4 * Kernel internal timers, kernel timekeeping, basic process system calls
6 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 1997-01-28 Modified by Finn Arne Gangstad to make timers scale better.
10 * 1997-09-10 Updated NTP code according to technical memorandum Jan '96
11 * "A Kernel Model for Precision Timekeeping" by Dave Mills
12 * 1998-12-24 Fixed a xtime SMP race (we need the xtime_lock rw spinlock to
13 * serialize accesses to xtime/lost_ticks).
14 * Copyright (C) 1998 Andrea Arcangeli
15 * 1999-03-10 Improved NTP compatibility by Ulrich Windl
16 * 2002-05-31 Move sys_sysinfo here and make its locking sane, Robert Love
17 * 2000-10-05 Implemented scalable SMP per-CPU timer handling.
18 * Copyright (C) 2000, 2001, 2002 Ingo Molnar
19 * Designed by David S. Miller, Alexey Kuznetsov and Ingo Molnar
22 #include <linux/kernel_stat.h>
23 #include <linux/module.h>
24 #include <linux/interrupt.h>
25 #include <linux/percpu.h>
26 #include <linux/init.h>
28 #include <linux/swap.h>
29 #include <linux/notifier.h>
30 #include <linux/thread_info.h>
31 #include <linux/time.h>
32 #include <linux/jiffies.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cpu.h>
35 #include <linux/syscalls.h>
36 #include <linux/delay.h>
38 #include <asm/uaccess.h>
39 #include <asm/unistd.h>
40 #include <asm/div64.h>
41 #include <asm/timex.h>
44 #ifdef CONFIG_TIME_INTERPOLATION
45 static void time_interpolator_update(long delta_nsec
);
47 #define time_interpolator_update(x)
50 u64 jiffies_64 __cacheline_aligned_in_smp
= INITIAL_JIFFIES
;
52 EXPORT_SYMBOL(jiffies_64
);
55 * per-CPU timer vector definitions:
57 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
58 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
59 #define TVN_SIZE (1 << TVN_BITS)
60 #define TVR_SIZE (1 << TVR_BITS)
61 #define TVN_MASK (TVN_SIZE - 1)
62 #define TVR_MASK (TVR_SIZE - 1)
64 typedef struct tvec_s
{
65 struct list_head vec
[TVN_SIZE
];
68 typedef struct tvec_root_s
{
69 struct list_head vec
[TVR_SIZE
];
72 struct tvec_t_base_s
{
74 struct timer_list
*running_timer
;
75 unsigned long timer_jiffies
;
81 } ____cacheline_aligned_in_smp
;
83 typedef struct tvec_t_base_s tvec_base_t
;
85 tvec_base_t boot_tvec_bases
;
86 EXPORT_SYMBOL(boot_tvec_bases
);
87 static DEFINE_PER_CPU(tvec_base_t
*, tvec_bases
) = &boot_tvec_bases
;
89 static inline void set_running_timer(tvec_base_t
*base
,
90 struct timer_list
*timer
)
93 base
->running_timer
= timer
;
97 static void internal_add_timer(tvec_base_t
*base
, struct timer_list
*timer
)
99 unsigned long expires
= timer
->expires
;
100 unsigned long idx
= expires
- base
->timer_jiffies
;
101 struct list_head
*vec
;
103 if (idx
< TVR_SIZE
) {
104 int i
= expires
& TVR_MASK
;
105 vec
= base
->tv1
.vec
+ i
;
106 } else if (idx
< 1 << (TVR_BITS
+ TVN_BITS
)) {
107 int i
= (expires
>> TVR_BITS
) & TVN_MASK
;
108 vec
= base
->tv2
.vec
+ i
;
109 } else if (idx
< 1 << (TVR_BITS
+ 2 * TVN_BITS
)) {
110 int i
= (expires
>> (TVR_BITS
+ TVN_BITS
)) & TVN_MASK
;
111 vec
= base
->tv3
.vec
+ i
;
112 } else if (idx
< 1 << (TVR_BITS
+ 3 * TVN_BITS
)) {
113 int i
= (expires
>> (TVR_BITS
+ 2 * TVN_BITS
)) & TVN_MASK
;
114 vec
= base
->tv4
.vec
+ i
;
115 } else if ((signed long) idx
< 0) {
117 * Can happen if you add a timer with expires == jiffies,
118 * or you set a timer to go off in the past
120 vec
= base
->tv1
.vec
+ (base
->timer_jiffies
& TVR_MASK
);
123 /* If the timeout is larger than 0xffffffff on 64-bit
124 * architectures then we use the maximum timeout:
126 if (idx
> 0xffffffffUL
) {
128 expires
= idx
+ base
->timer_jiffies
;
130 i
= (expires
>> (TVR_BITS
+ 3 * TVN_BITS
)) & TVN_MASK
;
131 vec
= base
->tv5
.vec
+ i
;
136 list_add_tail(&timer
->entry
, vec
);
140 * init_timer - initialize a timer.
141 * @timer: the timer to be initialized
143 * init_timer() must be done to a timer prior calling *any* of the
144 * other timer functions.
146 void fastcall
init_timer(struct timer_list
*timer
)
148 timer
->entry
.next
= NULL
;
149 timer
->base
= __raw_get_cpu_var(tvec_bases
);
151 EXPORT_SYMBOL(init_timer
);
153 static inline void detach_timer(struct timer_list
*timer
,
156 struct list_head
*entry
= &timer
->entry
;
158 __list_del(entry
->prev
, entry
->next
);
161 entry
->prev
= LIST_POISON2
;
165 * We are using hashed locking: holding per_cpu(tvec_bases).lock
166 * means that all timers which are tied to this base via timer->base are
167 * locked, and the base itself is locked too.
169 * So __run_timers/migrate_timers can safely modify all timers which could
170 * be found on ->tvX lists.
172 * When the timer's base is locked, and the timer removed from list, it is
173 * possible to set timer->base = NULL and drop the lock: the timer remains
176 static tvec_base_t
*lock_timer_base(struct timer_list
*timer
,
177 unsigned long *flags
)
178 __acquires(timer
->base
->lock
)
184 if (likely(base
!= NULL
)) {
185 spin_lock_irqsave(&base
->lock
, *flags
);
186 if (likely(base
== timer
->base
))
188 /* The timer has migrated to another CPU */
189 spin_unlock_irqrestore(&base
->lock
, *flags
);
195 int __mod_timer(struct timer_list
*timer
, unsigned long expires
)
197 tvec_base_t
*base
, *new_base
;
201 BUG_ON(!timer
->function
);
203 base
= lock_timer_base(timer
, &flags
);
205 if (timer_pending(timer
)) {
206 detach_timer(timer
, 0);
210 new_base
= __get_cpu_var(tvec_bases
);
212 if (base
!= new_base
) {
214 * We are trying to schedule the timer on the local CPU.
