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 u32 preempt_count
= preempt_count();
494 if (preempt_count
!= preempt_count()) {
495 printk("huh, entered %p with %08x, exited with %08x?\n", fn
, preempt_count
, preempt_count());
499 spin_lock_irq(&base
->t_base
.lock
);
502 set_running_timer(base
, NULL
);
503 spin_unlock_irq(&base
->t_base
.lock
);
506 #ifdef CONFIG_NO_IDLE_HZ
508 * Find out when the next timer event is due to happen. This
509 * is used on S/390 to stop all activity when a cpus is idle.
510 * This functions needs to be called disabled.
512 unsigned long next_timer_interrupt(void)
515 struct list_head
*list
;
516 struct timer_list
*nte
;
517 unsigned long expires
;
521 base
= &__get_cpu_var(tvec_bases
);
522 spin_lock(&base
->t_base
.lock
);
523 expires
= base
->timer_jiffies
+ (LONG_MAX
>> 1);
526 /* Look for timer events in tv1. */
527 j
= base
->timer_jiffies
& TVR_MASK
;
529 list_for_each_entry(nte
, base
->tv1
.vec
+ j
, entry
) {
530 expires
= nte
->expires
;
531 if (j
< (base
->timer_jiffies
& TVR_MASK
))
532 list
= base
->tv2
.vec
+ (INDEX(0));
535 j
= (j
+ 1) & TVR_MASK
;
536 } while (j
!= (base
->timer_jiffies
& TVR_MASK
));
539 varray
[0] = &base
->tv2
;
540 varray
[1] = &base
->tv3
;
541 varray
[2] = &base
->tv4
;
542 varray
[3] = &base
->tv5
;
543 for (i
= 0; i
< 4; i
++) {
546 if (list_empty(varray
[i
]->vec
+ j
)) {
547 j
= (j
+ 1) & TVN_MASK
;
550 list_for_each_entry(nte
, varray
[i
]->vec
+ j
, entry
)
551 if (time_before(nte
->expires
, expires
))
552 expires
= nte
->expires
;
553 if (j
< (INDEX(i
)) && i
< 3)
554 list
= varray
[i
+ 1]->vec
+ (INDEX(i
+ 1));
556 } while (j
!= (INDEX(i
)));
561 * The search wrapped. We need to look at the next list
562 * from next tv element that would cascade into tv element
563 * where we found the timer element.
565 list_for_each_entry(nte
, list
, entry
) {
566 if (time_before(nte
->expires
, expires
))
567 expires
= nte
->expires
;
570 spin_unlock(&base
->t_base
.lock
);
575 /******************************************************************/
578 * Timekeeping variables
580 unsigned long tick_usec
= TICK_USEC
; /* USER_HZ period (usec) */
581 unsigned long tick_nsec
= TICK_NSEC
; /* ACTHZ period (nsec) */
585 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
586 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
587 * at zero at system boot time, so wall_to_monotonic will be negative,
588 * however, we will ALWAYS keep the tv_nsec part positive so we can use
589 * the usual normalization.
591 struct timespec xtime
__attribute__ ((aligned (16)));
592 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
594 EXPORT_SYMBOL(xtime
);
596 /* Don't completely fail for HZ > 500. */
597 int tickadj
= 500/HZ
? : 1; /* microsecs */
601 * phase-lock loop variables
603 /* TIME_ERROR prevents overwriting the CMOS clock */
604 int time_state
= TIME_OK
; /* clock synchronization status */
605 int time_status
= STA_UNSYNC
; /* clock status bits */
606 long time_offset
; /* time adjustment (us) */
607 long time_constant
= 2; /* pll time constant */
608 long time_tolerance
= MAXFREQ
; /* frequency tolerance (ppm) */
609 long time_precision
= 1; /* clock precision (us) */
610 long time_maxerror
= NTP_PHASE_LIMIT
; /* maximum error (us) */
611 long time_esterror
= NTP_PHASE_LIMIT
; /* estimated error (us) */
612 static long time_phase
; /* phase offset (scaled us) */
613 long time_freq
= (((NSEC_PER_SEC
+ HZ
/2) % HZ
- HZ
/2) << SHIFT_USEC
) / NSEC_PER_USEC
;
614 /* frequency offset (scaled ppm)*/
615 static long time_adj
; /* tick adjust (scaled 1 / HZ) */
616 long time_reftime
; /* time at last adjustment (s) */
618 long time_next_adjust
;
621 * this routine handles the overflow of the microsecond field
623 * The tricky bits of code to handle the accurate clock support
624 * were provided by Dave Mills (Mills@UDEL.EDU) of NTP fame.
