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/sched.h>
17 #include <linux/syscore_ops.h>
18 #include <linux/clocksource.h>
19 #include <linux/jiffies.h>
20 #include <linux/time.h>
21 #include <linux/tick.h>
22 #include <linux/stop_machine.h>
24 /* Structure holding internal timekeeping values. */
26 /* Current clocksource used for timekeeping. */
27 struct clocksource
*clock
;
28 /* NTP adjusted clock multiplier */
30 /* The shift value of the current clocksource. */
32 /* Number of clock cycles in one NTP interval. */
33 cycle_t cycle_interval
;
34 /* Number of clock shifted nano seconds in one NTP interval. */
36 /* shifted nano seconds left over when rounding cycle_interval */
38 /* Raw nano seconds accumulated per NTP interval. */
41 /* Current CLOCK_REALTIME time in seconds */
43 /* Clock shifted nano seconds */
46 /* Difference between accumulated time and NTP time in ntp
47 * shifted nano seconds. */
49 /* Shift conversion between clock shifted nano seconds and
50 * ntp shifted nano seconds. */
54 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
55 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
56 * at zero at system boot time, so wall_to_monotonic will be negative,
57 * however, we will ALWAYS keep the tv_nsec part positive so we can use
58 * the usual normalization.
60 * wall_to_monotonic is moved after resume from suspend for the
61 * monotonic time not to jump. We need to add total_sleep_time to
62 * wall_to_monotonic to get the real boot based time offset.
64 * - wall_to_monotonic is no longer the boot time, getboottime must be
67 struct timespec wall_to_monotonic
;
68 /* Offset clock monotonic -> clock realtime */
70 /* time spent in suspend */
71 struct timespec total_sleep_time
;
72 /* Offset clock monotonic -> clock boottime */
74 /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */
75 struct timespec raw_time
;
76 /* Seqlock for all timekeeper values */
80 static struct timekeeper timekeeper
;
83 * This read-write spinlock protects us from races in SMP while
86 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
88 /* flag for if timekeeping is suspended */
89 int __read_mostly timekeeping_suspended
;
91 static inline void tk_normalize_xtime(struct timekeeper
*tk
)
93 while (tk
->xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->shift
)) {
94 tk
->xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->shift
;
99 static struct timespec
tk_xtime(struct timekeeper
*tk
)
103 ts
.tv_sec
= tk
->xtime_sec
;
104 ts
.tv_nsec
= (long)(tk
->xtime_nsec
>> tk
->shift
);
108 static void tk_set_xtime(struct timekeeper
*tk
, const struct timespec
*ts
)
110 tk
->xtime_sec
= ts
->tv_sec
;
111 tk
->xtime_nsec
= (u64
)ts
->tv_nsec
<< tk
->shift
;
114 static void tk_xtime_add(struct timekeeper
*tk
, const struct timespec
*ts
)
116 tk
->xtime_sec
+= ts
->tv_sec
;
117 tk
->xtime_nsec
+= (u64
)ts
->tv_nsec
<< tk
->shift
;
118 tk_normalize_xtime(tk
);
121 static void tk_set_wall_to_mono(struct timekeeper
*tk
, struct timespec wtm
)
126 * Verify consistency of: offset_real = -wall_to_monotonic
127 * before modifying anything
129 set_normalized_timespec(&tmp
, -tk
->wall_to_monotonic
.tv_sec
,
130 -tk
->wall_to_monotonic
.tv_nsec
);
131 WARN_ON_ONCE(tk
->offs_real
.tv64
!= timespec_to_ktime(tmp
).tv64
);
132 tk
->wall_to_monotonic
= wtm
;
133 set_normalized_timespec(&tmp
, -wtm
.tv_sec
, -wtm
.tv_nsec
);
134 tk
->offs_real
= timespec_to_ktime(tmp
);
137 static void tk_set_sleep_time(struct timekeeper
*tk
, struct timespec t
)
139 /* Verify consistency before modifying */
140 WARN_ON_ONCE(tk
->offs_boot
.tv64
!= timespec_to_ktime(tk
->total_sleep_time
).tv64
);
142 tk
->total_sleep_time
= t
;
143 tk
->offs_boot
= timespec_to_ktime(t
);
147 * timekeeper_setup_internals - Set up internals to use clocksource clock.
149 * @clock: Pointer to clocksource.
151 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
152 * pair and interval request.
154 * Unless you're the timekeeping code, you should not be using this!
156 static void tk_setup_internals(struct timekeeper
*tk
, struct clocksource
*clock
)
159 u64 tmp
, ntpinterval
;
160 struct clocksource
*old_clock
;
162 old_clock
= tk
->clock
;
164 clock
->cycle_last
= clock
->read(clock
);
166 /* Do the ns -> cycle conversion first, using original mult */
167 tmp
= NTP_INTERVAL_LENGTH
;
168 tmp
<<= clock
->shift
;
170 tmp
+= clock
->mult
/2;
171 do_div(tmp
, clock
->mult
);
175 interval
= (cycle_t
) tmp
;
176 tk
->cycle_interval
= interval
;
178 /* Go back from cycles -> shifted ns */
179 tk
->xtime_interval
= (u64
) interval
* clock
->mult
;
180 tk
->xtime_remainder
= ntpinterval
- tk
->xtime_interval
;
182 ((u64
) interval
* clock
->mult
) >> clock
->shift
;
184 /* if changing clocks, convert xtime_nsec shift units */
186 int shift_change
= clock
->shift
- old_clock
->shift
;
187 if (shift_change
< 0)
188 tk
->xtime_nsec
>>= -shift_change
;
190 tk
->xtime_nsec
<<= shift_change
;
192 tk
->shift
= clock
->shift
;
195 tk
->ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
198 * The timekeeper keeps its own mult values for the currently
199 * active clocksource. These value will be adjusted via NTP
200 * to counteract clock drifting.