215 * However we can't change timer's base while it is running,
216 * otherwise del_timer_sync() can't detect that the timer's
217 * handler yet has not finished. This also guarantees that
218 * the timer is serialized wrt itself.
220 if (likely(base
->running_timer
!= timer
)) {
221 /* See the comment in lock_timer_base() */
223 spin_unlock(&base
->lock
);
225 spin_lock(&base
->lock
);
230 timer
->expires
= expires
;
231 internal_add_timer(base
, timer
);
232 spin_unlock_irqrestore(&base
->lock
, flags
);
237 EXPORT_SYMBOL(__mod_timer
);
240 * add_timer_on - start a timer on a particular CPU
241 * @timer: the timer to be added
242 * @cpu: the CPU to start it on
244 * This is not very scalable on SMP. Double adds are not possible.
246 void add_timer_on(struct timer_list
*timer
, int cpu
)
248 tvec_base_t
*base
= per_cpu(tvec_bases
, cpu
);
251 BUG_ON(timer_pending(timer
) || !timer
->function
);
252 spin_lock_irqsave(&base
->lock
, flags
);
254 internal_add_timer(base
, timer
);
255 spin_unlock_irqrestore(&base
->lock
, flags
);
260 * mod_timer - modify a timer's timeout
261 * @timer: the timer to be modified
262 * @expires: new timeout in jiffies
264 * mod_timer is a more efficient way to update the expire field of an
265 * active timer (if the timer is inactive it will be activated)
267 * mod_timer(timer, expires) is equivalent to:
269 * del_timer(timer); timer->expires = expires; add_timer(timer);
271 * Note that if there are multiple unserialized concurrent users of the
272 * same timer, then mod_timer() is the only safe way to modify the timeout,
273 * since add_timer() cannot modify an already running timer.
275 * The function returns whether it has modified a pending timer or not.
276 * (ie. mod_timer() of an inactive timer returns 0, mod_timer() of an
277 * active timer returns 1.)
279 int mod_timer(struct timer_list
*timer
, unsigned long expires
)
281 BUG_ON(!timer
->function
);
284 * This is a common optimization triggered by the
285 * networking code - if the timer is re-modified
286 * to be the same thing then just return:
288 if (timer
->expires
== expires
&& timer_pending(timer
))
291 return __mod_timer(timer
, expires
);
294 EXPORT_SYMBOL(mod_timer
);
297 * del_timer - deactive a timer.
298 * @timer: the timer to be deactivated
300 * del_timer() deactivates a timer - this works on both active and inactive
303 * The function returns whether it has deactivated a pending timer or not.
304 * (ie. del_timer() of an inactive timer returns 0, del_timer() of an
305 * active timer returns 1.)
307 int del_timer(struct timer_list
*timer
)
313 if (timer_pending(timer
)) {
314 base
= lock_timer_base(timer
, &flags
);
315 if (timer_pending(timer
)) {
316 detach_timer(timer
, 1);
319 spin_unlock_irqrestore(&base
->lock
, flags
);
325 EXPORT_SYMBOL(del_timer
);
329 * try_to_del_timer_sync - Try to deactivate a timer
330 * @timer: timer do del
332 * This function tries to deactivate a timer. Upon successful (ret >= 0)
333 * exit the timer is not queued and the handler is not running on any CPU.
335 * It must not be called from interrupt contexts.
337 int try_to_del_timer_sync(struct timer_list
*timer
)
343 base
= lock_timer_base(timer
, &flags
);
345 if (base
->running_timer
== timer
)
349 if (timer_pending(timer
)) {
350 detach_timer(timer
, 1);
354 spin_unlock_irqrestore(&base
->lock
, flags
);
360 * del_timer_sync - deactivate a timer and wait for the handler to finish.
361 * @timer: the timer to be deactivated
363 * This function only differs from del_timer() on SMP: besides deactivating
364 * the timer it also makes sure the handler has finished executing on other
367 * Synchronization rules: callers must prevent restarting of the timer,
368 * otherwise this function is meaningless. It must not be called from
369 * interrupt contexts. The caller must not hold locks which would prevent
370 * completion of the timer's handler. The timer's handler must not call
371 * add_timer_on(). Upon exit the timer is not queued and the handler is
372 * not running on any CPU.
374 * The function returns whether it has deactivated a pending timer or not.
376 int del_timer_sync(struct timer_list
*timer
)
379 int ret
= try_to_del_timer_sync(timer
);
386 EXPORT_SYMBOL(del_timer_sync
);
389 static int cascade(tvec_base_t
*base
, tvec_t
*tv
, int index
)
391 /* cascade all the timers from tv up one level */
392 struct timer_list
*timer
, *tmp
;
393 struct list_head tv_list
;
395 list_replace_init(tv
->vec
+ index
, &tv_list
);
398 * We are removing _all_ timers from the list, so we
399 * don't have to detach them individually.
401 list_for_each_entry_safe(timer
, tmp
, &tv_list
, entry
) {
402 BUG_ON(timer
->base
!= base
);
403 internal_add_timer(base
, timer
);
409 #define INDEX(N) ((base->timer_jiffies >> (TVR_BITS + (N) * TVN_BITS)) & TVN_MASK)
412 * __run_timers - run all expired timers (if any) on this CPU.
413 * @base: the timer vector to be processed.