625 * They were originally developed for SUN and DEC kernels.
626 * All the kudos should go to Dave for this stuff.
629 static void second_overflow(void)
633 /* Bump the maxerror field */
634 time_maxerror
+= time_tolerance
>> SHIFT_USEC
;
635 if ( time_maxerror
> NTP_PHASE_LIMIT
) {
636 time_maxerror
= NTP_PHASE_LIMIT
;
637 time_status
|= STA_UNSYNC
;
641 * Leap second processing. If in leap-insert state at
642 * the end of the day, the system clock is set back one
643 * second; if in leap-delete state, the system clock is
644 * set ahead one second. The microtime() routine or
645 * external clock driver will insure that reported time
646 * is always monotonic. The ugly divides should be
649 switch (time_state
) {
652 if (time_status
& STA_INS
)
653 time_state
= TIME_INS
;
654 else if (time_status
& STA_DEL
)
655 time_state
= TIME_DEL
;
659 if (xtime
.tv_sec
% 86400 == 0) {
661 wall_to_monotonic
.tv_sec
++;
662 /* The timer interpolator will make time change gradually instead
663 * of an immediate jump by one second.
665 time_interpolator_update(-NSEC_PER_SEC
);
666 time_state
= TIME_OOP
;
668 printk(KERN_NOTICE
"Clock: inserting leap second 23:59:60 UTC\n");
673 if ((xtime
.tv_sec
+ 1) % 86400 == 0) {
675 wall_to_monotonic
.tv_sec
--;
676 /* Use of time interpolator for a gradual change of time */
677 time_interpolator_update(NSEC_PER_SEC
);
678 time_state
= TIME_WAIT
;
680 printk(KERN_NOTICE
"Clock: deleting leap second 23:59:59 UTC\n");
685 time_state
= TIME_WAIT
;
689 if (!(time_status
& (STA_INS
| STA_DEL
)))
690 time_state
= TIME_OK
;
694 * Compute the phase adjustment for the next second. In
695 * PLL mode, the offset is reduced by a fixed factor
696 * times the time constant. In FLL mode the offset is
697 * used directly. In either mode, the maximum phase
698 * adjustment for each second is clamped so as to spread
699 * the adjustment over not more than the number of
700 * seconds between updates.
702 if (time_offset
< 0) {
703 ltemp
= -time_offset
;
704 if (!(time_status
& STA_FLL
))
705 ltemp
>>= SHIFT_KG
+ time_constant
;
706 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
707 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
708 time_offset
+= ltemp
;
709 time_adj
= -ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
712 if (!(time_status
& STA_FLL
))
713 ltemp
>>= SHIFT_KG
+ time_constant
;
714 if (ltemp
> (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
)
715 ltemp
= (MAXPHASE
/ MINSEC
) << SHIFT_UPDATE
;
716 time_offset
-= ltemp
;
717 time_adj
= ltemp
<< (SHIFT_SCALE
- SHIFT_HZ
- SHIFT_UPDATE
);
721 * Compute the frequency estimate and additional phase
722 * adjustment due to frequency error for the next
723 * second. When the PPS signal is engaged, gnaw on the
724 * watchdog counter and update the frequency computed by
725 * the pll and the PPS signal.
728 if (pps_valid
== PPS_VALID
) { /* PPS signal lost */
729 pps_jitter
= MAXTIME
;
730 pps_stabil
= MAXFREQ
;
731 time_status
&= ~(STA_PPSSIGNAL
| STA_PPSJITTER
|
732 STA_PPSWANDER
| STA_PPSERROR
);
734 ltemp
= time_freq
+ pps_freq
;
736 time_adj
-= -ltemp
>>
737 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
740 (SHIFT_USEC
+ SHIFT_HZ
- SHIFT_SCALE
);
743 /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
744 * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
747 time_adj
-= (-time_adj
>> 2) + (-time_adj
>> 5);
749 time_adj
+= (time_adj
>> 2) + (time_adj
>> 5);
752 /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
753 * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
756 time_adj
-= (-time_adj
>> 6) + (-time_adj
>> 7);
758 time_adj
+= (time_adj
>> 6) + (time_adj
>> 7);
762 /* in the NTP reference this is called "hardclock()" */
763 static void update_wall_time_one_tick(void)
765 long time_adjust_step
, delta_nsec
;
767 if ( (time_adjust_step
= time_adjust
) != 0 ) {
768 /* We are doing an adjtime thing.