202 tk
->mult
= clock
->mult
;
205 /* Timekeeper helper functions. */
206 static inline s64
timekeeping_get_ns(struct timekeeper
*tk
)
208 cycle_t cycle_now
, cycle_delta
;
209 struct clocksource
*clock
;
212 /* read clocksource: */
214 cycle_now
= clock
->read(clock
);
216 /* calculate the delta since the last update_wall_time: */
217 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
219 nsec
= cycle_delta
* tk
->mult
+ tk
->xtime_nsec
;
222 /* If arch requires, add in gettimeoffset() */
223 return nsec
+ arch_gettimeoffset();
226 static inline s64
timekeeping_get_ns_raw(struct timekeeper
*tk
)
228 cycle_t cycle_now
, cycle_delta
;
229 struct clocksource
*clock
;
232 /* read clocksource: */
234 cycle_now
= clock
->read(clock
);
236 /* calculate the delta since the last update_wall_time: */
237 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
239 /* convert delta to nanoseconds. */
240 nsec
= clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
242 /* If arch requires, add in gettimeoffset() */
243 return nsec
+ arch_gettimeoffset();
246 /* must hold write on timekeeper.lock */
247 static void timekeeping_update(struct timekeeper
*tk
, bool clearntp
)
256 update_vsyscall(&xt
, &tk
->wall_to_monotonic
, tk
->clock
, tk
->mult
);
260 * timekeeping_forward_now - update clock to the current time
262 * Forward the current clock to update its state since the last call to
263 * update_wall_time(). This is useful before significant clock changes,
264 * as it avoids having to deal with this time offset explicitly.
266 static void timekeeping_forward_now(struct timekeeper
*tk
)
268 cycle_t cycle_now
, cycle_delta
;
269 struct clocksource
*clock
;
273 cycle_now
= clock
->read(clock
);
274 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
275 clock
->cycle_last
= cycle_now
;
277 tk
->xtime_nsec
+= cycle_delta
* tk
->mult
;
279 /* If arch requires, add in gettimeoffset() */
280 tk
->xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->shift
;
282 tk_normalize_xtime(tk
);
284 nsec
= clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
285 timespec_add_ns(&tk
->raw_time
, nsec
);
289 * getnstimeofday - Returns the time of day in a timespec
290 * @ts: pointer to the timespec to be set
292 * Returns the time of day in a timespec.
294 void getnstimeofday(struct timespec
*ts
)
296 struct timekeeper
*tk
= &timekeeper
;
300 WARN_ON(timekeeping_suspended
);
303 seq
= read_seqbegin(&tk
->lock
);
305 ts
->tv_sec
= tk
->xtime_sec
;
306 ts
->tv_nsec
= timekeeping_get_ns(tk
);
308 } while (read_seqretry(&tk
->lock
, seq
));
310 timespec_add_ns(ts
, nsecs
);
312 EXPORT_SYMBOL(getnstimeofday
);
314 ktime_t
ktime_get(void)
316 struct timekeeper
*tk
= &timekeeper
;
320 WARN_ON(timekeeping_suspended
);
323 seq
= read_seqbegin(&tk
->lock
);
324 secs
= tk
->xtime_sec
+ tk
->wall_to_monotonic
.tv_sec
;
325 nsecs
= timekeeping_get_ns(tk
) + tk
->wall_to_monotonic
.tv_nsec
;
327 } while (read_seqretry(&tk
->lock
, seq
));
329 * Use ktime_set/ktime_add_ns to create a proper ktime on
330 * 32-bit architectures without CONFIG_KTIME_SCALAR.
332 return ktime_add_ns(ktime_set(secs
, 0), nsecs
);
334 EXPORT_SYMBOL_GPL(ktime_get
);
337 * ktime_get_ts - get the monotonic clock in timespec format
338 * @ts: pointer to timespec variable
340 * The function calculates the monotonic clock from the realtime
341 * clock and the wall_to_monotonic offset and stores the result
342 * in normalized timespec format in the variable pointed to by @ts.