415 * This function cascades all vectors and executes all expired timer
418 static inline void __run_timers(tvec_base_t
*base
)
420 struct timer_list
*timer
;
422 spin_lock_irq(&base
->lock
);
423 while (time_after_eq(jiffies
, base
->timer_jiffies
)) {
424 struct list_head work_list
;
425 struct list_head
*head
= &work_list
;
426 int index
= base
->timer_jiffies
& TVR_MASK
;
432 (!cascade(base
, &base
->tv2
, INDEX(0))) &&
433 (!cascade(base
, &base
->tv3
, INDEX(1))) &&
434 !cascade(base
, &base
->tv4
, INDEX(2)))
435 cascade(base
, &base
->tv5
, INDEX(3));
436 ++base
->timer_jiffies
;
437 list_replace_init(base
->tv1
.vec
+ index
, &work_list
);
438 while (!list_empty(head
)) {
439 void (*fn
)(unsigned long);
442 timer
= list_entry(head
->next
,struct timer_list
,entry
);
443 fn
= timer
->function
;
446 set_running_timer(base
, timer
);
447 detach_timer(timer
, 1);
448 spin_unlock_irq(&base
->lock
);
450 int preempt_count
= preempt_count();
452 if (preempt_count
!= preempt_count()) {
453 printk(KERN_WARNING
"huh, entered %p "
454 "with preempt_count %08x, exited"
461 spin_lock_irq(&base
->lock
);
464 set_running_timer(base
, NULL
);
465 spin_unlock_irq(&base
->lock
);
468 #ifdef CONFIG_NO_IDLE_HZ
470 * Find out when the next timer event is due to happen. This
471 * is used on S/390 to stop all activity when a cpus is idle.
472 * This functions needs to be called disabled.
474 unsigned long next_timer_interrupt(void)
477 struct list_head
*list
;
478 struct timer_list
*nte
;
479 unsigned long expires
;
480 unsigned long hr_expires
= MAX_JIFFY_OFFSET
;
485 hr_delta
= hrtimer_get_next_event();
486 if (hr_delta
.tv64
!= KTIME_MAX
) {
487 struct timespec tsdelta
;
488 tsdelta
= ktime_to_timespec(hr_delta
);
489 hr_expires
= timespec_to_jiffies(&tsdelta
);
491 return hr_expires
+ jiffies
;
493 hr_expires
+= jiffies
;
495 base
= __get_cpu_var(tvec_bases
);
496 spin_lock(&base
->lock
);
497 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
500 /* Look for timer events in tv1. */
501 j
= base
->timer_jiffies
& TVR_MASK
;
503 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
504 expires
= nte
->expires
;
505 if (j
< (base
->timer_jiffies
& TVR_MASK
))
506 list
= base
->tv2
.vec
+ (INDEX(0));
509 j
= (j
+ 1) & TVR_MASK
;
510 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
513 varray
[0] = &base
->tv2
;
514 varray
[1] = &base
->tv3
;
515 varray
[2] = &base
->tv4
;
516 varray
[3] = &base
->tv5
;
517 for (i
= 0; i
< 4; i
++) {
520 if (list_empty(varray
[i
]->vec
+ j
)) {
521 j
= (j
+ 1) & TVN_MASK
;
524 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
525 if (time_before(nte
->expires
, expires
))
526 expires
= nte
->expires
;
527 if (j
< (INDEX(i
)) && i
< 3)
528 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
530 } while (j
!= (INDEX(i
)));
535 * The search wrapped. We need to look at the next list
536 * from next tv element that would cascade into tv element
537 * where we found the timer element.
539 list_for_each_entry(nte
, list
, entry
) {
540 if (time_before(nte
->expires
, expires
))
541 expires
= nte
->expires
;
544 spin_unlock(&base
->lock
);
547 * It can happen that other CPUs service timer IRQs and increment
548 * jiffies, but we have not yet got a local timer tick to process
549 * the timer wheels. In that case, the expiry time can be before
550 * jiffies, but since the high-resolution timer here is relative to
551 * jiffies, the default expression when high-resolution timers are
554 * time_before(MAX_JIFFY_OFFSET + jiffies, expires)
556 * would falsely evaluate to true. If that is the case, just
557 * return jiffies so that we can immediately fire the local timer
559 if (time_before(expires
, jiffies
))
562 if (time_before(hr_expires
, expires
))
569 /******************************************************************/
572 * Timekeeping variables
574 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
575 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
579 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
580 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
581 * at zero at system boot time, so wall_to_monotonic will be negative,
582 * however, we will ALWAYS keep the tv_nsec part positive so we can use
583 * the usual normalization.
585 struct timespec xtime
__attribute__ ((aligned (16)));
586 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
588 EXPORT_SYMBOL(xtime
);
590 /* Don't completely fail for HZ > 500. */
591 int tickadj
= 500/HZ
? : 1; /* microsecs */
595 * phase-lock loop variables
597 /* TIME_ERROR prevents overwriting the CMOS clock */
598 int time_state
= TIME_OK
; /* clock synchronization status */
599 int time_status
= STA_UNSYNC
; /* clock status bits */
600 long time_offset
; /* time adjustment (us) */
601 long time_constant
= 2; /* pll time constant */
602 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
603 long time_precision
= 1; /* clock precision (us) */
604 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
605 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
606 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
607 /* frequency offset (scaled ppm)*/
608 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
609 long time_reftime
; /* time at last adjustment (s) */
611 long time_next_adjust
;
614 * this routine handles the overflow of the microsecond field
616 * The tricky bits of code to handle the accurate clock support
617 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
618 * They were originally developed for SUN and DEC kernels.
619 * All the kudos should go to Dave for this stuff.
622 static void second_overflow(void)
626 /* Bump the maxerror field */
627 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
628 if (time_maxerror
> NTP_PHASE_LIMIT
) {
629 time_maxerror
= NTP_PHASE_LIMIT
;
630 time_status
|= STA_UNSYNC
;
634 * Leap second processing. If in leap-insert state at the end of the
635 * day, the system clock is set back one second; if in leap-delete
636 * state, the system clock is set ahead one second. The microtime()
637 * routine or external clock driver will insure that reported time is
638 * always monotonic. The ugly divides should be replaced.
640 switch (time_state
) {
642 if (time_status
& STA_INS
)
643 time_state
= TIME_INS
;
644 else if (time_status
& STA_DEL
)
645 time_state
= TIME_DEL
;
648 if (xtime
.tv_sec
% 86400 == 0) {
650 wall_to_monotonic
.tv_sec
++;
652 * The timer interpolator will make time change
653 * gradually instead of an immediate jump by one second
655 time_interpolator_update(-NSEC_PER_SEC
);
656 time_state
= TIME_OOP
;
658 printk(KERN_NOTICE
"Clock: inserting leap second "
663 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
665 wall_to_monotonic
.tv_sec
--;
667 * Use of time interpolator for a gradual change of
670 time_interpolator_update(NSEC_PER_SEC
);
671 time_state
= TIME_WAIT
;
673 printk(KERN_NOTICE
"Clock: deleting leap second "
678 time_state
= TIME_WAIT
;
681 if (!(time_status
& (STA_INS
| STA_DEL
)))
682 time_state
= TIME_OK
;
686 * Compute the phase adjustment for the next second. In PLL mode, the
687 * offset is reduced by a fixed factor times the time constant. In FLL
688 * mode the offset is used directly. In either mode, the maximum phase
689 * adjustment for each second is clamped so as to spread the adjustment
690 * over not more than the number of seconds between updates.