770 * Prepare time_adjust_step to be within bounds.
771 * Note that a positive time_adjust means we want the clock
774 * Limit the amount of the step to be in the range
775 * -tickadj .. +tickadj
777 if (time_adjust
> tickadj
)
778 time_adjust_step
= tickadj
;
779 else if (time_adjust
< -tickadj
)
780 time_adjust_step
= -tickadj
;
782 /* Reduce by this step the amount of time left */
783 time_adjust
-= time_adjust_step
;
785 delta_nsec
= tick_nsec
+ time_adjust_step
* 1000;
787 * Advance the phase, once it gets to one microsecond, then
788 * advance the tick more.
790 time_phase
+= time_adj
;
791 if (time_phase
<= -FINENSEC
) {
792 long ltemp
= -time_phase
>> (SHIFT_SCALE
- 10);
793 time_phase
+= ltemp
<< (SHIFT_SCALE
- 10);
796 else if (time_phase
>= FINENSEC
) {
797 long ltemp
= time_phase
>> (SHIFT_SCALE
- 10);
798 time_phase
-= ltemp
<< (SHIFT_SCALE
- 10);
801 xtime
.tv_nsec
+= delta_nsec
;
802 time_interpolator_update(delta_nsec
);
804 /* Changes by adjtime() do not take effect till next tick. */
805 if (time_next_adjust
!= 0) {
806 time_adjust
= time_next_adjust
;
807 time_next_adjust
= 0;
812 * Using a loop looks inefficient, but "ticks" is
813 * usually just one (we shouldn't be losing ticks,
814 * we're doing this this way mainly for interrupt
815 * latency reasons, not because we think we'll
816 * have lots of lost timer ticks
818 static void update_wall_time(unsigned long ticks
)
822 update_wall_time_one_tick();
823 if (xtime
.tv_nsec
>= 1000000000) {
824 xtime
.tv_nsec
-= 1000000000;
832 * Called from the timer interrupt handler to charge one tick to the current
833 * process. user_tick is 1 if the tick is user time, 0 for system.
835 void update_process_times(int user_tick
)
837 struct task_struct
*p
= current
;
838 int cpu
= smp_processor_id();
840 /* Note: this timer irq context must be accounted for as well. */
842 account_user_time(p
, jiffies_to_cputime(1));
844 account_system_time(p
, HARDIRQ_OFFSET
, jiffies_to_cputime(1));
846 if (rcu_pending(cpu
))
847 rcu_check_callbacks(cpu
, user_tick
);
849 run_posix_cpu_timers(p
);
853 * Nr of active tasks - counted in fixed-point numbers
855 static unsigned long count_active_tasks(void)
857 return (nr_running() + nr_uninterruptible()) * FIXED_1
;
861 * Hmm.. Changed this, as the GNU make sources (load.c) seems to
862 * imply that avenrun[] is the standard name for this kind of thing.
863 * Nothing else seems to be standardized: the fractional size etc
864 * all seem to differ on different machines.
866 * Requires xtime_lock to access.
868 unsigned long avenrun
[3];
870 EXPORT_SYMBOL(avenrun
);
873 * calc_load - given tick count, update the avenrun load estimates.
874 * This is called while holding a write_lock on xtime_lock.
876 static inline void calc_load(unsigned long ticks
)
878 unsigned long active_tasks
; /* fixed-point */
879 static int count
= LOAD_FREQ
;
884 active_tasks
= count_active_tasks();
885 CALC_LOAD(avenrun
[0], EXP_1
, active_tasks
);
886 CALC_LOAD(avenrun
[1], EXP_5
, active_tasks
);
887 CALC_LOAD(avenrun
[2], EXP_15
, active_tasks
);
891 /* jiffies at the most recent update of wall time */
892 unsigned long wall_jiffies
= INITIAL_JIFFIES
;
895 * This read-write spinlock protects us from races in SMP while
896 * playing with xtime and avenrun.
