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 /* time spent in suspend */
69 struct timespec total_sleep_time;
70 /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */
71 struct timespec raw_time;
72 /* Offset clock monotonic -> clock realtime */
74 /* Offset clock monotonic -> clock boottime */
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 = 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 += ts->tv_nsec << tk->shift;
121 * timekeeper_setup_internals - Set up internals to use clocksource clock.
123 * @clock: Pointer to clocksource.
125 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
126 * pair and interval request.
128 * Unless you're the timekeeping code, you should not be using this!
130 static void timekeeper_setup_internals(struct clocksource *clock)
133 u64 tmp, ntpinterval;
134 struct clocksource *old_clock;
136 old_clock = timekeeper.clock;
137 timekeeper.clock = clock;
138 clock->cycle_last = clock->read(clock);
140 /* Do the ns -> cycle conversion first, using original mult */
141 tmp = NTP_INTERVAL_LENGTH;
142 tmp <<= clock->shift;
144 tmp += clock->mult/2;
145 do_div(tmp, clock->mult);
149 interval = (cycle_t) tmp;
150 timekeeper.cycle_interval = interval;
152 /* Go back from cycles -> shifted ns */
153 timekeeper.xtime_interval = (u64) interval * clock->mult;
154 timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval;
155 timekeeper.raw_interval =
156 ((u64) interval * clock->mult) >> clock->shift;
158 /* if changing clocks, convert xtime_nsec shift units */
160 int shift_change = clock->shift - old_clock->shift;
161 if (shift_change < 0)
162 timekeeper.xtime_nsec >>= -shift_change;
164 timekeeper.xtime_nsec <<= shift_change;
166 timekeeper.shift = clock->shift;
168 timekeeper.ntp_error = 0;
169 timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
172 * The timekeeper keeps its own mult values for the currently
173 * active clocksource. These value will be adjusted via NTP
174 * to counteract clock drifting.
176 timekeeper.mult = clock->mult;
179 /* Timekeeper helper functions. */
180 static inline s64 timekeeping_get_ns(void)
182 cycle_t cycle_now, cycle_delta;
183 struct clocksource *clock;
186 /* read clocksource: */
187 clock = timekeeper.clock;
188 cycle_now = clock->read(clock);
190 /* calculate the delta since the last update_wall_time: */
191 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
193 nsec = cycle_delta * timekeeper.mult + timekeeper.xtime_nsec;
194 nsec >>= timekeeper.shift;
196 /* If arch requires, add in gettimeoffset() */
197 return nsec + arch_gettimeoffset();
200 static inline s64 timekeeping_get_ns_raw(void)
202 cycle_t cycle_now, cycle_delta;
203 struct clocksource *clock;
206 /* read clocksource: */
207 clock = timekeeper.clock;
208 cycle_now = clock->read(clock);
210 /* calculate the delta since the last update_wall_time: */
211 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
213 /* convert delta to nanoseconds. */
214 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
216 /* If arch requires, add in gettimeoffset() */
217 return nsec + arch_gettimeoffset();
220 static void update_rt_offset(void)
222 struct timespec tmp, *wtm = &timekeeper.wall_to_monotonic;
224 set_normalized_timespec(&tmp, -wtm->tv_sec, -wtm->tv_nsec);
225 timekeeper.offs_real = timespec_to_ktime(tmp);
228 /* must hold write on timekeeper.lock */
229 static void timekeeping_update(bool clearntp)
234 timekeeper.ntp_error = 0;
238 xt = tk_xtime(&timekeeper);
239 update_vsyscall(&xt, &timekeeper.wall_to_monotonic,
240 timekeeper.clock, timekeeper.mult);
245 * timekeeping_forward_now - update clock to the current time
247 * Forward the current clock to update its state since the last call to
248 * update_wall_time(). This is useful before significant clock changes,
249 * as it avoids having to deal with this time offset explicitly.