693 if (!(time_status
& STA_FLL
))
694 ltemp
= shift_right(ltemp
, SHIFT_KG
+ time_constant
);
695 ltemp
= min(ltemp
, (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
696 ltemp
= max(ltemp
, -(MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
);
697 time_offset
-= ltemp
;
698 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
701 * Compute the frequency estimate and additional phase adjustment due
702 * to frequency error for the next second.
705 time_adj
+= shift_right(ltemp
,(SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
));
709 * Compensate for (HZ==100) != (1 << SHIFT_HZ). Add 25% and 3.125% to
710 * get 128.125; => only 0.125% error (p. 14)
712 time_adj
+= shift_right(time_adj
, 2) + shift_right(time_adj
, 5);
716 * Compensate for (HZ==250) != (1 << SHIFT_HZ). Add 1.5625% and
717 * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
719 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
723 * Compensate for (HZ==1000) != (1 << SHIFT_HZ). Add 1.5625% and
724 * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
726 time_adj
+= shift_right(time_adj
, 6) + shift_right(time_adj
, 7);
731 * Returns how many microseconds we need to add to xtime this tick
732 * in doing an adjustment requested with adjtime.
734 static long adjtime_adjustment(void)
736 long time_adjust_step
;
738 time_adjust_step
= time_adjust
;
739 if (time_adjust_step
) {
741 * We are doing an adjtime thing. Prepare time_adjust_step to
742 * be within bounds. Note that a positive time_adjust means we
743 * want the clock to run faster.
745 * Limit the amount of the step to be in the range
746 * -tickadj .. +tickadj
748 time_adjust_step
= min(time_adjust_step
, (long)tickadj
);
749 time_adjust_step
= max(time_adjust_step
, (long)-tickadj
);
751 return time_adjust_step
;
754 /* in the NTP reference this is called "hardclock()" */
755 static void update_ntp_one_tick(void)
757 long time_adjust_step
;
759 time_adjust_step
= adjtime_adjustment();
760 if (time_adjust_step
)
761 /* Reduce by this step the amount of time left */
762 time_adjust
-= time_adjust_step
;
764 /* Changes by adjtime() do not take effect till next tick. */
765 if (time_next_adjust
!= 0) {
766 time_adjust
= time_next_adjust
;
767 time_next_adjust
= 0;
772 * Return how long ticks are at the moment, that is, how much time
773 * update_wall_time_one_tick will add to xtime next time we call it
774 * (assuming no calls to do_adjtimex in the meantime).
775 * The return value is in fixed-point nanoseconds shifted by the
776 * specified number of bits to the right of the binary point.
777 * This function has no side-effects.
779 u64
current_tick_length(void)
784 /* calculate the finest interval NTP will allow.
785 * ie: nanosecond value shifted by (SHIFT_SCALE - 10)
787 delta_nsec
= tick_nsec
+ adjtime_adjustment() * 1000;
788 ret
= (u64
)delta_nsec
<< TICK_LENGTH_SHIFT
;
789 ret
+= (s64
)time_adj
<< (TICK_LENGTH_SHIFT
- (SHIFT_SCALE
- 10));
794 /* XXX - all of this timekeeping code should be later moved to time.c */
795 #include <linux/clocksource.h>
796 static struct clocksource
*clock
; /* pointer to current clocksource */
798 #ifdef CONFIG_GENERIC_TIME
800 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
802 * private function, must hold xtime_lock lock when being
803 * called. Returns the number of nanoseconds since the
804 * last call to update_wall_time() (adjusted by NTP scaling)
806 static inline s64
__get_nsec_offset(void)
808 cycle_t cycle_now
, cycle_delta
;
811 /* read clocksource: */
812 cycle_now
= clocksource_read(clock
);
814 /* calculate the delta since the last update_wall_time: */
815 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
817 /* convert to nanoseconds: */
818 ns_offset
= cyc2ns(clock
, cycle_delta
);
824 * __get_realtime_clock_ts - Returns the time of day in a timespec
825 * @ts: pointer to the timespec to be set
827 * Returns the time of day in a timespec. Used by
828 * do_gettimeofday() and get_realtime_clock_ts().
830 static inline void __get_realtime_clock_ts(struct timespec
*ts
)
836 seq
= read_seqbegin(&xtime_lock
);
839 nsecs
= __get_nsec_offset();
841 } while (read_seqretry(&xtime_lock
, seq
));
843 timespec_add_ns(ts
, nsecs
);
847 * getnstimeofday - Returns the time of day in a timespec
848 * @ts: pointer to the timespec to be set
850 * Returns the time of day in a timespec.