898 #ifndef ARCH_HAVE_XTIME_LOCK
899 seqlock_t xtime_lock __cacheline_aligned_in_smp
= SEQLOCK_UNLOCKED
;
901 EXPORT_SYMBOL(xtime_lock
);
905 * This function runs timers and the timer-tq in bottom half context.
907 static void run_timer_softirq(struct softirq_action
*h
)
909 tvec_base_t
*base
= &__get_cpu_var(tvec_bases
);
911 if (time_after_eq(jiffies
, base
->timer_jiffies
))
916 * Called by the local, per-CPU timer interrupt on SMP.
918 void run_local_timers(void)
920 raise_softirq(TIMER_SOFTIRQ
);
924 * Called by the timer interrupt. xtime_lock must already be taken
927 static inline void update_times(void)
931 ticks
= jiffies
- wall_jiffies
;
933 wall_jiffies
+= ticks
;
934 update_wall_time(ticks
);
940 * The 64-bit jiffies value is not atomic - you MUST NOT read it
941 * without sampling the sequence number in xtime_lock.
942 * jiffies is defined in the linker script...
945 void do_timer(struct pt_regs
*regs
)
951 #ifdef __ARCH_WANT_SYS_ALARM
954 * For backwards compatibility? This can be done in libc so Alpha
955 * and all newer ports shouldn't need it.
957 asmlinkage
unsigned long sys_alarm(unsigned int seconds
)
959 struct itimerval it_new
, it_old
;
960 unsigned int oldalarm
;
962 it_new
.it_interval
.tv_sec
= it_new
.it_interval
.tv_usec
= 0;
963 it_new
.it_value
.tv_sec
= seconds
;
964 it_new
.it_value
.tv_usec
= 0;
965 do_setitimer(ITIMER_REAL
, &it_new
, &it_old
);
966 oldalarm
= it_old
.it_value
.tv_sec
;
967 /* ehhh.. We can't return 0 if we have an alarm pending.. */
968 /* And we'd better return too much than too little anyway */
969 if ((!oldalarm
&& it_old
.it_value
.tv_usec
) || it_old
.it_value
.tv_usec
>= 500000)
979 * The Alpha uses getxpid, getxuid, and getxgid instead. Maybe this
980 * should be moved into arch/i386 instead?
984 * sys_getpid - return the thread group id of the current process
986 * Note, despite the name, this returns the tgid not the pid. The tgid and
987 * the pid are identical unless CLONE_THREAD was specified on clone() in
988 * which case the tgid is the same in all threads of the same group.
990 * This is SMP safe as current->tgid does not change.
992 asmlinkage
long sys_getpid(void)
994 return current
->tgid
;
998 * Accessing ->group_leader->real_parent is not SMP-safe, it could
999 * change from under us. However, rather than getting any lock
1000 * we can use an optimistic algorithm: get the parent
1001 * pid, and go back and check that the parent is still
1002 * the same. If it has changed (which is extremely unlikely
1003 * indeed), we just try again..
1005 * NOTE! This depends on the fact that even if we _do_
1006 * get an old value of "parent", we can happily dereference
1007 * the pointer (it was and remains a dereferencable kernel pointer
1008 * no matter what): we just can't necessarily trust the result
1009 * until we know that the parent pointer is valid.
1011 * NOTE2: ->group_leader never changes from under us.
1013 asmlinkage
long sys_getppid(void)
1016 struct task_struct
*me
= current
;
1017 struct task_struct
*parent
;
1019 parent
= me
->group_leader
->real_parent
;
1024 struct task_struct
*old
= parent
;
1027 * Make sure we read the pid before re-reading the
1031 parent
= me
->group_leader
->real_parent
;
1041 asmlinkage
long sys_getuid(void)
1043 /* Only we change this so SMP safe */
1044 return current
->uid
;
1047 asmlinkage
long sys_geteuid(void)
1049 /* Only we change this so SMP safe */
1050 return current
->euid
;
1053 asmlinkage
long sys_getgid(void)
1055 /* Only we change this so SMP safe */
1056 return current
->gid
;
1059 asmlinkage
long sys_getegid(void)
1061 /* Only we change this so SMP safe */
1062 return current
->egid
;
1067 static void process_timeout(unsigned long __data
)
1069 wake_up_process((task_t
*)__data
);
1073 * schedule_timeout - sleep until timeout
1074 * @timeout: timeout value in jiffies
1076 * Make the current task sleep until @timeout jiffies have
1077 * elapsed. The routine will return immediately unless
1078 * the current task state has been set (see set_current_state()).
