2 * linux/kernel/time/timekeeping.c
4 * Kernel timekeeping code and accessor functions
6 * This code was moved from linux/kernel/timer.c.
7 * Please see that file for copyright and history logs.
11 #include <linux/module.h>
12 #include <linux/interrupt.h>
13 #include <linux/percpu.h>
14 #include <linux/init.h>
16 #include <linux/sysdev.h>
17 #include <linux/clocksource.h>
18 #include <linux/jiffies.h>
19 #include <linux/time.h>
20 #include <linux/tick.h>
24 * This read-write spinlock protects us from races in SMP while
25 * playing with xtime and avenrun.
27 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
32 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
33 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
34 * at zero at system boot time, so wall_to_monotonic will be negative,
35 * however, we will ALWAYS keep the tv_nsec part positive so we can use
36 * the usual normalization.
38 * wall_to_monotonic is moved after resume from suspend for the monotonic
39 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
40 * to get the real boot based time offset.
42 * - wall_to_monotonic is no longer the boot time, getboottime must be
45 struct timespec xtime
__attribute__ ((aligned (16)));
46 struct timespec wall_to_monotonic
__attribute__ ((aligned (16)));
47 static unsigned long total_sleep_time
; /* seconds */
49 static struct timespec xtime_cache
__attribute__ ((aligned (16)));
50 void update_xtime_cache(u64 nsec
)
53 timespec_add_ns(&xtime_cache
, nsec
);
56 struct clocksource
*clock
;
59 #ifdef CONFIG_GENERIC_TIME
61 * clocksource_forward_now - update clock to the current time
63 * Forward the current clock to update its state since the last call to
64 * update_wall_time(). This is useful before significant clock changes,
65 * as it avoids having to deal with this time offset explicitly.
67 static void clocksource_forward_now(void)
69 cycle_t cycle_now
, cycle_delta
;
72 cycle_now
= clocksource_read(clock
);
73 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
74 clock
->cycle_last
= cycle_now
;
76 nsec
= cyc2ns(clock
, cycle_delta
);
77 timespec_add_ns(&xtime
, nsec
);
79 nsec
= ((s64
)cycle_delta
* clock
->mult_orig
) >> clock
->shift
;
80 clock
->raw_time
.tv_nsec
+= nsec
;
84 * getnstimeofday - Returns the time of day in a timespec
85 * @ts: pointer to the timespec to be set
87 * Returns the time of day in a timespec.
89 void getnstimeofday(struct timespec
*ts
)
91 cycle_t cycle_now
, cycle_delta
;
96 seq
= read_seqbegin(&xtime_lock
);
100 /* read clocksource: */
101 cycle_now
= clocksource_read(clock
);
103 /* calculate the delta since the last update_wall_time: */
104 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
106 /* convert to nanoseconds: */
107 nsecs
= cyc2ns(clock
, cycle_delta
);
109 } while (read_seqretry(&xtime_lock
, seq
));
111 timespec_add_ns(ts
, nsecs
);
114 EXPORT_SYMBOL(getnstimeofday
);
117 * do_gettimeofday - Returns the time of day in a timeval
118 * @tv: pointer to the timeval to be set
120 * NOTE: Users should be converted to using getnstimeofday()
122 void do_gettimeofday(struct timeval
*tv
)
126 getnstimeofday(&now
);
127 tv
->tv_sec
= now
.tv_sec
;
128 tv
->tv_usec
= now
.tv_nsec
/1000;
131 EXPORT_SYMBOL(do_gettimeofday
);
133 * do_settimeofday - Sets the time of day
134 * @tv: pointer to the timespec variable containing the new time
136 * Sets the time of day to the new time and update NTP and notify hrtimers
138 int do_settimeofday(struct timespec
*tv
)
140 struct timespec ts_delta
;
143 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
146 write_seqlock_irqsave(&xtime_lock
, flags
);
148 clocksource_forward_now();
150 ts_delta
.tv_sec
= tv
->tv_sec
- xtime
.tv_sec
;
151 ts_delta
.tv_nsec
= tv
->tv_nsec
- xtime
.tv_nsec
;
152 wall_to_monotonic
= timespec_sub(wall_to_monotonic
, ts_delta
);
156 update_xtime_cache(0);
161 update_vsyscall(&xtime
, clock
);
163 write_sequnlock_irqrestore(&xtime_lock
, flags
);
165 /* signal hrtimers about time change */
171 EXPORT_SYMBOL(do_settimeofday
);
174 * change_clocksource - Swaps clocksources if a new one is available
176 * Accumulates current time interval and initializes new clocksource
178 static void change_clocksource(void)
180 struct clocksource
*new;
182 new = clocksource_get_next();
187 clocksource_forward_now();
189 new->raw_time
= clock
->raw_time
;
192 clock
->cycle_last
= 0;
193 clock
->cycle_last
= clocksource_read(new);
195 clock
->xtime_nsec
= 0;
196 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
201 * We're holding xtime lock and waking up klogd would deadlock
202 * us on enqueue. So no printing!