852 void getnstimeofday(struct timespec
*ts
)
854 __get_realtime_clock_ts(ts
);
857 EXPORT_SYMBOL(getnstimeofday
);
860 * do_gettimeofday - Returns the time of day in a timeval
861 * @tv: pointer to the timeval to be set
863 * NOTE: Users should be converted to using get_realtime_clock_ts()
865 void do_gettimeofday(struct timeval
*tv
)
869 __get_realtime_clock_ts(&now
);
870 tv
->tv_sec
= now
.tv_sec
;
871 tv
->tv_usec
= now
.tv_nsec
/1000;
874 EXPORT_SYMBOL(do_gettimeofday
);
876 * do_settimeofday - Sets the time of day
877 * @tv: pointer to the timespec variable containing the new time
879 * Sets the time of day to the new time and update NTP and notify hrtimers
881 int do_settimeofday(struct timespec
*tv
)
884 time_t wtm_sec
, sec
= tv
->tv_sec
;
885 long wtm_nsec
, nsec
= tv
->tv_nsec
;
887 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
890 write_seqlock_irqsave(&xtime_lock
, flags
);
892 nsec
-= __get_nsec_offset();
894 wtm_sec
= wall_to_monotonic
.tv_sec
+ (xtime
.tv_sec
- sec
);
895 wtm_nsec
= wall_to_monotonic
.tv_nsec
+ (xtime
.tv_nsec
- nsec
);
897 set_normalized_timespec(&xtime
, sec
, nsec
);
898 set_normalized_timespec(&wall_to_monotonic
, wtm_sec
, wtm_nsec
);
903 write_sequnlock_irqrestore(&xtime_lock
, flags
);
905 /* signal hrtimers about time change */
911 EXPORT_SYMBOL(do_settimeofday
);
914 * change_clocksource - Swaps clocksources if a new one is available
916 * Accumulates current time interval and initializes new clocksource
918 static int change_clocksource(void)
920 struct clocksource
*new;
923 new = clocksource_get_next();
925 now
= clocksource_read(new);
926 nsec
= __get_nsec_offset();
927 timespec_add_ns(&xtime
, nsec
);
930 clock
->cycle_last
= now
;
931 printk(KERN_INFO
"Time: %s clocksource has been installed.\n",
934 } else if (clock
->update_callback
) {
935 return clock
->update_callback();
940 #define change_clocksource() (0)
944 * timeofday_is_continuous - check to see if timekeeping is free running
946 int timekeeping_is_continuous(void)
952 seq
= read_seqbegin(&xtime_lock
);
954 ret
= clock
->is_continuous
;
956 } while (read_seqretry(&xtime_lock
, seq
));
962 * timekeeping_init - Initializes the clocksource and common timekeeping values
964 void __init
timekeeping_init(void)
968 write_seqlock_irqsave(&xtime_lock
, flags
);
969 clock
= clocksource_get_next();
970 clocksource_calculate_interval(clock
, tick_nsec
);
971 clock
->cycle_last
= clocksource_read(clock
);
973 write_sequnlock_irqrestore(&xtime_lock
, flags
);
977 static int timekeeping_suspended
;
979 * timekeeping_resume - Resumes the generic timekeeping subsystem.
982 * This is for the generic clocksource timekeeping.
983 * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
984 * still managed by arch specific suspend/resume code.
986 static int timekeeping_resume(struct sys_device
*dev
)
990 write_seqlock_irqsave(&xtime_lock
, flags
);
991 /* restart the last cycle value */
992 clock
->cycle_last
= clocksource_read(clock
);
994 timekeeping_suspended
= 0;
995 write_sequnlock_irqrestore(&xtime_lock
, flags
);
999 static int timekeeping_suspend(struct sys_device
*dev
, pm_message_t state
)
1001 unsigned long flags
;
1003 write_seqlock_irqsave(&xtime_lock
, flags
);
1004 timekeeping_suspended
= 1;
1005 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1009 /* sysfs resume/suspend bits for timekeeping */
1010 static struct sysdev_class timekeeping_sysclass
= {
1011 .resume
= timekeeping_resume
,
1012 .suspend
= timekeeping_suspend
,
1013 set_kset_name("timekeeping"),
1016 static struct sys_device device_timer
= {
1018 .cls
= &timekeeping_sysclass
,
1021 static int __init
timekeeping_init_device(void)
1023 int error
= sysdev_class_register(&timekeeping_sysclass
);
1025 error
= sysdev_register(&device_timer
);
1029 device_initcall(timekeeping_init_device
);
1032 * If the error is already larger, we look ahead even further
1033 * to compensate for late or lost adjustments.
1035 static __always_inline
int clocksource_bigadjust(s64 error
, s64
*interval
, s64
*offset
)
1038 u32 look_ahead
, adj
;
1042 * Use the current error value to determine how much to look ahead.
1043 * The larger the error the slower we adjust for it to avoid problems
1044 * with losing too many ticks, otherwise we would overadjust and
1045 * produce an even larger error. The smaller the adjustment the
1046 * faster we try to adjust for it, as lost ticks can do less harm
1047 * here. This is tuned so that an error of about 1 msec is adusted
1048 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1050 error2
= clock
->error
>> (TICK_LENGTH_SHIFT
+ 22 - 2 * SHIFT_HZ
);
1051 error2
= abs(error2
);
1052 for (look_ahead
= 0; error2
> 0; look_ahead
++)
1056 * Now calculate the error in (1 << look_ahead) ticks, but first
1057 * remove the single look ahead already included in the error.
1059 tick_error
= current_tick_length() >> (TICK_LENGTH_SHIFT
- clock
->shift
+ 1);
1060 tick_error
-= clock
->xtime_interval
>> 1;
1061 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
1063 /* Finally calculate the adjustment shift value. */
1068 *interval
= -*interval
;
1072 for (adj
= 0; error
> i
; adj
++)
1081 * Adjust the multiplier to reduce the error value,
1082 * this is optimized for the most common adjustments of -1,0,1,
1083 * for other values we can do a bit more work.
1085 static void clocksource_adjust(struct clocksource
*clock
, s64 offset
)
1087 s64 error
, interval
= clock
->cycle_interval
;
1090 error
= clock
->error
>> (TICK_LENGTH_SHIFT
- clock
->shift
- 1);
1091 if (error
> interval
) {
1093 if (likely(error
<= interval
))
1096 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1097 } else if (error
< -interval
) {
1099 if (likely(error
>= -interval
)) {
1101 interval
= -interval
;
1104 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
1109 clock
->xtime_interval
+= interval
;
1110 clock
->xtime_nsec
-= offset
;
1111 clock
->error
-= (interval
- offset
) << (TICK_LENGTH_SHIFT
- clock
->shift
);
1115 * update_wall_time - Uses the current clocksource to increment the wall time
1117 * Called from the timer interrupt, must hold a write on xtime_lock.
1119 static void update_wall_time(void)
1123 /* Make sure we're fully resumed: */
1124 if (unlikely(timekeeping_suspended
))
1127 #ifdef CONFIG_GENERIC_TIME
1128 offset
= (clocksource_read(clock
) - clock
->cycle_last
) & clock
->mask
;
1130 offset
= clock
->cycle_interval
;
1132 clock
->xtime_nsec
+= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1134 /* normally this loop will run just once, however in the
1135 * case of lost or late ticks, it will accumulate correctly.