1080 * You can set the task state as follows -
1082 * %TASK_UNINTERRUPTIBLE - at least @timeout jiffies are guaranteed to
1083 * pass before the routine returns. The routine will return 0
1085 * %TASK_INTERRUPTIBLE - the routine may return early if a signal is
1086 * delivered to the current task. In this case the remaining time
1087 * in jiffies will be returned, or 0 if the timer expired in time
1089 * The current task state is guaranteed to be TASK_RUNNING when this
1092 * Specifying a @timeout value of %MAX_SCHEDULE_TIMEOUT will schedule
1093 * the CPU away without a bound on the timeout. In this case the return
1094 * value will be %MAX_SCHEDULE_TIMEOUT.
1096 * In all cases the return value is guaranteed to be non-negative.
1098 fastcall
signed long __sched
schedule_timeout(signed long timeout
)
1100 struct timer_list timer
;
1101 unsigned long expire
;
1105 case MAX_SCHEDULE_TIMEOUT
:
1107 * These two special cases are useful to be comfortable
1108 * in the caller. Nothing more. We could take
1109 * MAX_SCHEDULE_TIMEOUT from one of the negative value
1110 * but I' d like to return a valid offset (>=0) to allow
1111 * the caller to do everything it want with the retval.
1117 * Another bit of PARANOID. Note that the retval will be
1118 * 0 since no piece of kernel is supposed to do a check
1119 * for a negative retval of schedule_timeout() (since it
1120 * should never happens anyway). You just have the printk()
1121 * that will tell you if something is gone wrong and where.
1125 printk(KERN_ERR
"schedule_timeout: wrong timeout "
1126 "value %lx from %p\n", timeout
,
1127 __builtin_return_address(0));
1128 current
->state
= TASK_RUNNING
;
1133 expire
= timeout
+ jiffies
;
1136 timer
.expires
= expire
;
1137 timer
.data
= (unsigned long) current
;
1138 timer
.function
= process_timeout
;
1142 del_singleshot_timer_sync(&timer
);
1144 timeout
= expire
- jiffies
;
1147 return timeout
< 0 ? 0 : timeout
;
1150 EXPORT_SYMBOL(schedule_timeout
);
1152 /* Thread ID - the internal kernel "pid" */
1153 asmlinkage
long sys_gettid(void)
1155 return current
->pid
;
1158 static long __sched
nanosleep_restart(struct restart_block
*restart
)
1160 unsigned long expire
= restart
->arg0
, now
= jiffies
;
1161 struct timespec __user
*rmtp
= (struct timespec __user
*) restart
->arg1
;
1164 /* Did it expire while we handled signals? */
1165 if (!time_after(expire
, now
))
1168 current
->state
= TASK_INTERRUPTIBLE
;
1169 expire
= schedule_timeout(expire
- now
);
1174 jiffies_to_timespec(expire
, &t
);
1176 ret
= -ERESTART_RESTARTBLOCK
;
1177 if (rmtp
&& copy_to_user(rmtp
, &t
, sizeof(t
)))
1179 /* The 'restart' block is already filled in */
1184 asmlinkage
long sys_nanosleep(struct timespec __user
*rqtp
, struct timespec __user
*rmtp
)
1187 unsigned long expire
;
1190 if (copy_from_user(&t
, rqtp
, sizeof(t
)))
1193 if ((t
.tv_nsec
>= 1000000000L) || (t
.tv_nsec
< 0) || (t
.tv_sec
< 0))
1196 expire
= timespec_to_jiffies(&t
) + (t
.tv_sec
|| t
.tv_nsec
);
1197 current
->state
= TASK_INTERRUPTIBLE
;
1198 expire
= schedule_timeout(expire
);
1202 struct restart_block
*restart
;
1203 jiffies_to_timespec(expire
, &t
);
1204 if (rmtp
&& copy_to_user(rmtp
, &t
, sizeof(t
)))
1207 restart
= ¤t_thread_info()->restart_block
;
1208 restart
->fn
= nanosleep_restart
;
1209 restart
->arg0
= jiffies
+ expire
;
1210 restart
->arg1
= (unsigned long) rmtp
;
1211 ret