251 static void timekeeping_forward_now(void)
253 cycle_t cycle_now, cycle_delta;
254 struct clocksource *clock;
257 clock = timekeeper.clock;
258 cycle_now = clock->read(clock);
259 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
260 clock->cycle_last = cycle_now;
262 timekeeper.xtime_nsec += cycle_delta * timekeeper.mult;
264 /* If arch requires, add in gettimeoffset() */
265 timekeeper.xtime_nsec += arch_gettimeoffset() << timekeeper.shift;
267 tk_normalize_xtime(&timekeeper);
269 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
270 timespec_add_ns(&timekeeper.raw_time, nsec);
274 * getnstimeofday - Returns the time of day in a timespec
275 * @ts: pointer to the timespec to be set
277 * Returns the time of day in a timespec.
279 void getnstimeofday(struct timespec *ts)
284 WARN_ON(timekeeping_suspended);
287 seq = read_seqbegin(&timekeeper.lock);
289 ts->tv_sec = timekeeper.xtime_sec;
290 ts->tv_nsec = timekeeping_get_ns();
292 } while (read_seqretry(&timekeeper.lock, seq));
294 timespec_add_ns(ts, nsecs);
296 EXPORT_SYMBOL(getnstimeofday);
298 ktime_t ktime_get(void)
303 WARN_ON(timekeeping_suspended);
306 seq = read_seqbegin(&timekeeper.lock);
307 secs = timekeeper.xtime_sec +
308 timekeeper.wall_to_monotonic.tv_sec;
309 nsecs = timekeeping_get_ns() +
310 timekeeper.wall_to_monotonic.tv_nsec;
312 } while (read_seqretry(&timekeeper.lock, seq));
314 * Use ktime_set/ktime_add_ns to create a proper ktime on
315 * 32-bit architectures without CONFIG_KTIME_SCALAR.
317 return ktime_add_ns(ktime_set(secs, 0), nsecs);
319 EXPORT_SYMBOL_GPL(ktime_get);
322 * ktime_get_ts - get the monotonic clock in timespec format
323 * @ts: pointer to timespec variable
325 * The function calculates the monotonic clock from the realtime
326 * clock and the wall_to_monotonic offset and stores the result
327 * in normalized timespec format in the variable pointed to by @ts.
329 void ktime_get_ts(struct timespec *ts)
331 struct timespec tomono;
334 WARN_ON(timekeeping_suspended);
337 seq = read_seqbegin(&timekeeper.lock);
338 ts->tv_sec = timekeeper.xtime_sec;
339 ts->tv_nsec = timekeeping_get_ns();
340 tomono = timekeeper.wall_to_monotonic;
342 } while (read_seqretry(&timekeeper.lock, seq));
344 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
345 ts->tv_nsec + tomono.tv_nsec);
347 EXPORT_SYMBOL_GPL(ktime_get_ts);
349 #ifdef CONFIG_NTP_PPS
352 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
353 * @ts_raw: pointer to the timespec to be set to raw monotonic time
354 * @ts_real: pointer to the timespec to be set to the time of day
356 * This function reads both the time of day and raw monotonic time at the
357 * same time atomically and stores the resulting timestamps in timespec
360 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
363 s64 nsecs_raw, nsecs_real;
365 WARN_ON_ONCE(timekeeping_suspended);
368 seq = read_seqbegin(&timekeeper.lock);
370 *ts_raw = timekeeper.raw_time;
371 ts_real->tv_sec = timekeeper.xtime_sec;
372 ts_real->tv_nsec = 0;
374 nsecs_raw = timekeeping_get_ns_raw();
375 nsecs_real = timekeeping_get_ns();
377 } while (read_seqretry(&timekeeper.lock, seq));
379 timespec_add_ns(ts_raw, nsecs_raw);
380 timespec_add_ns(ts_real, nsecs_real);
382 EXPORT_SYMBOL(getnstime_raw_and_real);
384 #endif /* CONFIG_NTP_PPS */
387 * do_gettimeofday - Returns the time of day in a timeval
388 * @tv: pointer to the timeval to be set
390 * NOTE: Users should be converted to using getnstimeofday()
392 void do_gettimeofday(struct timeval *tv)
396 getnstimeofday(&now);
397 tv->tv_sec = now.tv_sec;
398 tv->tv_usec = now.tv_nsec/1000;
400 EXPORT_SYMBOL(do_gettimeofday);
403 * do_settimeofday - Sets the time of day
404 * @tv: pointer to the timespec variable containing the new time
406 * Sets the time of day to the new time and update NTP and notify hrtimers
408 int do_settimeofday(const struct timespec *tv)
410 struct timespec ts_delta, xt;
413 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
416 write_seqlock_irqsave(&timekeeper.lock, flags);
418 timekeeping_forward_now();
420 xt = tk_xtime(&timekeeper);
421 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
422 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
424 timekeeper.wall_to_monotonic =
425 timespec_sub(timekeeper.wall_to_monotonic, ts_delta);
427 tk_set_xtime(&timekeeper, tv);
429 timekeeping_update(true);
431 write_sequnlock_irqrestore(&timekeeper.lock, flags);
433 /* signal hrtimers about time change */
438 EXPORT_SYMBOL(do_settimeofday);
442 * timekeeping_inject_offset - Adds or subtracts from the current time.