344 void ktime_get_ts(struct timespec
*ts
)
346 struct timekeeper
*tk
= &timekeeper
;
347 struct timespec tomono
;
350 WARN_ON(timekeeping_suspended
);
353 seq
= read_seqbegin(&tk
->lock
);
354 ts
->tv_sec
= tk
->xtime_sec
;
355 ts
->tv_nsec
= timekeeping_get_ns(tk
);
356 tomono
= tk
->wall_to_monotonic
;
358 } while (read_seqretry(&tk
->lock
, seq
));
360 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
361 ts
->tv_nsec
+ tomono
.tv_nsec
);
363 EXPORT_SYMBOL_GPL(ktime_get_ts
);
365 #ifdef CONFIG_NTP_PPS
368 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
369 * @ts_raw: pointer to the timespec to be set to raw monotonic time
370 * @ts_real: pointer to the timespec to be set to the time of day
372 * This function reads both the time of day and raw monotonic time at the
373 * same time atomically and stores the resulting timestamps in timespec
376 void getnstime_raw_and_real(struct timespec
*ts_raw
, struct timespec
*ts_real
)
378 struct timekeeper
*tk
= &timekeeper
;
380 s64 nsecs_raw
, nsecs_real
;
382 WARN_ON_ONCE(timekeeping_suspended
);
385 seq
= read_seqbegin(&tk
->lock
);
387 *ts_raw
= tk
->raw_time
;
388 ts_real
->tv_sec
= tk
->xtime_sec
;
389 ts_real
->tv_nsec
= 0;
391 nsecs_raw
= timekeeping_get_ns_raw(tk
);
392 nsecs_real
= timekeeping_get_ns(tk
);
394 } while (read_seqretry(&tk
->lock
, seq
));
396 timespec_add_ns(ts_raw
, nsecs_raw
);
397 timespec_add_ns(ts_real
, nsecs_real
);
399 EXPORT_SYMBOL(getnstime_raw_and_real
);
401 #endif /* CONFIG_NTP_PPS */
404 * do_gettimeofday - Returns the time of day in a timeval
405 * @tv: pointer to the timeval to be set
407 * NOTE: Users should be converted to using getnstimeofday()
409 void do_gettimeofday(struct timeval
*tv
)
413 getnstimeofday(&now
);
414 tv
->tv_sec
= now
.tv_sec
;
415 tv
->tv_usec
= now
.tv_nsec
/1000;
417 EXPORT_SYMBOL(do_gettimeofday
);
420 * do_settimeofday - Sets the time of day
421 * @tv: pointer to the timespec variable containing the new time
423 * Sets the time of day to the new time and update NTP and notify hrtimers
425 int do_settimeofday(const struct timespec
*tv
)
427 struct timekeeper
*tk
= &timekeeper
;
428 struct timespec ts_delta
, xt
;
431 if (!timespec_valid_strict(tv
))
434 write_seqlock_irqsave(&tk
->lock
, flags
);
436 timekeeping_forward_now(tk
);
439 ts_delta
.tv_sec
= tv
->tv_sec
- xt
.tv_sec
;
440 ts_delta
.tv_nsec
= tv
->tv_nsec
- xt
.tv_nsec
;
442 tk_set_wall_to_mono(tk
, timespec_sub(tk
->wall_to_monotonic
, ts_delta
));
444 tk_set_xtime(tk
, tv
);
446 timekeeping_update(tk
, true);
448 write_sequnlock_irqrestore(&tk
->lock
, flags
);
450 /* signal hrtimers about time change */
455 EXPORT_SYMBOL(do_settimeofday
);
458 * timekeeping_inject_offset - Adds or subtracts from the current time.
459 * @tv: pointer to the timespec variable containing the offset
461 * Adds or subtracts an offset value from the current time.
463 int timekeeping_inject_offset(struct timespec
*ts
)
465 struct timekeeper
*tk
= &timekeeper
;
470 if ((unsigned long)ts
->tv_nsec
>= NSEC_PER_SEC
)
473 write_seqlock_irqsave(&tk
->lock
, flags
);
475 timekeeping_forward_now(tk
);
477 /* Make sure the proposed value is valid */
478 tmp
= timespec_add(tk_xtime(tk
), *ts
);
479 if (!timespec_valid_strict(&tmp
)) {
484 tk_xtime_add(tk
, ts
);
485 tk_set_wall_to_mono(tk
, timespec_sub(tk
->wall_to_monotonic
, *ts
));
487 error
: /* even if we error out, we forwarded the time, so call update */
488 timekeeping_update(tk
, true);
490 write_sequnlock_irqrestore(&tk
->lock
, flags
);
492 /* signal hrtimers about time change */
497 EXPORT_SYMBOL(timekeeping_inject_offset
);
500 * change_clocksource - Swaps clocksources if a new one is available
502 * Accumulates current time interval and initializes new clocksource
504 static int change_clocksource(void *data
)
506 struct timekeeper
*tk
= &timekeeper
;
507 struct clocksource
*new, *old
;
510 new = (struct clocksource
*) data
;
512 write_seqlock_irqsave(&tk
->lock
, flags
);
514 timekeeping_forward_now(tk
);
515 if (!new->enable
|| new->enable(new) == 0) {
517 tk_setup_internals(tk
, new);
521 timekeeping_update(tk
, true);
523 write_sequnlock_irqrestore(&tk
->lock
, flags
);
529 * timekeeping_notify - Install a new clock source
530 * @clock: pointer to the clock source
532 * This function is called from clocksource.c after a new, better clock
533 * source has been registered. The caller holds the clocksource_mutex.