= -ERESTART_RESTARTBLOCK
;
1217 * sys_sysinfo - fill in sysinfo struct
1219 asmlinkage
long sys_sysinfo(struct sysinfo __user
*info
)
1222 unsigned long mem_total
, sav_total
;
1223 unsigned int mem_unit
, bitcount
;
1226 memset((char *)&val
, 0, sizeof(struct sysinfo
));
1230 seq
= read_seqbegin(&xtime_lock
);
1233 * This is annoying. The below is the same thing
1234 * posix_get_clock_monotonic() does, but it wants to
1235 * take the lock which we want to cover the loads stuff
1239 getnstimeofday(&tp
);
1240 tp
.tv_sec
+= wall_to_monotonic
.tv_sec
;
1241 tp
.tv_nsec
+= wall_to_monotonic
.tv_nsec
;
1242 if (tp
.tv_nsec
- NSEC_PER_SEC
>= 0) {
1243 tp
.tv_nsec
= tp
.tv_nsec
- NSEC_PER_SEC
;
1246 val
.uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
1248 val
.loads
[0] = avenrun
[0] << (SI_LOAD_SHIFT
- FSHIFT
);
1249 val
.loads
[1] = avenrun
[1] << (SI_LOAD_SHIFT
- FSHIFT
);
1250 val
.loads
[2] = avenrun
[2] << (SI_LOAD_SHIFT
- FSHIFT
);
1252 val
.procs
= nr_threads
;
1253 } while (read_seqretry(&xtime_lock
, seq
));
1259 * If the sum of all the available memory (i.e. ram + swap)
1260 * is less than can be stored in a 32 bit unsigned long then
1261 * we can be binary compatible with 2.2.x kernels. If not,
1262 * well, in that case 2.2.x was broken anyways...
1264 * -Erik Andersen <andersee@debian.org>
1267 mem_total
= val
.totalram
+ val
.totalswap
;
1268 if (mem_total
< val
.totalram
|| mem_total
< val
.totalswap
)
1271 mem_unit
= val
.mem_unit
;
1272 while (mem_unit
> 1) {
1275 sav_total
= mem_total
;
1277 if (mem_total
< sav_total
)
1282 * If mem_total did not overflow, multiply all memory values by
1283 * val.mem_unit and set it to 1. This leaves things compatible
1284 * with 2.2.x, and also retains compatibility with earlier 2.4.x
1289 val
.totalram
<<= bitcount
;
1290 val
.freeram
<<= bitcount
;
1291 val
.sharedram
<<= bitcount
;
1292 val
.bufferram
<<= bitcount
;
1293 val
.totalswap
<<= bitcount
;
1294 val
.freeswap
<<= bitcount
;
1295 val
.totalhigh
<<= bitcount
;
1296 val
.freehigh
<<= bitcount
;
1299 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
1305 static void __devinit
init_timers_cpu(int cpu
)
1310 base
= &per_cpu(tvec_bases
, cpu
);
1311 spin_lock_init(&base
->t_base
.lock
);
1312 for (j
= 0; j
< TVN_SIZE
; j
++) {
1313 INIT_LIST_HEAD(base
->tv5
.vec
+ j
);
1314 INIT_LIST_HEAD(base
->tv4
.vec
+ j
);
1315 INIT_LIST_HEAD(base
->tv3
.vec
+ j
);
1316 INIT_LIST_HEAD(base
->tv2
.vec
+ j
);
1318 for (j
= 0; j
< TVR_SIZE
; j
++)
1319 INIT_LIST_HEAD(base
->tv1
.vec
+ j
);
1321 base
->timer_jiffies
= jiffies
;
1324 #ifdef CONFIG_HOTPLUG_CPU
1325 static void migrate_timer_list(tvec_base_t
*new_base
, struct list_head
*head
)
1327 struct timer_list
*timer
;
1329 while (!list_empty(head
)) {
1330 timer
= list_entry(head
->next
, struct timer_list
, entry
);
1331 detach_timer(timer
, 0);
1332 timer
->base
= &new_base
->t_base
;
1333 internal_add_timer(new_base
, timer
);
1337 static void __devinit
migrate_timers(int cpu
)
1339 tvec_base_t
*old_base
;
1340 tvec_base_t
*new_base
;
1343 BUG_ON(cpu_online(cpu
));
1344 old_base
= &per_cpu(tvec_bases
, cpu
);
1345 new_base
= &get_cpu_var(tvec_bases
);
1347 local_irq_disable();
1348 spin_lock(&new_base
->t_base
.lock
);
1349 spin_lock(&old_base
->t_base
.lock
);
1351 if (old_base
->t_base
.running_timer
)
1353 for (i
= 0; i
< TVR_SIZE
; i
++)
1354 migrate_timer_list(new_base