443 * @tv: pointer to the timespec variable containing the offset
445 * Adds or subtracts an offset value from the current time.
447 int timekeeping_inject_offset(struct timespec *ts)
451 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
454 write_seqlock_irqsave(&timekeeper.lock, flags);
456 timekeeping_forward_now();
459 tk_xtime_add(&timekeeper, ts);
460 timekeeper.wall_to_monotonic =
461 timespec_sub(timekeeper.wall_to_monotonic, *ts);
463 timekeeping_update(true);
465 write_sequnlock_irqrestore(&timekeeper.lock, flags);
467 /* signal hrtimers about time change */
472 EXPORT_SYMBOL(timekeeping_inject_offset);
475 * change_clocksource - Swaps clocksources if a new one is available
477 * Accumulates current time interval and initializes new clocksource
479 static int change_clocksource(void *data)
481 struct clocksource *new, *old;
484 new = (struct clocksource *) data;
486 write_seqlock_irqsave(&timekeeper.lock, flags);
488 timekeeping_forward_now();
489 if (!new->enable || new->enable(new) == 0) {
490 old = timekeeper.clock;
491 timekeeper_setup_internals(new);
495 timekeeping_update(true);
497 write_sequnlock_irqrestore(&timekeeper.lock, flags);
503 * timekeeping_notify - Install a new clock source
504 * @clock: pointer to the clock source
506 * This function is called from clocksource.c after a new, better clock
507 * source has been registered. The caller holds the clocksource_mutex.
509 void timekeeping_notify(struct clocksource *clock)
511 if (timekeeper.clock == clock)
513 stop_machine(change_clocksource, clock, NULL);
518 * ktime_get_real - get the real (wall-) time in ktime_t format
520 * returns the time in ktime_t format
522 ktime_t ktime_get_real(void)
526 getnstimeofday(&now);
528 return timespec_to_ktime(now);
530 EXPORT_SYMBOL_GPL(ktime_get_real);
533 * getrawmonotonic - Returns the raw monotonic time in a timespec
534 * @ts: pointer to the timespec to be set
536 * Returns the raw monotonic time (completely un-modified by ntp)
538 void getrawmonotonic(struct timespec *ts)
544 seq = read_seqbegin(&timekeeper.lock);
545 nsecs = timekeeping_get_ns_raw();
546 *ts = timekeeper.raw_time;
548 } while (read_seqretry(&timekeeper.lock, seq));
550 timespec_add_ns(ts, nsecs);
552 EXPORT_SYMBOL(getrawmonotonic);
556 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
558 int timekeeping_valid_for_hres(void)
564 seq = read_seqbegin(&timekeeper.lock);
566 ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
568 } while (read_seqretry(&timekeeper.lock, seq));
574 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
576 u64 timekeeping_max_deferment(void)
582 seq = read_seqbegin(&timekeeper.lock);
584 ret = timekeeper.clock->max_idle_ns;
586 } while (read_seqretry(&timekeeper.lock, seq));
592 * read_persistent_clock - Return time from the persistent clock.