535 void timekeeping_notify(struct clocksource
*clock
)
537 struct timekeeper
*tk
= &timekeeper
;
539 if (tk
->clock
== clock
)
541 stop_machine(change_clocksource
, clock
, NULL
);
546 * ktime_get_real - get the real (wall-) time in ktime_t format
548 * returns the time in ktime_t format
550 ktime_t
ktime_get_real(void)
554 getnstimeofday(&now
);
556 return timespec_to_ktime(now
);
558 EXPORT_SYMBOL_GPL(ktime_get_real
);
561 * getrawmonotonic - Returns the raw monotonic time in a timespec
562 * @ts: pointer to the timespec to be set
564 * Returns the raw monotonic time (completely un-modified by ntp)
566 void getrawmonotonic(struct timespec
*ts
)
568 struct timekeeper
*tk
= &timekeeper
;
573 seq
= read_seqbegin(&tk
->lock
);
574 nsecs
= timekeeping_get_ns_raw(tk
);
577 } while (read_seqretry(&tk
->lock
, seq
));
579 timespec_add_ns(ts
, nsecs
);
581 EXPORT_SYMBOL(getrawmonotonic
);
584 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
586 int timekeeping_valid_for_hres(void)
588 struct timekeeper
*tk
= &timekeeper
;
593 seq
= read_seqbegin(&tk
->lock
);
595 ret
= tk
->clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
597 } while (read_seqretry(&tk
->lock
, seq
));
603 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
605 u64
timekeeping_max_deferment(void)
607 struct timekeeper
*tk
= &timekeeper
;
612 seq
= read_seqbegin(&tk
->lock
);
614 ret
= tk
->clock
->max_idle_ns
;
616 } while (read_seqretry(&tk
->lock
, seq
));
622 * read_persistent_clock - Return time from the persistent clock.
624 * Weak dummy function for arches that do not yet support it.
625 * Reads the time from the battery backed persistent clock.
626 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
628 * XXX - Do be sure to remove it once all arches implement it.
630 void __attribute__((weak
)) read_persistent_clock(struct timespec
*ts
)
637 * read_boot_clock - Return time of the system start.
639 * Weak dummy function for arches that do not yet support it.
640 * Function to read the exact time the system has been started.
641 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
643 * XXX - Do be sure to remove it once all arches implement it.
645 void __attribute__((weak
)) read_boot_clock(struct timespec
*ts
)
652 * timekeeping_init - Initializes the clocksource and common timekeeping values
654 void __init
timekeeping_init(void)
656 struct timekeeper
*tk
= &timekeeper
;
657 struct clocksource
*clock
;
659 struct timespec now
, boot
, tmp
;
661 read_persistent_clock(&now
);
662 if (!timespec_valid_strict(&now
)) {
663 pr_warn("WARNING: Persistent clock returned invalid value!\n"
664 " Check your CMOS/BIOS settings.\n");
669 read_boot_clock(&boot
);
670 if (!timespec_valid_strict(&boot
)) {
671 pr_warn("WARNING: Boot clock returned invalid value!\n"
672 " Check your CMOS/BIOS settings.\n");
677 seqlock_init(&tk
->lock
);
681 write_seqlock_irqsave(&tk
->lock
, flags
);
682 clock
= clocksource_default_clock();
684 clock
->enable(clock
);
685 tk_setup_internals(tk
, clock
);
687 tk_set_xtime(tk
, &now
);
688 tk
->raw_time
.tv_sec
= 0;
689 tk
->raw_time
.tv_nsec
= 0;
690 if (boot
.tv_sec
== 0 && boot
.tv_nsec
== 0)
693 set_normalized_timespec(&tmp
, -boot
.tv_sec
, -boot
.tv_nsec
);
694 tk_set_wall_to_mono(tk
, tmp
);
698 tk_set_sleep_time(tk
, tmp
);
700 write_sequnlock_irqrestore(&tk
->lock
, flags
);
703 /* time in seconds when suspend began */
704 static struct timespec timekeeping_suspend_time
;
707 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
708 * @delta: pointer to a timespec delta value
710 * Takes a timespec offset measuring a suspend interval and properly
711 * adds the sleep offset to the timekeeping variables.
713 static void __timekeeping_inject_sleeptime(struct timekeeper
*tk
,
714 struct timespec
*delta
)
716 if (!timespec_valid_strict(delta
)) {
717 printk(KERN_WARNING
"__timekeeping_inject_sleeptime: Invalid "
718 "sleep delta value!\n");
721 tk_xtime_add(tk
, delta
);
722 tk_set_wall_to_mono(tk
, timespec_sub(tk
->wall_to_monotonic
, *delta
));
723 tk_set_sleep_time(tk
, timespec_add(tk
->total_sleep_time
, *delta
));
727 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
728 * @delta: pointer to a timespec delta value
730 * This hook is for architectures that cannot support read_persistent_clock
731 * because their RTC/persistent clock is only accessible when irqs are enabled.
733 * This function should only be called by rtc_resume(), and allows
734 * a suspend offset to be injected into the timekeeping values.
736 void timekeeping_inject_sleeptime(struct timespec
*delta
)
738 struct timekeeper
*tk
= &timekeeper
;
742 /* Make sure we don't set the clock twice */
743 read_persistent_clock(&ts
);
744 if (!(ts
.tv_sec
== 0 && ts
.tv_nsec
== 0))
747 write_seqlock_irqsave(&tk
->lock
, flags
);
749 timekeeping_forward_now(tk
);
751 __timekeeping_inject_sleeptime(tk
, delta
);
753 timekeeping_update(tk
, true);
755 write_sequnlock_irqrestore(&tk
->lock
, flags
);
757 /* signal hrtimers about time change */
762 * timekeeping_resume - Resumes the generic timekeeping subsystem.