203 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
208 static inline void clocksource_forward_now(void) { }
209 static inline void change_clocksource(void) { }
213 * getrawmonotonic - Returns the raw monotonic time in a timespec
214 * @ts: pointer to the timespec to be set
216 * Returns the raw monotonic time (completely un-modified by ntp)
218 void getrawmonotonic(struct timespec
*ts
)
222 cycle_t cycle_now
, cycle_delta
;
225 seq
= read_seqbegin(&xtime_lock
);
227 /* read clocksource: */
228 cycle_now
= clocksource_read(clock
);
230 /* calculate the delta since the last update_wall_time: */
231 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
233 /* convert to nanoseconds: */
234 nsecs
= ((s64
)cycle_delta
* clock
->mult_orig
) >> clock
->shift
;
236 *ts
= clock
->raw_time
;
238 } while (read_seqretry(&xtime_lock
, seq
));
240 timespec_add_ns(ts
, nsecs
);
242 EXPORT_SYMBOL(getrawmonotonic
);
246 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
248 int timekeeping_valid_for_hres(void)
254 seq
= read_seqbegin(&xtime_lock
);
256 ret
= clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
258 } while (read_seqretry(&xtime_lock
, seq
));
264 * read_persistent_clock - Return time in seconds from the persistent clock.
266 * Weak dummy function for arches that do not yet support it.
267 * Returns seconds from epoch using the battery backed persistent clock.
268 * Returns zero if unsupported.
270 * XXX - Do be sure to remove it once all arches implement it.
272 unsigned long __attribute__((weak
)) read_persistent_clock(void)
278 * timekeeping_init - Initializes the clocksource and common timekeeping values
280 void __init
timekeeping_init(void)
283 unsigned long sec
= read_persistent_clock();
285 write_seqlock_irqsave(&xtime_lock
, flags
);
289 clock
= clocksource_get_next();
290 clocksource_calculate_interval(clock
, NTP_INTERVAL_LENGTH
);
291 clock
->cycle_last
= clocksource_read(clock
);
295 set_normalized_timespec(&wall_to_monotonic
,
296 -xtime
.tv_sec
, -xtime
.tv_nsec
);
297 update_xtime_cache(0);
298 total_sleep_time
= 0;
299 write_sequnlock_irqrestore(&xtime_lock
, flags
);
302 /* flag for if timekeeping is suspended */
303 static int timekeeping_suspended
;
304 /* time in seconds when suspend began */
305 static unsigned long timekeeping_suspend_time
;
308 * timekeeping_resume - Resumes the generic timekeeping subsystem.
311 * This is for the generic clocksource timekeeping.
312 * xtime/wall_to_monotonic/jiffies/etc are
313 * still managed by arch specific suspend/resume code.