1137 while (offset
>= clock
->cycle_interval
) {
1138 /* accumulate one interval */
1139 clock
->xtime_nsec
+= clock
->xtime_interval
;
1140 clock
->cycle_last
+= clock
->cycle_interval
;
1141 offset
-= clock
->cycle_interval
;
1143 if (clock
->xtime_nsec
>= (u64
)NSEC_PER_SEC
<< clock
->shift
) {
1144 clock
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< clock
->shift
;
1149 /* interpolator bits */
1150 time_interpolator_update(clock
->xtime_interval
1152 /* increment the NTP state machine */
1153 update_ntp_one_tick();
1155 /* accumulate error between NTP and clock interval */
1156 clock
->error
+= current_tick_length();
1157 clock
->error
-= clock
->xtime_interval
<< (TICK_LENGTH_SHIFT
- clock
->shift
);
1160 /* correct the clock when NTP error is too big */
1161 clocksource_adjust(clock
, offset
);
1163 /* store full nanoseconds into xtime */
1164 xtime
.tv_nsec
= (s64
)clock
->xtime_nsec
>> clock
->shift
;
1165 clock
->xtime_nsec
-= (s64
)xtime
.tv_nsec
<< clock
->shift
;
1167 /* check to see if there is a new clocksource to use */
1168 if (change_clocksource()) {
1170 clock
->xtime_nsec
= 0;
1171 clocksource_calculate_interval(clock
, tick_nsec
);
1176 * Called from the timer interrupt handler to charge one tick to the current
1177 * process. user_tick is 1 if the tick is user time, 0 for system.
1179 void update_process_times(int user_tick
)
1181 struct task_struct
*p
= current
;
1182 int cpu
= smp_processor_id();
1184 /* Note: this timer irq context must be accounted for as well. */
1186 account_user_time(p
, jiffies_to_cputime(1));
1188 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
1190 if (rcu_pending(cpu
))
1191 rcu_check_callbacks(cpu
, user_tick
);
1193 run_posix_cpu_timers(p
);
1197 * Nr of active tasks - counted in fixed-point numbers
1199 static unsigned long count_active_tasks(void)
1201 return nr_active() * FIXED_1
;
1205 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
1206 * imply that avenrun[] is the standard name for this kind of thing.
1207 * Nothing else seems to be standardized: the fractional size etc
1208 * all seem to differ on different machines.
1210 * Requires xtime_lock to access.
1212 unsigned long avenrun
[3];
1214 EXPORT_SYMBOL(avenrun
);
1217 * calc_load - given tick count, update the avenrun load estimates.
1218 * This is called while holding a write_lock on xtime_lock.
1220 static inline void calc_load(unsigned long ticks
)
1222 unsigned long active_tasks
; /* fixed-point */
1223 static int count
= LOAD_FREQ
;
1225 active_tasks
= count_active_tasks();
1226 for (count
-= ticks
; count
< 0; count
+= LOAD_FREQ
) {
1227 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
1228 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
1229 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
1233 /* jiffies at the most recent update of wall time */
1234 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
1237 * This read-write spinlock protects us from races in SMP while
1238 * playing with xtime and avenrun.
1240 #ifndef ARCH_HAVE_XTIME_LOCK
1241 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
1243 EXPORT_SYMBOL(xtime_lock
);
1247 * This function runs timers and the timer-tq in bottom half context.
1249 static void run_timer_softirq(struct softirq_action
*h
)
1251 tvec_base_t
*base
= __get_cpu_var(tvec_bases
);
1253 hrtimer_run_queues();
1254 if (time_after_eq(jiffies
, base
->timer_jiffies
))
1259 * Called by the local, per-CPU timer interrupt on SMP.
1261 void run_local_timers(void)
1263 raise_softirq(TIMER_SOFTIRQ
);
1268 * Called by the timer interrupt. xtime_lock must already be taken
1271 static inline void update_times(unsigned long ticks
)
1273 wall_jiffies
+= ticks
;
1279 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1280 * without sampling the sequence number in xtime_lock.
1281 * jiffies is defined in the linker script...
1284 void do_timer(unsigned long ticks
)
1286 jiffies_64
+= ticks
;
1287 update_times(ticks
);
1290 #ifdef __ARCH_WANT_SYS_ALARM
1293 * For backwards compatibility? This can be done in libc so Alpha
1294 * and all newer ports shouldn't need it.
1296 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
1298 return alarm_setitimer(seconds
);
1306 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
1307 * should be moved into arch/i386 instead?
1311 * sys_getpid - return the thread group id of the current process
1313 * Note, despite the name, this returns the tgid not the pid. The tgid and
1314 * the pid are identical unless CLONE_THREAD was specified on clone() in
1315 * which case the tgid is the same in all threads of the same group.
1317 * This is SMP safe as current->tgid does not change.
1319 asmlinkage
long sys_getpid(void)
1321 return current
->tgid
;
1325 * Accessing ->real_parent is not SMP-safe, it could
1326 * change from under us. However, we can use a stale
1327 * value of ->real_parent under rcu_read_lock(), see
1328 * release_task()->call_rcu(delayed_put_task_struct).
1330 asmlinkage
long sys_getppid(void)
1335 pid
= rcu_dereference(current
->real_parent
)->tgid
;
1341 asmlinkage
long sys_getuid(void)
1343 /* Only we change this so SMP safe */
1344 return current
->uid
;
1347 asmlinkage
long sys_geteuid(void)
1349 /* Only we change this so SMP safe */
1350 return current
->euid
;
1353 asmlinkage
long sys_getgid(void)
1355 /* Only we change this so SMP safe */
1356 return current
->gid
;
1359 asmlinkage
long sys_getegid(void)
1361 /* Only we change this so SMP safe */
1362 return current
->egid
;
1367 static void process_timeout(unsigned long __data
)
1369 wake_up_process((struct task_struct
*)__data
);
1373 * schedule_timeout - sleep until timeout
1374 * @timeout: timeout value in jiffies
1376 * Make the current task sleep until @timeout jiffies have
1377 * elapsed. The routine will return immediately unless
1378 * the current task state has been set (see set_current_state()).
1380 * You can set the task state as follows -
1382 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1383 * pass before the routine returns. The routine will return 0
1385 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1386 * delivered to the current task. In this case the remaining time
1387 * in jiffies will be returned, or 0 if the timer expired in time
1389 * The current task state is guaranteed to be TASK_RUNNING when this
1392 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1393 * the CPU away without a bound on the timeout. In this case the return
1394 * value will be %MAX_SCHEDULE_TIMEOUT.
1396 * In all cases the return value is guaranteed to be non-negative.