764 * This is for the generic clocksource timekeeping.
765 * xtime/wall_to_monotonic/jiffies/etc are
766 * still managed by arch specific suspend/resume code.
768 static void timekeeping_resume(void)
770 struct timekeeper
*tk
= &timekeeper
;
774 read_persistent_clock(&ts
);
776 clocksource_resume();
778 write_seqlock_irqsave(&tk
->lock
, flags
);
780 if (timespec_compare(&ts
, &timekeeping_suspend_time
) > 0) {
781 ts
= timespec_sub(ts
, timekeeping_suspend_time
);
782 __timekeeping_inject_sleeptime(tk
, &ts
);
784 /* re-base the last cycle value */
785 tk
->clock
->cycle_last
= tk
->clock
->read(tk
->clock
);
787 timekeeping_suspended
= 0;
788 timekeeping_update(tk
, false);
789 write_sequnlock_irqrestore(&tk
->lock
, flags
);
791 touch_softlockup_watchdog();
793 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
795 /* Resume hrtimers */
799 static int timekeeping_suspend(void)
801 struct timekeeper
*tk
= &timekeeper
;
803 struct timespec delta
, delta_delta
;
804 static struct timespec old_delta
;
806 read_persistent_clock(&timekeeping_suspend_time
);
808 write_seqlock_irqsave(&tk
->lock
, flags
);
809 timekeeping_forward_now(tk
);
810 timekeeping_suspended
= 1;
813 * To avoid drift caused by repeated suspend/resumes,
814 * which each can add ~1 second drift error,
815 * try to compensate so the difference in system time
816 * and persistent_clock time stays close to constant.
818 delta
= timespec_sub(tk_xtime(tk
), timekeeping_suspend_time
);
819 delta_delta
= timespec_sub(delta
, old_delta
);
820 if (abs(delta_delta
.tv_sec
) >= 2) {
822 * if delta_delta is too large, assume time correction
823 * has occured and set old_delta to the current delta.
827 /* Otherwise try to adjust old_system to compensate */
828 timekeeping_suspend_time
=
829 timespec_add(timekeeping_suspend_time
, delta_delta
);
831 write_sequnlock_irqrestore(&tk
->lock
, flags
);
833 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
834 clocksource_suspend();
839 /* sysfs resume/suspend bits for timekeeping */
840 static struct syscore_ops timekeeping_syscore_ops
= {
841 .resume
= timekeeping_resume
,
842 .suspend
= timekeeping_suspend
,
845 static int __init
timekeeping_init_ops(void)
847 register_syscore_ops(&timekeeping_syscore_ops
);
851 device_initcall(timekeeping_init_ops
);
854 * If the error is already larger, we look ahead even further
855 * to compensate for late or lost adjustments.
857 static __always_inline
int timekeeping_bigadjust(struct timekeeper
*tk
,
858 s64 error
, s64
*interval
,
866 * Use the current error value to determine how much to look ahead.
867 * The larger the error the slower we adjust for it to avoid problems
868 * with losing too many ticks, otherwise we would overadjust and
869 * produce an even larger error. The smaller the adjustment the
870 * faster we try to adjust for it, as lost ticks can do less harm
871 * here. This is tuned so that an error of about 1 msec is adjusted
872 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
874 error2
= tk
->ntp_error
>> (NTP_SCALE_SHIFT
+ 22 - 2 * SHIFT_HZ
);
875 error2
= abs(error2
);
876 for (look_ahead
= 0; error2
> 0; look_ahead
++)
880 * Now calculate the error in (1 << look_ahead) ticks, but first
881 * remove the single look ahead already included in the error.
883 tick_error
= ntp_tick_length() >> (tk
->ntp_error_shift
+ 1);
884 tick_error
-= tk
->xtime_interval
>> 1;
885 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
887 /* Finally calculate the adjustment shift value. */
892 *interval
= -*interval
;
896 for (adj
= 0; error
> i
; adj
++)
905 * Adjust the multiplier to reduce the error value,
906 * this is optimized for the most common adjustments of -1,0,1,
907 * for other values we can do a bit more work.
909 static void timekeeping_adjust(struct timekeeper
*tk
, s64 offset
)
911 s64 error
, interval
= tk
->cycle_interval
;
915 * The point of this is to check if the error is greater than half
918 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
920 * Note we subtract one in the shift, so that error is really error*2.
921 * This "saves" dividing(shifting) interval twice, but keeps the
922 * (error > interval) comparison as still measuring if error is
923 * larger than half an interval.
925 * Note: It does not "save" on aggravation when reading the code.
927 error
= tk
->ntp_error
>> (tk
->ntp_error_shift
- 1);
928 if (error
> interval
) {
930 * We now divide error by 4(via shift), which checks if
931 * the error is greater than twice the interval.
932 * If it is greater, we need a bigadjust, if its smaller,
933 * we can adjust by 1.
937 * XXX - In update_wall_time, we round up to the next
938 * nanosecond, and store the amount rounded up into
939 * the error. This causes the likely below to be unlikely.