, old_base
->tv1
.vec
+ i
);
1355 for (i
= 0; i
< TVN_SIZE
; i
++) {
1356 migrate_timer_list(new_base
, old_base
->tv2
.vec
+ i
);
1357 migrate_timer_list(new_base
, old_base
->tv3
.vec
+ i
);
1358 migrate_timer_list(new_base
, old_base
->tv4
.vec
+ i
);
1359 migrate_timer_list(new_base
, old_base
->tv5
.vec
+ i
);
1362 spin_unlock(&old_base
->t_base
.lock
);
1363 spin_unlock(&new_base
->t_base
.lock
);
1365 put_cpu_var(tvec_bases
);
1367 #endif /* CONFIG_HOTPLUG_CPU */
1369 static int __devinit
timer_cpu_notify(struct notifier_block
*self
,
1370 unsigned long action
, void *hcpu
)
1372 long cpu
= (long)hcpu
;
1374 case CPU_UP_PREPARE
:
1375 init_timers_cpu(cpu
);
1377 #ifdef CONFIG_HOTPLUG_CPU
1379 migrate_timers(cpu
);
1388 static struct notifier_block __devinitdata timers_nb
= {
1389 .notifier_call
= timer_cpu_notify
,
1393 void __init
init_timers(void)
1395 timer_cpu_notify(&timers_nb
, (unsigned long)CPU_UP_PREPARE
,
1396 (void *)(long)smp_processor_id());
1397 register_cpu_notifier(&timers_nb
);
1398 open_softirq(TIMER_SOFTIRQ
, run_timer_softirq
, NULL
);
1401 #ifdef CONFIG_TIME_INTERPOLATION
1403 struct time_interpolator
*time_interpolator
;
1404 static struct time_interpolator
*time_interpolator_list
;
1405 static DEFINE_SPINLOCK(time_interpolator_lock
);
1407 static inline u64
time_interpolator_get_cycles(unsigned int src
)
1409 unsigned long (*x
)(void);
1413 case TIME_SOURCE_FUNCTION
:
1414 x
= time_interpolator
->addr
;
1417 case TIME_SOURCE_MMIO64
:
1418 return readq((void __iomem
*) time_interpolator
->addr
);
1420 case TIME_SOURCE_MMIO32
:
1421 return readl((void __iomem
*) time_interpolator
->addr
);
1423 default: return get_cycles();
1427 static inline u64
time_interpolator_get_counter(void)
1429 unsigned int src
= time_interpolator
->source
;
1431 if (time_interpolator
->jitter
)
1437 lcycle
= time_interpolator
->last_cycle
;
1438 now
= time_interpolator_get_cycles(src
);
1439 if (lcycle
&& time_after(lcycle
, now
))
1441 /* Keep track of the last timer value returned. The use of cmpxchg here
1442 * will cause contention in an SMP environment.
1444 } while (unlikely(cmpxchg(&time_interpolator
->last_cycle
, lcycle
, now
) != lcycle
));
1448 return time_interpolator_get_cycles(src
);
1451 void time_interpolator_reset(void)
1453 time_interpolator
->offset
= 0;
1454 time_interpolator
->last_counter
= time_interpolator_get_counter();
1457 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
1459 unsigned long time_interpolator_get_offset(void)
1461 /* If we do not have a time interpolator set up then just return zero */
1462 if (!time_interpolator
)
1465 return time_interpolator
->offset
+
1466 GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator
);
1469 #define INTERPOLATOR_ADJUST 65536
1470 #define INTERPOLATOR_MAX_SKIP 10*INTERPOLATOR_ADJUST
1472 static void time_interpolator_update(long delta_nsec
)
1475 unsigned long offset
;
1477 /* If there is no time interpolator set up then do nothing */
1478 if (!time_interpolator
)
1481 /* The interpolator compensates for late ticks by accumulating
1482 * the late time in time_interpolator->offset. A tick earlier than
1483 * expected will lead to a reset of the offset and a corresponding
1484 * jump of the clock forward. Again this only works if the
1485 * interpolator clock is running slightly slower than the regular clock
1486 * and the tuning logic insures that.
