base, old_base->tv2.vec + i);
1349 migrate_timer_list(new_base, old_base->tv3.vec + i);
1350 migrate_timer_list(new_base, old_base->tv4.vec + i);
1351 migrate_timer_list(new_base, old_base->tv5.vec + i);
1354 spin_unlock(&old_base->t_base.lock);
1355 spin_unlock(&new_base->t_base.lock);
1357 put_cpu_var(tvec_bases);
1359 #endif /* CONFIG_HOTPLUG_CPU */
1361 static int __devinit timer_cpu_notify(struct notifier_block *self,
1362 unsigned long action, void *hcpu)
1364 long cpu = (long)hcpu;
1366 case CPU_UP_PREPARE:
1367 init_timers_cpu(cpu);
1369 #ifdef CONFIG_HOTPLUG_CPU
1371 migrate_timers(cpu);
1380 static struct notifier_block __devinitdata timers_nb = {
1381 .notifier_call = timer_cpu_notify,
1385 void __init init_timers(void)
1387 timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1388 (void *)(long)smp_processor_id());
1389 register_cpu_notifier(&timers_nb);
1390 open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
1393 #ifdef CONFIG_TIME_INTERPOLATION
1395 struct time_interpolator *time_interpolator;
1396 static struct time_interpolator *time_interpolator_list;
1397 static DEFINE_SPINLOCK(time_interpolator_lock);
1399 static inline u64 time_interpolator_get_cycles(unsigned int src)
1401 unsigned long (*x)(void);
1405 case TIME_SOURCE_FUNCTION:
1406 x = time_interpolator->addr;
1409 case TIME_SOURCE_MMIO64 :
1410 return readq((void __iomem *) time_interpolator->addr);
1412 case TIME_SOURCE_MMIO32 :
1413 return readl((void __iomem *) time_interpolator->addr);
1415 default: return get_cycles();
1419 static inline u64 time_interpolator_get_counter(int writelock)
1421 unsigned int src = time_interpolator->source;
1423 if (time_interpolator->jitter)
1429 lcycle = time_interpolator->last_cycle;
1430 now = time_interpolator_get_cycles(src);
1431 if (lcycle && time_after(lcycle, now))
1434 /* When holding the xtime write lock, there's no need
1435 * to add the overhead of the cmpxchg. Readers are
1436 * force to retry until the write lock is released.
1439 time_interpolator->last_cycle = now;
1442 /* Keep track of the last timer value returned. The use of cmpxchg here
1443 * will cause contention in an SMP environment.
1445 } while (unlikely(cmpxchg(&time_interpolator->last_cycle, lcycle, now) != lcycle));
1449 return time_interpolator_get_cycles(src);
1452 void time_interpolator_reset(void)
1454 time_interpolator->offset = 0;
1455 time_interpolator->last_counter = time_interpolator_get_counter(1);
1458 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1460 unsigned long time_interpolator_get_offset(void)
1462 /* If we do not have a time interpolator set up then just return zero */
1463 if (!time_interpolator)
1466 return time_interpolator->offset +
1467 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
1470 #define INTERPOLATOR_ADJUST 65536
1471 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1473 static void time_interpolator_update(long delta_nsec)
1476 unsigned long offset;
1478 /* If there is no time interpolator set up then do nothing */
1479 if (!time_interpolator)
1483 * The interpolator compensates for late ticks by accumulating the late
1484 * time in time_interpolator->offset. A tick earlier than expected will
1485 * lead to a reset of the offset and a corresponding jump of the clock
1486 * forward. Again this only works if the interpolator clock is running
1487 * slightly slower than the regular clock and the tuning logic insures
1491 counter = time_interpolator_get_counter(1);
1492 offset = time_interpolator->offset +
1493 GET_TI_NSECS(counter, time_interpolator);
1495 if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
1496 time_interpolator->offset = offset - delta_nsec;
1498 time_interpolator->skips++;
1499 time_interpolator->ns_skipped += delta_nsec - offset;
1500 time_interpolator->offset = 0;
1502 time_interpolator->last_counter = counter;
1504 /* Tuning logic for time interpolator invoked every minute or so.
1505 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1506 * Increase interpolator clock speed if we skip too much time.
1508 if (jiffies % INTERPOLATOR_ADJUST == 0)
1510 if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
1511 time_interpolator->nsec_per_cyc--;
1512 if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
1513 time_interpolator->nsec_per_cyc++;
1514 time_interpolator->skips = 0;
1515 time_interpolator->ns_skipped = 0;
1520 is_better_time_interpolator(struct time_interpolator *new)
1522 if (!time_interpolator)
1524 return new->frequency > 2*time_interpolator->frequency ||
1525 (unsigned long)new->drift < (unsigned long)time_interpolator->drift;
1529 register_time_interpolator(struct time_interpolator *ti)
1531 unsigned long flags;
1534 if (ti->frequency == 0 || ti->mask == 0)
1537 ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
1538 spin_lock(&time_interpolator_lock);
1539 write_seqlock_irqsave(&xtime_lock, flags);
1540 if (is_better_time_interpolator(ti)) {
1541 time_interpolator = ti;
1542 time_interpolator_reset();
1544 write_sequnlock_irqrestore(&xtime_lock, flags);
1546 ti->next = time_interpolator_list;
1547 time_interpolator_list = ti;
1548 spin_unlock(&time_interpolator_lock);
1552 unregister_time_interpolator(struct time_interpolator *ti)
1554 struct time_interpolator *curr, **prev;
1555 unsigned long flags;
1557 spin_lock(&time_interpolator_lock);
1558 prev = &time_interpolator_list;
1559 for (curr = *prev; curr; curr = curr->next) {
1567 write_seqlock_irqsave(&xtime_lock, flags);
1568 if (ti == time_interpolator) {
1569 /* we lost the best time-interpolator: */
1570 time_interpolator = NULL;
1571 /* find the next-best interpolator */
1572 for (curr = time_interpolator_list; curr; curr = curr->next)
1573 if (is_better_time_interpolator(curr))
1574 time_interpolator = curr;
1575 time_interpolator_reset();
1577 write_sequnlock_irqrestore(&xtime_lock, flags);
1578 spin_unlock(&time_interpolator_lock);
1580 #endif /* CONFIG_TIME_INTERPOLATION */
1583 * msleep - sleep safely even with waitqueue interruptions
1584 * @msecs: Time in milliseconds to sleep for
1586 void msleep(unsigned int msecs)
1588 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1591 timeout = schedule_timeout_uninterruptible(timeout);
1594 EXPORT_SYMBOL(msleep);
1597 * msleep_interruptible - sleep waiting for signals
1598 * @msecs: Time in milliseconds to sleep for
1600 unsigned long msleep_interruptible(unsigned int msecs)
1602 unsigned long timeout = msecs_to_jiffies(msecs) + 1;
1604 while (timeout && !signal_pending(current))
1605 timeout = schedule_timeout_interruptible(timeout);
1606 return jiffies_to_msecs(timeout);
1609 EXPORT_SYMBOL(msleep_interruptible);