315 static int timekeeping_resume(struct sys_device
*dev
)
318 unsigned long now
= read_persistent_clock();
320 clocksource_resume();
322 write_seqlock_irqsave(&xtime_lock
, flags
);
324 if (now
&& (now
> timekeeping_suspend_time
)) {
325 unsigned long sleep_length
= now
- timekeeping_suspend_time
;
327 xtime
.tv_sec
+= sleep_length
;
328 wall_to_monotonic
.tv_sec
-= sleep_length
;
329 total_sleep_time
+= sleep_length
;
331 update_xtime_cache(0);
332 /* re-base the last cycle value */
333 clock
->cycle_last
= 0;
334 clock
->cycle_last
= clocksource_read(clock
);
336 timekeeping_suspended
= 0;
337 write_sequnlock_irqrestore(&xtime_lock
, flags
);
339 touch_softlockup_watchdog();
341 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
343 /* Resume hrtimers */
344 hres_timers_resume();
349 static int timekeeping_suspend(struct sys_device
*dev
, pm_message_t state
)
353 timekeeping_suspend_time
= read_persistent_clock();
355 write_seqlock_irqsave(&xtime_lock
, flags
);
356 clocksource_forward_now();
357 timekeeping_suspended
= 1;
358 write_sequnlock_irqrestore(&xtime_lock
, flags
);
360 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
365 /* sysfs resume/suspend bits for timekeeping */
366 static struct sysdev_class timekeeping_sysclass
= {
367 .name
= "timekeeping",
368 .resume
= timekeeping_resume
,
369 .suspend
= timekeeping_suspend
,
372 static struct sys_device device_timer
= {
374 .cls
= &timekeeping_sysclass
,
377 static int __init
timekeeping_init_device(void)
379 int error
= sysdev_class_register(&timekeeping_sysclass
);
381 error
= sysdev_register(&device_timer
);
385 device_initcall(timekeeping_init_device
);
388 * If the error is already larger, we look ahead even further
389 * to compensate for late or lost adjustments.
391 static __always_inline
int clocksource_bigadjust(s64 error
, s64
*interval
,
399 * Use the current error value to determine how much to look ahead.
400 * The larger the error the slower we adjust for it to avoid problems
401 * with losing too many ticks, otherwise we would overadjust and
402 * produce an even larger error. The smaller the adjustment the
403 * faster we try to adjust for it, as lost ticks can do less harm
404 * here. This is tuned so that an error of about 1 msec is adjusted
405 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
407 error2
= clock
->error
>> (NTP_SCALE_SHIFT
+ 22 - 2 * SHIFT_HZ
);
408 error2
= abs(error2
);
409 for (look_ahead
= 0; error2
> 0; look_ahead
++)
413 * Now calculate the error in (1 << look_ahead) ticks, but first
414 * remove the single look ahead already included in the error.
416 tick_error
= tick_length
>> (NTP_SCALE_SHIFT
- clock
->shift
+ 1);
417 tick_error
-= clock
->xtime_interval
>> 1;
418 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
420 /* Finally calculate the adjustment shift value. */
425 *interval
= -*interval
;
429 for (adj
= 0; error
> i
; adj
++)
438 * Adjust the multiplier to reduce the error value,
439 * this is optimized for the most common adjustments of -1,0,1,
440 * for other values we can do a bit more work.
442 static void clocksource_adjust(s64 offset
)
444 s64 error
, interval
= clock
->cycle_interval
;
447 error
= clock
->error
>> (NTP_SCALE_SHIFT
- clock
->shift
- 1);
448 if (error
> interval
) {
450 if (likely(error
<= interval
))
453 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
454 } else if (error
< -interval
) {
456 if (likely(error
>= -interval
)) {
458 interval
= -interval
;
461 adj
= clocksource_bigadjust(error
, &interval
, &offset
);
466 clock
->xtime_interval
+= interval
;
467 clock
->xtime_nsec
-= offset
;
468 clock
->error
-= (interval
- offset
) <<
469 (NTP_SCALE_SHIFT
- clock
->shift
);
473 * update_wall_time - Uses the current clocksource to increment the wall time
475 * Called from the timer interrupt, must hold a write on xtime_lock.
477 void update_wall_time(void)
481 /* Make sure we're fully resumed: */
482 if (unlikely(timekeeping_suspended
))
485 #ifdef CONFIG_GENERIC_TIME
486 offset
= (clocksource_read(clock
) - clock
->cycle_last
) & clock
->mask
;
488 offset
= clock
->cycle_interval
;
490 clock
->xtime_nsec
= (s64
)xtime
.tv_nsec
<< clock
->shift
;
492 /* normally this loop will run just once, however in the
493 * case of lost or late ticks, it will accumulate correctly.