1398 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1400 struct timer_list timer
;
1401 unsigned long expire
;
1405 case MAX_SCHEDULE_TIMEOUT
:
1407 * These two special cases are useful to be comfortable
1408 * in the caller. Nothing more. We could take
1409 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1410 * but I' d like to return a valid offset (>=0) to allow
1411 * the caller to do everything it want with the retval.
1417 * Another bit of PARANOID. Note that the retval will be
1418 * 0 since no piece of kernel is supposed to do a check
1419 * for a negative retval of schedule_timeout() (since it
1420 * should never happens anyway). You just have the printk()
1421 * that will tell you if something is gone wrong and where.
1425 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1426 "value %lx from %p\n", timeout
,
1427 __builtin_return_address(0));
1428 current
->state
= TASK_RUNNING
;
1433 expire
= timeout
+ jiffies
;
1435 setup_timer(&timer
, process_timeout
, (unsigned long)current
);
1436 __mod_timer(&timer
, expire
);
1438 del_singleshot_timer_sync(&timer
);
1440 timeout
= expire
- jiffies
;
1443 return timeout
< 0 ? 0 : timeout
;
1445 EXPORT_SYMBOL(schedule_timeout
);
1448 * We can use __set_current_state() here because schedule_timeout() calls
1449 * schedule() unconditionally.
1451 signed long __sched
schedule_timeout_interruptible(signed long timeout
)
1453 __set_current_state(TASK_INTERRUPTIBLE
);
1454 return schedule_timeout(timeout
);
1456 EXPORT_SYMBOL(schedule_timeout_interruptible
);
1458 signed long __sched
schedule_timeout_uninterruptible(signed long timeout
)
1460 __set_current_state(TASK_UNINTERRUPTIBLE
);
1461 return schedule_timeout(timeout
);
1463 EXPORT_SYMBOL(schedule_timeout_uninterruptible
);
1465 /* Thread ID - the internal kernel "pid" */
1466 asmlinkage
long sys_gettid(void)
1468 return current
->pid
;
1472 * sys_sysinfo - fill in sysinfo struct
1473 * @info: pointer to buffer to fill
1475 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1478 unsigned long mem_total
, sav_total
;
1479 unsigned int mem_unit
, bitcount
;
1482 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1486 seq
= read_seqbegin(&xtime_lock
);
1489 * This is annoying. The below is the same thing
1490 * posix_get_clock_monotonic() does, but it wants to
1491 * take the lock which we want to cover the loads stuff
1495 getnstimeofday(&tp
);
1496 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1497 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1498 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1499 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1502 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1504 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1505 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1506 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1508 val
.procs
= nr_threads
;
1509 } while (read_seqretry(&xtime_lock
, seq
));
1515 * If the sum of all the available memory (i.e. ram + swap)
1516 * is less than can be stored in a 32 bit unsigned long then
1517 * we can be binary compatible with 2.2.x kernels. If not,
1518 * well, in that case 2.2.x was broken anyways...
1520 * -Erik Andersen <andersee@debian.org>
1523 mem_total
= val
.totalram
+ val
.totalswap
;
1524 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1527 mem_unit
= val
.mem_unit
;
1528 while (mem_unit
> 1) {
1531 sav_total
= mem_total
;
1533 if (mem_total
< sav_total
)
1538 * If mem_total did not overflow, multiply all memory values by
1539 * val.mem_unit and set it to 1. This leaves things compatible
1540 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1545 val
.totalram
<<= bitcount
;
1546 val
.freeram
<<= bitcount
;
1547 val
.sharedram
<<= bitcount
;
1548 val
.bufferram
<<= bitcount
;
1549 val
.totalswap
<<= bitcount
;
1550 val
.freeswap
<<= bitcount
;
1551 val
.totalhigh
<<= bitcount
;
1552 val
.freehigh
<<= bitcount
;
1555 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1562 * lockdep: we want to track each per-CPU base as a separate lock-class,
1563 * but timer-bases are kmalloc()-ed, so we need to attach separate
1566 static struct lock_class_key base_lock_keys
[NR_CPUS
];
1568 static int __devinit
init_timers_cpu(int cpu
)
1572 static char __devinitdata tvec_base_done
[NR_CPUS
];
1574 if (!tvec_base_done
[cpu
]) {
1575 static char boot_done
;
1579 * The APs use this path later in boot
1581 base
= kmalloc_node(sizeof(*base
), GFP_KERNEL
,
1585 memset(base
, 0, sizeof(*base
));
1586 per_cpu(tvec_bases
, cpu
) = base
;
1589 * This is for the boot CPU - we use compile-time
1590 * static initialisation because per-cpu memory isn't
1591 * ready yet and because the memory allocators are not
1592 * initialised either.