941 * The proper fix is to avoid rounding up by using
942 * the high precision tk->xtime_nsec instead of
943 * xtime.tv_nsec everywhere. Fixing this will take some
946 if (likely(error
<= interval
))
949 adj
= timekeeping_bigadjust(tk
, error
, &interval
, &offset
);
951 if (error
< -interval
) {
952 /* See comment above, this is just switched for the negative */
954 if (likely(error
>= -interval
)) {
956 interval
= -interval
;
959 adj
= timekeeping_bigadjust(tk
, error
, &interval
, &offset
);
966 if (unlikely(tk
->clock
->maxadj
&&
967 (tk
->mult
+ adj
> tk
->clock
->mult
+ tk
->clock
->maxadj
))) {
968 printk_once(KERN_WARNING
969 "Adjusting %s more than 11%% (%ld vs %ld)\n",
970 tk
->clock
->name
, (long)tk
->mult
+ adj
,
971 (long)tk
->clock
->mult
+ tk
->clock
->maxadj
);
974 * So the following can be confusing.
976 * To keep things simple, lets assume adj == 1 for now.
978 * When adj != 1, remember that the interval and offset values
979 * have been appropriately scaled so the math is the same.
981 * The basic idea here is that we're increasing the multiplier
982 * by one, this causes the xtime_interval to be incremented by
983 * one cycle_interval. This is because:
984 * xtime_interval = cycle_interval * mult
985 * So if mult is being incremented by one:
986 * xtime_interval = cycle_interval * (mult + 1)
988 * xtime_interval = (cycle_interval * mult) + cycle_interval
989 * Which can be shortened to:
990 * xtime_interval += cycle_interval
992 * So offset stores the non-accumulated cycles. Thus the current
993 * time (in shifted nanoseconds) is:
994 * now = (offset * adj) + xtime_nsec
995 * Now, even though we're adjusting the clock frequency, we have
996 * to keep time consistent. In other words, we can't jump back
997 * in time, and we also want to avoid jumping forward in time.
999 * So given the same offset value, we need the time to be the same
1000 * both before and after the freq adjustment.
1001 * now = (offset * adj_1) + xtime_nsec_1
1002 * now = (offset * adj_2) + xtime_nsec_2
1004 * (offset * adj_1) + xtime_nsec_1 =
1005 * (offset * adj_2) + xtime_nsec_2
1009 * (offset * adj_1) + xtime_nsec_1 =
1010 * (offset * (adj_1+1)) + xtime_nsec_2
1011 * (offset * adj_1) + xtime_nsec_1 =
1012 * (offset * adj_1) + offset + xtime_nsec_2
1013 * Canceling the sides:
1014 * xtime_nsec_1 = offset + xtime_nsec_2
1016 * xtime_nsec_2 = xtime_nsec_1 - offset
1017 * Which simplfies to:
1018 * xtime_nsec -= offset
1020 * XXX - TODO: Doc ntp_error calculation.
1023 tk
->xtime_interval
+= interval
;
1024 tk
->xtime_nsec
-= offset
;
1025 tk
->ntp_error
-= (interval
- offset
) << tk
->ntp_error_shift
;
1029 * It may be possible that when we entered this function, xtime_nsec
1030 * was very small. Further, if we're slightly speeding the clocksource
1031 * in the code above, its possible the required corrective factor to
1032 * xtime_nsec could cause it to underflow.
1034 * Now, since we already accumulated the second, cannot simply roll
1035 * the accumulated second back, since the NTP subsystem has been
1036 * notified via second_overflow. So instead we push xtime_nsec forward
1037 * by the amount we underflowed, and add that amount into the error.
1039 * We'll correct this error next time through this function, when
1040 * xtime_nsec is not as small.
1042 if (unlikely((s64
)tk
->xtime_nsec
< 0)) {
1043 s64 neg
= -(s64
)tk
->xtime_nsec
;
1045 tk
->ntp_error
+= neg
<< tk
->ntp_error_shift
;
1051 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1053 * Helper function that accumulates a the nsecs greater then a second
1054 * from the xtime_nsec field to the xtime_secs field.
1055 * It also calls into the NTP code to handle leapsecond processing.
1058 static inline void accumulate_nsecs_to_secs(struct timekeeper
*tk
)
1060 u64 nsecps
= (u64
)NSEC_PER_SEC
<< tk
->shift
;
1062 while (tk
->xtime_nsec
>= nsecps
) {
1065 tk
->xtime_nsec
-= nsecps
;
1068 /* Figure out if its a leap sec and apply if needed */
1069 leap
= second_overflow(tk
->xtime_sec
);
1070 if (unlikely(leap
)) {
1073 tk
->xtime_sec
+= leap
;
1077 tk_set_wall_to_mono(tk
,
1078 timespec_sub(tk
->wall_to_monotonic
, ts
));
1080 clock_was_set_delayed();
1086 * logarithmic_accumulation - shifted accumulation of cycles
1088 * This functions accumulates a shifted interval of cycles into
1089 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1092 * Returns the unconsumed cycles.