495 while (offset
>= clock
->cycle_interval
) {
496 /* accumulate one interval */
497 offset
-= clock
->cycle_interval
;
498 clock
->cycle_last
+= clock
->cycle_interval
;
500 clock
->xtime_nsec
+= clock
->xtime_interval
;
501 if (clock
->xtime_nsec
>= (u64
)NSEC_PER_SEC
<< clock
->shift
) {
502 clock
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< clock
->shift
;
507 clock
->raw_time
.tv_nsec
+= clock
->raw_interval
;
508 if (clock
->raw_time
.tv_nsec
>= NSEC_PER_SEC
) {
509 clock
->raw_time
.tv_nsec
-= NSEC_PER_SEC
;
510 clock
->raw_time
.tv_sec
++;
513 /* accumulate error between NTP and clock interval */
514 clock
->error
+= tick_length
;
515 clock
->error
-= clock
->xtime_interval
<< (NTP_SCALE_SHIFT
- clock
->shift
);
518 /* correct the clock when NTP error is too big */
519 clocksource_adjust(offset
);
522 * Since in the loop above, we accumulate any amount of time
523 * in xtime_nsec over a second into xtime.tv_sec, its possible for
524 * xtime_nsec to be fairly small after the loop. Further, if we're
525 * slightly speeding the clocksource up in clocksource_adjust(),
526 * its possible the required corrective factor to xtime_nsec could
527 * cause it to underflow.
529 * Now, we cannot simply roll the accumulated second back, since
530 * the NTP subsystem has been notified via second_overflow. So
531 * instead we push xtime_nsec forward by the amount we underflowed,
532 * and add that amount into the error.
534 * We'll correct this error next time through this function, when
535 * xtime_nsec is not as small.
537 if (unlikely((s64
)clock
->xtime_nsec
< 0)) {
538 s64 neg
= -(s64
)clock
->xtime_nsec
;
539 clock
->xtime_nsec
= 0;
540 clock
->error
+= neg
<< (NTP_SCALE_SHIFT
- clock
->shift
);
543 /* store full nanoseconds into xtime after rounding it up and
544 * add the remainder to the error difference.
546 xtime
.tv_nsec
= ((s64
)clock
->xtime_nsec
>> clock
->shift
) + 1;
547 clock
->xtime_nsec
-= (s64
)xtime
.tv_nsec
<< clock
->shift
;
548 clock
->error
+= clock
->xtime_nsec
<< (NTP_SCALE_SHIFT
- clock
->shift
);
550 update_xtime_cache(cyc2ns(clock
, offset
));
552 /* check to see if there is a new clocksource to use */
553 change_clocksource();
554 update_vsyscall(&xtime
, clock
);
558 * getboottime - Return the real time of system boot.
559 * @ts: pointer to the timespec to be set
561 * Returns the time of day in a timespec.
563 * This is based on the wall_to_monotonic offset and the total suspend
564 * time. Calls to settimeofday will affect the value returned (which
565 * basically means that however wrong your real time clock is at boot time,
566 * you get the right time here).
568 void getboottime(struct timespec
*ts
)
570 set_normalized_timespec(ts
,
571 - (wall_to_monotonic
.tv_sec
+ total_sleep_time
),
572 - wall_to_monotonic
.tv_nsec
);
576 * monotonic_to_bootbased - Convert the monotonic time to boot based.
577 * @ts: pointer to the timespec to be converted
579 void monotonic_to_bootbased(struct timespec
*ts
)
581 ts
->tv_sec
+= total_sleep_time
;
584 unsigned long get_seconds(void)
586 return xtime_cache
.tv_sec
;
588 EXPORT_SYMBOL(get_seconds
);
591 struct timespec
current_kernel_time(void)
597 seq
= read_seqbegin(&xtime_lock
);
600 } while (read_seqretry(&xtime_lock
, seq
));
604 EXPORT_SYMBOL(current_kernel_time
);