1489 counter
= time_interpolator_get_counter();
1490 offset
= time_interpolator
->offset
+ GET_TI_NSECS(counter
, time_interpolator
);
1492 if (delta_nsec
< 0 || (unsigned long) delta_nsec
< offset
)
1493 time_interpolator
->offset
= offset
- delta_nsec
;
1495 time_interpolator
->skips
++;
1496 time_interpolator
->ns_skipped
+= delta_nsec
- offset
;
1497 time_interpolator
->offset
= 0;
1499 time_interpolator
->last_counter
= counter
;
1501 /* Tuning logic for time interpolator invoked every minute or so.
1502 * Decrease interpolator clock speed if no skips occurred and an offset is carried.
1503 * Increase interpolator clock speed if we skip too much time.
1505 if (jiffies
% INTERPOLATOR_ADJUST
== 0)
1507 if (time_interpolator
->skips
== 0 && time_interpolator
->offset
> TICK_NSEC
)
1508 time_interpolator
->nsec_per_cyc
--;
1509 if (time_interpolator
->ns_skipped
> INTERPOLATOR_MAX_SKIP
&& time_interpolator
->offset
== 0)
1510 time_interpolator
->nsec_per_cyc
++;
1511 time_interpolator
->skips
= 0;
1512 time_interpolator
->ns_skipped
= 0;
1517 is_better_time_interpolator(struct time_interpolator
*new)
1519 if (!time_interpolator
)
1521 return new->frequency
> 2*time_interpolator
->frequency
||
1522 (unsigned long)new->drift
< (unsigned long)time_interpolator
->drift
;
1526 register_time_interpolator(struct time_interpolator
*ti
)
1528 unsigned long flags
;
1531 if (ti
->frequency
== 0 || ti
->mask
== 0)
1534 ti
->nsec_per_cyc
= ((u64
)NSEC_PER_SEC
<< ti
->shift
) / ti
->frequency
;
1535 spin_lock(&time_interpolator_lock
);
1536 write_seqlock_irqsave(&xtime_lock
, flags
);
1537 if (is_better_time_interpolator(ti
)) {
1538 time_interpolator
= ti
;
1539 time_interpolator_reset();
1541 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1543 ti
->next
= time_interpolator_list
;
1544 time_interpolator_list
= ti
;
1545 spin_unlock(&time_interpolator_lock
);
1549 unregister_time_interpolator(struct time_interpolator
*ti
)
1551 struct time_interpolator
*curr
, **prev
;
1552 unsigned long flags
;
1554 spin_lock(&time_interpolator_lock
);
1555 prev
= &time_interpolator_list
;
1556 for (curr
= *prev
; curr
; curr
= curr
->next
) {
1564 write_seqlock_irqsave(&xtime_lock
, flags
);
1565 if (ti
== time_interpolator
) {
1566 /* we lost the best time-interpolator: */
1567 time_interpolator
= NULL
;
1568 /* find the next-best interpolator */
1569 for (curr
= time_interpolator_list
; curr
; curr
= curr
->next
)
1570 if (is_better_time_interpolator(curr
))
1571 time_interpolator
= curr
;
1572 time_interpolator_reset();
1574 write_sequnlock_irqrestore(&xtime_lock
, flags
);
1575 spin_unlock(&time_interpolator_lock
);
1577 #endif /* CONFIG_TIME_INTERPOLATION */
1580 * msleep - sleep safely even with waitqueue interruptions
1581 * @msecs: Time in milliseconds to sleep for
1583 void msleep(unsigned int msecs
)
1585 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1588 set_current_state(TASK_UNINTERRUPTIBLE
);
1589 timeout
= schedule_timeout(timeout
);
1593 EXPORT_SYMBOL(msleep
);
1596 * msleep_interruptible - sleep waiting for waitqueue interruptions
1597 * @msecs: Time in milliseconds to sleep for
1599 unsigned long msleep_interruptible(unsigned int msecs
)
1601 unsigned long timeout
= msecs_to_jiffies(msecs
) + 1;
1603 while (timeout
&& !signal_pending(current
)) {
1604 set_current_state(TASK_INTERRUPTIBLE
);
1605 timeout
= schedule_timeout(timeout
);
1607 return jiffies_to_msecs(timeout
);
1610 EXPORT_SYMBOL(msleep_interruptible
);