1094 static cycle_t
logarithmic_accumulation(struct timekeeper
*tk
, cycle_t offset
,
1099 /* If the offset is smaller then a shifted interval, do nothing */
1100 if (offset
< tk
->cycle_interval
<<shift
)
1103 /* Accumulate one shifted interval */
1104 offset
-= tk
->cycle_interval
<< shift
;
1105 tk
->clock
->cycle_last
+= tk
->cycle_interval
<< shift
;
1107 tk
->xtime_nsec
+= tk
->xtime_interval
<< shift
;
1108 accumulate_nsecs_to_secs(tk
);
1110 /* Accumulate raw time */
1111 raw_nsecs
= tk
->raw_interval
<< shift
;
1112 raw_nsecs
+= tk
->raw_time
.tv_nsec
;
1113 if (raw_nsecs
>= NSEC_PER_SEC
) {
1114 u64 raw_secs
= raw_nsecs
;
1115 raw_nsecs
= do_div(raw_secs
, NSEC_PER_SEC
);
1116 tk
->raw_time
.tv_sec
+= raw_secs
;
1118 tk
->raw_time
.tv_nsec
= raw_nsecs
;
1120 /* Accumulate error between NTP and clock interval */
1121 tk
->ntp_error
+= ntp_tick_length() << shift
;
1122 tk
->ntp_error
-= (tk
->xtime_interval
+ tk
->xtime_remainder
) <<
1123 (tk
->ntp_error_shift
+ shift
);
1129 * update_wall_time - Uses the current clocksource to increment the wall time
1132 static void update_wall_time(void)
1134 struct clocksource
*clock
;
1135 struct timekeeper
*tk
= &timekeeper
;
1137 int shift
= 0, maxshift
;
1138 unsigned long flags
;
1141 write_seqlock_irqsave(&tk
->lock
, flags
);
1143 /* Make sure we're fully resumed: */
1144 if (unlikely(timekeeping_suspended
))
1149 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1150 offset
= tk
->cycle_interval
;
1152 offset
= (clock
->read(clock
) - clock
->cycle_last
) & clock
->mask
;
1155 /* Check if there's really nothing to do */
1156 if (offset
< tk
->cycle_interval
)
1160 * With NO_HZ we may have to accumulate many cycle_intervals
1161 * (think "ticks") worth of time at once. To do this efficiently,
1162 * we calculate the largest doubling multiple of cycle_intervals
1163 * that is smaller than the offset. We then accumulate that
1164 * chunk in one go, and then try to consume the next smaller
1167 shift
= ilog2(offset
) - ilog2(tk
->cycle_interval
);
1168 shift
= max(0, shift
);
1169 /* Bound shift to one less than what overflows tick_length */
1170 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
1171 shift
= min(shift
, maxshift
);
1172 while (offset
>= tk
->cycle_interval
) {
1173 offset
= logarithmic_accumulation(tk
, offset
, shift
);
1174 if (offset
< tk
->cycle_interval
<<shift
)
1178 /* correct the clock when NTP error is too big */
1179 timekeeping_adjust(tk
, offset
);
1183 * Store only full nanoseconds into xtime_nsec after rounding
1184 * it up and add the remainder to the error difference.
1185 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1186 * by truncating the remainder in vsyscalls. However, it causes
1187 * additional work to be done in timekeeping_adjust(). Once
1188 * the vsyscall implementations are converted to use xtime_nsec
1189 * (shifted nanoseconds), this can be killed.
1191 remainder
= tk
->xtime_nsec
& ((1ULL << tk
->shift
) - 1);
1192 tk
->xtime_nsec
-= remainder
;
1193 tk
->xtime_nsec
+= 1ULL << tk
->shift
;
1194 tk
->ntp_error
+= remainder
<< tk
->ntp_error_shift
;
1197 * Finally, make sure that after the rounding
1198 * xtime_nsec isn't larger than NSEC_PER_SEC
1200 accumulate_nsecs_to_secs(tk
);
1202 timekeeping_update(tk
, false);
1205 write_sequnlock_irqrestore(&tk
->lock
, flags
);
1210 * getboottime - Return the real time of system boot.
1211 * @ts: pointer to the timespec to be set
1213 * Returns the wall-time of boot in a timespec.
1215 * This is based on the wall_to_monotonic offset and the total suspend
1216 * time. Calls to settimeofday will affect the value returned (which
1217 * basically means that however wrong your real time clock is at boot time,
1218 * you get the right time here).
1220 void getboottime(struct timespec
*ts
)
1222 struct timekeeper
*tk
= &timekeeper
;
1223 struct timespec boottime
= {
1224 .tv_sec
= tk
->wall_to_monotonic
.tv_sec
+
1225 tk
->total_sleep_time
.tv_sec
,
1226 .tv_nsec
= tk
->wall_to_monotonic
.tv_nsec
+
1227 tk
->total_sleep_time
.tv_nsec
1230 set_normalized_timespec(ts
, -boottime
.tv_sec
, -boottime
.tv_nsec
);
1232 EXPORT_SYMBOL_GPL(getboottime
);
1235 * get_monotonic_boottime - Returns monotonic time since boot
1236 * @ts: pointer to the timespec to be set
1238 * Returns the monotonic time since boot in a timespec.
1240 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1241 * includes the time spent in suspend.