1595 base
= &boot_tvec_bases
;
1597 tvec_base_done
[cpu
] = 1;
1599 base
= per_cpu(tvec_bases
, cpu
);
1602 spin_lock_init(&base
->lock
);
1603 lockdep_set_class(&base
->lock
, base_lock_keys
+ cpu
);
1605 for (j
= 0; j
< TVN_SIZE
; j
++) {
1606 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1607 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1608 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1609 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1611 for (j
= 0; j
< TVR_SIZE
; j
++)
1612 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1614 base
->timer_jiffies
= jiffies
;
1618 #ifdef CONFIG_HOTPLUG_CPU
1619 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1621 struct timer_list
*timer
;
1623 while (!list_empty(head
)) {
1624 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1625 detach_timer(timer
, 0);
1626 timer
->base
= new_base
;
1627 internal_add_timer(new_base
, timer
);
1631 static void __devinit
migrate_timers(int cpu
)
1633 tvec_base_t
*old_base
;
1634 tvec_base_t
*new_base
;
1637 BUG_ON(cpu_online(cpu
));
1638 old_base
= per_cpu(tvec_bases
, cpu
);
1639 new_base
= get_cpu_var(tvec_bases
);
1641 local_irq_disable();
1642 spin_lock(&new_base
->lock
);
1643 spin_lock(&old_base
->lock
);
1645 BUG_ON(old_base
->running_timer
);
1647 for (i
= 0; i
< TVR_SIZE
; i
++)
1648 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1649 for (i
= 0; i
< TVN_SIZE
; i
++) {
1650 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1651 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1652 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1653 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1656 spin_unlock(&old_base
->lock
);
1657 spin_unlock(&new_base
->lock
);
1659 put_cpu_var(tvec_bases
);
1661 #endif /* CONFIG_HOTPLUG_CPU */
1663 static int __cpuinit
timer_cpu_notify(struct notifier_block
*self
,
1664 unsigned long action
, void *hcpu
)
1666 long cpu
= (long)hcpu
;
1668 case CPU_UP_PREPARE
:
1669 if (init_timers_cpu(cpu
) < 0)
1672 #ifdef CONFIG_HOTPLUG_CPU
1674 migrate_timers(cpu
);
1683 static struct notifier_block __cpuinitdata timers_nb
= {
1684 .notifier_call
= timer_cpu_notify
,
1688 void __init
init_timers(void)
1690 int err
= timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1691 (void *)(long)smp_processor_id());
1693 BUG_ON(err
== NOTIFY_BAD
);
1694 register_cpu_notifier(&timers_nb
);
1695 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1698 #ifdef CONFIG_TIME_INTERPOLATION
1700 struct time_interpolator
*time_interpolator __read_mostly
;
1701 static struct time_interpolator
*time_interpolator_list __read_mostly
;
1702 static DEFINE_SPINLOCK(time_interpolator_lock
);
1704 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1706 unsigned long (*x
)(void);
1710 case TIME_SOURCE_FUNCTION
:
1711 x
= time_interpolator
->addr
;
1714 case TIME_SOURCE_MMIO64
:
1715 return readq_relaxed((void __iomem
*)time_interpolator
->addr
);
1717 case TIME_SOURCE_MMIO32
:
1718 return readl_relaxed((void __iomem
*)time_interpolator
->addr
);
1720 default: return get_cycles();
1724 static inline u64
time_interpolator_get_counter(int writelock
)
1726 unsigned int src
= time_interpolator
->source
;
1728 if (time_interpolator
->jitter
)
1734 lcycle
= time_interpolator
->last_cycle
;
1735 now
= time_interpolator_get_cycles(src
);
1736 if (lcycle
&& time_after(lcycle
, now
))
1739 /* When holding the xtime write lock, there's no need
1740 * to add the overhead of the cmpxchg. Readers are
1741 * force to retry until the write lock is released.
1744 time_interpolator
->last_cycle
= now
;
1747 /* Keep track of the last timer value returned. The use of cmpxchg here
1748 * will cause contention in an SMP environment.
1750 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1754 return time_interpolator_get_cycles(src
);
1757 void time_interpolator_reset(void)
1759 time_interpolator
->offset
= 0;
1760 time_interpolator
->last_counter
= time_interpolator_get_counter(1);
1763 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1765 unsigned long time_interpolator_get_offset(void)
1767 /* If we do not have a time interpolator set up then just return zero */
1768 if (!time_interpolator
)
1771 return time_interpolator
->offset
+
1772 GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator
);
1775 #define INTERPOLATOR_ADJUST 65536
1776 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1778 static void time_interpolator_update(long delta_nsec
)
1781 unsigned long offset
;
1783 /* If there is no time interpolator set up then do nothing */
1784 if (!time_interpolator
)
1788 * The interpolator compensates for late ticks by accumulating the late
1789 * time in time_interpolator->offset. A tick earlier than expected will
1790 * lead to a reset of the offset and a corresponding jump of the clock
1791 * forward. Again this only works if the interpolator clock is running
1792 * slightly slower than the regular clock and the tuning logic insures
1796 counter
= time_interpolator_get_counter(1);
1797 offset
= time_interpolator
->offset
+
1798 GET_TI_NSECS(counter
, time_interpolator
);
1800 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1801 time_interpolator
->offset
= offset
- delta_nsec
;
1803 time_interpolator
->skips
++;
1804 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1805 time_interpolator
->offset
= 0;
1807 time_interpolator
->last_counter
= counter
;
1809 /* Tuning logic for time interpolator invoked every minute or so.
1810 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1811 * Increase interpolator clock speed if we skip too much time.
1813 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1815 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> tick_nsec
)
1816 time_interpolator
->nsec_per_cyc
--;
1817 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1818 time_interpolator
->nsec_per_cyc
++;
1819 time_interpolator
->skips
= 0;
1820 time_interpolator
->ns_skipped
= 0;
1825 is_better_time_interpolator(struct time_interpolator
*new)
1827 if (!time_interpolator
)
1829 return new->frequency
> 2*time_interpolator
->frequency
||
1830 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1834 register_time_interpolator(struct time_interpolator
*ti
)
1836 unsigned long flags
;
1839 BUG_ON(ti
->frequency
== 0 || ti
->mask
== 0);
1841 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1842 spin_lock(&time_interpolator_lock
);
1843 write_seqlock_irqsave(&xtime_lock
, flags
);
1844 if (is_better_time_interpolator(ti
)) {
1845 time_interpolator
= ti
;
1846 time_interpolator_reset();
1848 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1850 ti
->next
= time_interpolator_list
;
1851 time_interpolator_list
= ti
;
1852 spin_unlock(&time_interpolator_lock
);
1856 unregister_time_interpolator(struct time_interpolator
*ti
)
1858 struct time_interpolator
*curr
, **prev
;
1859 unsigned long flags
;
1861 spin_lock(&time_interpolator_lock
);
1862 prev
= &time_interpolator_list
;
1863 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1871 write_seqlock_irqsave(&xtime_lock
, flags
);
1872 if (ti
== time_interpolator
) {
1873 /* we lost the best time-interpolator: */
1874 time_interpolator
= NULL
;
1875 /* find the next-best interpolator */
1876 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1877 if (is_better_time_interpolator(curr
))
1878 time_interpolator
= curr
;
1879 time_interpolator_reset();
1881 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1882 spin_unlock(&time_interpolator_lock
);
1884 #endif /* CONFIG_TIME_INTERPOLATION */
1887 * msleep - sleep safely even with waitqueue interruptions
1888 * @msecs: Time in milliseconds to sleep for
1890 void msleep(unsigned int msecs
)
1892 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1895 timeout
= schedule_timeout_uninterruptible(timeout
);
1898 EXPORT_SYMBOL(msleep
);
1901 * msleep_interruptible - sleep waiting for signals
1902 * @msecs: Time in milliseconds to sleep for
1904 unsigned long msleep_interruptible(unsigned int msecs
)
1906 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1908 while (timeout
&& !signal_pending(current
))
1909 timeout
= schedule_timeout_interruptible(timeout
);
1910 return jiffies_to_msecs(timeout
);
1913 EXPORT_SYMBOL(msleep_interruptible
);