594 * Weak dummy function for arches that do not yet support it.
595 * Reads the time from the battery backed persistent clock.
596 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
598 * XXX - Do be sure to remove it once all arches implement it.
600 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
607 * read_boot_clock - Return time of the system start.
609 * Weak dummy function for arches that do not yet support it.
610 * Function to read the exact time the system has been started.
611 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
613 * XXX - Do be sure to remove it once all arches implement it.
615 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
622 * timekeeping_init - Initializes the clocksource and common timekeeping values
624 void __init timekeeping_init(void)
626 struct clocksource *clock;
628 struct timespec now, boot;
630 read_persistent_clock(&now);
631 read_boot_clock(&boot);
633 seqlock_init(&timekeeper.lock);
637 write_seqlock_irqsave(&timekeeper.lock, flags);
638 clock = clocksource_default_clock();
640 clock->enable(clock);
641 timekeeper_setup_internals(clock);
643 tk_set_xtime(&timekeeper, &now);
644 timekeeper.raw_time.tv_sec = 0;
645 timekeeper.raw_time.tv_nsec = 0;
646 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
647 boot = tk_xtime(&timekeeper);
649 set_normalized_timespec(&timekeeper.wall_to_monotonic,
650 -boot.tv_sec, -boot.tv_nsec);
652 timekeeper.total_sleep_time.tv_sec = 0;
653 timekeeper.total_sleep_time.tv_nsec = 0;
654 write_sequnlock_irqrestore(&timekeeper.lock, flags);
657 /* time in seconds when suspend began */
658 static struct timespec timekeeping_suspend_time;
660 static void update_sleep_time(struct timespec t)
662 timekeeper.total_sleep_time = t;
663 timekeeper.offs_boot = timespec_to_ktime(t);
667 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
668 * @delta: pointer to a timespec delta value
670 * Takes a timespec offset measuring a suspend interval and properly
671 * adds the sleep offset to the timekeeping variables.
673 static void __timekeeping_inject_sleeptime(struct timespec *delta)
675 if (!timespec_valid(delta)) {
676 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
677 "sleep delta value!\n");
681 tk_xtime_add(&timekeeper, delta);
682 timekeeper.wall_to_monotonic =
683 timespec_sub(timekeeper.wall_to_monotonic, *delta);
684 update_sleep_time(timespec_add(timekeeper.total_sleep_time, *delta));
689 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
690 * @delta: pointer to a timespec delta value
692 * This hook is for architectures that cannot support read_persistent_clock
693 * because their RTC/persistent clock is only accessible when irqs are enabled.
695 * This function should only be called by rtc_resume(), and allows
696 * a suspend offset to be injected into the timekeeping values.
698 void timekeeping_inject_sleeptime(struct timespec *delta)
703 /* Make sure we don't set the clock twice */
704 read_persistent_clock(&ts);
705 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
708 write_seqlock_irqsave(&timekeeper.lock, flags);
710 timekeeping_forward_now();
712 __timekeeping_inject_sleeptime(delta);
714 timekeeping_update(true);
716 write_sequnlock_irqrestore(&timekeeper.lock, flags);
718 /* signal hrtimers about time change */
724 * timekeeping_resume - Resumes the generic timekeeping subsystem.
726 * This is for the generic clocksource timekeeping.
727 * xtime/wall_to_monotonic/jiffies/etc are
728 * still managed by arch specific suspend/resume code.
730 static void timekeeping_resume(void)
735 read_persistent_clock(&ts);
737 clocksource_resume();
739 write_seqlock_irqsave(&timekeeper.lock, flags);
741 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
742 ts = timespec_sub(ts, timekeeping_suspend_time);
743 __timekeeping_inject_sleeptime(&ts);
745 /* re-base the last cycle value */
746 timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock);
747 timekeeper.ntp_error = 0;
748 timekeeping_suspended = 0;
749 write_sequnlock_irqrestore(&timekeeper.lock, flags);
751 touch_softlockup_watchdog();
753 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
755 /* Resume hrtimers */
759 static int timekeeping_suspend(void)
762 struct timespec delta, delta_delta;
763 static struct timespec old_delta;
765 read_persistent_clock(&timekeeping_suspend_time);
767 write_seqlock_irqsave(&timekeeper.lock, flags);
768 timekeeping_forward_now();
769 timekeeping_suspended = 1;
772 * To avoid drift caused by repeated suspend/resumes,
773 * which each can add ~1 second drift error,
774 * try to compensate so the difference in system time
775 * and persistent_clock time stays close to constant.