1243 void get_monotonic_boottime(struct timespec
*ts
)
1245 struct timekeeper
*tk
= &timekeeper
;
1246 struct timespec tomono
, sleep
;
1249 WARN_ON(timekeeping_suspended
);
1252 seq
= read_seqbegin(&tk
->lock
);
1253 ts
->tv_sec
= tk
->xtime_sec
;
1254 ts
->tv_nsec
= timekeeping_get_ns(tk
);
1255 tomono
= tk
->wall_to_monotonic
;
1256 sleep
= tk
->total_sleep_time
;
1258 } while (read_seqretry(&tk
->lock
, seq
));
1260 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
+ sleep
.tv_sec
,
1261 ts
->tv_nsec
+ tomono
.tv_nsec
+ sleep
.tv_nsec
);
1263 EXPORT_SYMBOL_GPL(get_monotonic_boottime
);
1266 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1268 * Returns the monotonic time since boot in a ktime
1270 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1271 * includes the time spent in suspend.
1273 ktime_t
ktime_get_boottime(void)
1277 get_monotonic_boottime(&ts
);
1278 return timespec_to_ktime(ts
);
1280 EXPORT_SYMBOL_GPL(ktime_get_boottime
);
1283 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1284 * @ts: pointer to the timespec to be converted
1286 void monotonic_to_bootbased(struct timespec
*ts
)
1288 struct timekeeper
*tk
= &timekeeper
;
1290 *ts
= timespec_add(*ts
, tk
->total_sleep_time
);
1292 EXPORT_SYMBOL_GPL(monotonic_to_bootbased
);
1294 unsigned long get_seconds(void)
1296 struct timekeeper
*tk
= &timekeeper
;
1298 return tk
->xtime_sec
;
1300 EXPORT_SYMBOL(get_seconds
);
1302 struct timespec
__current_kernel_time(void)
1304 struct timekeeper
*tk
= &timekeeper
;
1306 return tk_xtime(tk
);
1309 struct timespec
current_kernel_time(void)
1311 struct timekeeper
*tk
= &timekeeper
;
1312 struct timespec now
;
1316 seq
= read_seqbegin(&tk
->lock
);
1319 } while (read_seqretry(&tk
->lock
, seq
));
1323 EXPORT_SYMBOL(current_kernel_time
);
1325 struct timespec
get_monotonic_coarse(void)
1327 struct timekeeper
*tk
= &timekeeper
;
1328 struct timespec now
, mono
;
1332 seq
= read_seqbegin(&tk
->lock
);
1335 mono
= tk
->wall_to_monotonic
;
1336 } while (read_seqretry(&tk
->lock
, seq
));
1338 set_normalized_timespec(&now
, now
.tv_sec
+ mono
.tv_sec
,
1339 now
.tv_nsec
+ mono
.tv_nsec
);
1344 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1345 * without sampling the sequence number in xtime_lock.
1346 * jiffies is defined in the linker script...
1348 void do_timer(unsigned long ticks
)
1350 jiffies_64
+= ticks
;
1352 calc_global_load(ticks
);
1356 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1357 * and sleep offsets.
1358 * @xtim: pointer to timespec to be set with xtime
1359 * @wtom: pointer to timespec to be set with wall_to_monotonic
1360 * @sleep: pointer to timespec to be set with time in suspend
1362 void get_xtime_and_monotonic_and_sleep_offset(struct timespec
*xtim
,
1363 struct timespec
*wtom
, struct timespec
*sleep
)
1365 struct timekeeper
*tk
= &timekeeper
;
1369 seq
= read_seqbegin(&tk
->lock
);
1370 *xtim
= tk_xtime(tk
);
1371 *wtom
= tk
->wall_to_monotonic
;
1372 *sleep
= tk
->total_sleep_time
;
1373 } while (read_seqretry(&tk
->lock
, seq
));
1376 #ifdef CONFIG_HIGH_RES_TIMERS
1378 * ktime_get_update_offsets - hrtimer helper
1379 * @offs_real: pointer to storage for monotonic -> realtime offset
1380 * @offs_boot: pointer to storage for monotonic -> boottime offset
1382 * Returns current monotonic time and updates the offsets
1383 * Called from hrtimer_interupt() or retrigger_next_event()
1385 ktime_t
ktime_get_update_offsets(ktime_t
*offs_real
, ktime_t
*offs_boot
)
1387 struct timekeeper
*tk
= &timekeeper
;
1393 seq
= read_seqbegin(&tk
->lock
);
1395 secs
= tk
->xtime_sec
;
1396 nsecs
= timekeeping_get_ns(tk
);
1398 *offs_real
= tk
->offs_real
;
1399 *offs_boot
= tk
->offs_boot
;
1400 } while (read_seqretry(&tk
->lock
, seq
));
1402 now
= ktime_add_ns(ktime_set(secs
, 0), nsecs
);
1403 now
= ktime_sub(now
, *offs_real
);
1409 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1411 ktime_t
ktime_get_monotonic_offset(void)
1413 struct timekeeper
*tk
= &timekeeper
;
1415 struct timespec wtom
;
1418 seq
= read_seqbegin(&tk
->lock
);
1419 wtom
= tk
->wall_to_monotonic
;
1420 } while (read_seqretry(&tk
->lock
, seq
));
1422 return timespec_to_ktime(wtom
);
1424 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset
);
1427 * xtime_update() - advances the timekeeping infrastructure
1428 * @ticks: number of ticks, that have elapsed since the last call.
1430 * Must be called with interrupts disabled.
1432 void xtime_update(unsigned long ticks
)
1434 write_seqlock(&xtime_lock
);
1436 write_sequnlock(&xtime_lock
);