777 delta = timespec_sub(tk_xtime(&timekeeper), timekeeping_suspend_time);
778 delta_delta = timespec_sub(delta, old_delta);
779 if (abs(delta_delta.tv_sec) >= 2) {
781 * if delta_delta is too large, assume time correction
782 * has occured and set old_delta to the current delta.
786 /* Otherwise try to adjust old_system to compensate */
787 timekeeping_suspend_time =
788 timespec_add(timekeeping_suspend_time, delta_delta);
790 write_sequnlock_irqrestore(&timekeeper.lock, flags);
792 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
793 clocksource_suspend();
798 /* sysfs resume/suspend bits for timekeeping */
799 static struct syscore_ops timekeeping_syscore_ops = {
800 .resume = timekeeping_resume,
801 .suspend = timekeeping_suspend,
804 static int __init timekeeping_init_ops(void)
806 register_syscore_ops(&timekeeping_syscore_ops);
810 device_initcall(timekeeping_init_ops);
813 * If the error is already larger, we look ahead even further
814 * to compensate for late or lost adjustments.
816 static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval,
824 * Use the current error value to determine how much to look ahead.
825 * The larger the error the slower we adjust for it to avoid problems
826 * with losing too many ticks, otherwise we would overadjust and
827 * produce an even larger error. The smaller the adjustment the
828 * faster we try to adjust for it, as lost ticks can do less harm
829 * here. This is tuned so that an error of about 1 msec is adjusted
830 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
832 error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
833 error2 = abs(error2);
834 for (look_ahead = 0; error2 > 0; look_ahead++)
838 * Now calculate the error in (1 << look_ahead) ticks, but first
839 * remove the single look ahead already included in the error.
841 tick_error = ntp_tick_length() >> (timekeeper.ntp_error_shift + 1);
842 tick_error -= timekeeper.xtime_interval >> 1;
843 error = ((error - tick_error) >> look_ahead) + tick_error;
845 /* Finally calculate the adjustment shift value. */
850 *interval = -*interval;
854 for (adj = 0; error > i; adj++)
863 * Adjust the multiplier to reduce the error value,
864 * this is optimized for the most common adjustments of -1,0,1,
865 * for other values we can do a bit more work.
867 static void timekeeping_adjust(s64 offset)
869 s64 error, interval = timekeeper.cycle_interval;
873 * The point of this is to check if the error is greater than half
876 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
878 * Note we subtract one in the shift, so that error is really error*2.
879 * This "saves" dividing(shifting) interval twice, but keeps the
880 * (error > interval) comparison as still measuring if error is
881 * larger than half an interval.
883 * Note: It does not "save" on aggravation when reading the code.
885 error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1);
886 if (error > interval) {
888 * We now divide error by 4(via shift), which checks if
889 * the error is greater than twice the interval.
890 * If it is greater, we need a bigadjust, if its smaller,
891 * we can adjust by 1.
895 * XXX - In update_wall_time, we round up to the next
896 * nanosecond, and store the amount rounded up into
897 * the error. This causes the likely below to be unlikely.
899 * The proper fix is to avoid rounding up by using
900 * the high precision timekeeper.xtime_nsec instead of
901 * xtime.tv_nsec everywhere. Fixing this will take some
904 if (likely(error <= interval))
907 adj = timekeeping_bigadjust(error, &interval, &offset);
908 } else if (error < -interval) {
909 /* See comment above, this is just switched for the negative */
911 if (likely(error >= -interval)) {
913 interval = -interval;
916 adj = timekeeping_bigadjust(error, &interval, &offset);
917 } else /* No adjustment needed */
920 if (unlikely(timekeeper.clock->maxadj &&
921 (timekeeper.mult + adj >
922 timekeeper.clock->mult + timekeeper.clock->maxadj))) {
923 printk_once(KERN_WARNING
924 "Adjusting %s more than 11%% (%ld vs %ld)\n",
925 timekeeper.clock->name, (long)timekeeper.mult + adj,
926 (long)timekeeper.clock->mult +
927 timekeeper.clock->maxadj);
930 * So the following can be confusing.
932 * To keep things simple, lets assume adj == 1 for now.
934 * When adj != 1, remember that the interval and offset values
935 * have been appropriately scaled so the math is the same.
937 * The basic idea here is that we're increasing the multiplier
938 * by one, this causes the xtime_interval to be incremented by
939 * one cycle_interval. This is because:
940 * xtime_interval = cycle_interval * mult
941 * So if mult is being incremented by one:
942 * xtime_interval = cycle_interval * (mult + 1)
944 * xtime_interval = (cycle_interval * mult) + cycle_interval
945 * Which can be shortened to:
946 * xtime_interval += cycle_interval
948 * So offset stores the non-accumulated cycles. Thus the current
949 * time (in shifted nanoseconds) is:
950 * now = (offset * adj) + xtime_nsec
951 * Now, even though we're adjusting the clock frequency, we have
952 * to keep time consistent. In other words, we can't jump back
953 * in time, and we also want to avoid jumping forward in time.
955 * So given the same offset value, we need the time to be the same
956 * both before and after the freq adjustment.
957 * now = (offset * adj_1) + xtime_nsec_1
958 * now = (offset * adj_2) + xtime_nsec_2
960 * (offset * adj_1) + xtime_nsec_1 =
961 * (offset * adj_2) + xtime_nsec_2
965 * (offset * adj_1) + xtime_nsec_1 =
966 * (offset * (adj_1+1)) + xtime_nsec_2
967 * (offset * adj_1) + xtime_nsec_1 =
968 * (offset * adj_1) + offset + xtime_nsec_2
969 * Canceling the sides:
970 * xtime_nsec_1 = offset + xtime_nsec_2
972 * xtime_nsec_2 = xtime_nsec_1 - offset
973 * Which simplfies to:
974 * xtime_nsec -= offset
976 * XXX - TODO: Doc ntp_error calculation.
978 timekeeper.mult += adj;
979 timekeeper.xtime_interval += interval;
980 timekeeper.xtime_nsec -= offset;
981 timekeeper.ntp_error -= (interval - offset) <<
982 timekeeper.ntp_error_shift;
985 * It may be possible that when we entered this function, xtime_nsec
986 * was very small. Further, if we're slightly speeding the clocksource
987 * in the code above, its possible the required corrective factor to
988 * xtime_nsec could cause it to underflow.
990 * Now, since we already accumulated the second, cannot simply roll
991 * the accumulated second back, since the NTP subsystem has been
992 * notified via second_overflow. So instead we push xtime_nsec forward
993 * by the amount we underflowed, and add that amount into the error.
995 * We'll correct this error next time through this function, when
996 * xtime_nsec is not as small.
998 if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
999 s64 neg = -(s64)timekeeper.xtime_nsec;
1000 timekeeper.xtime_nsec = 0;
1001 timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
1008 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1010 * Helper function that accumulates a the nsecs greater then a second
1011 * from the xtime_nsec field to the xtime_secs field.
1012 * It also calls into the NTP code to handle leapsecond processing.
1015 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1017 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1019 while (tk->xtime_nsec >= nsecps) {
1022 tk->xtime_nsec -= nsecps;
1025 /* Figure out if its a leap sec and apply if needed */
1026 leap = second_overflow(tk->xtime_sec);
1027 tk->xtime_sec += leap;
1028 tk->wall_to_monotonic.tv_sec -= leap;
1030 clock_was_set_delayed();
1037 * logarithmic_accumulation - shifted accumulation of cycles
1039 * This functions accumulates a shifted interval of cycles into
1040 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1043 * Returns the unconsumed cycles.
1045 static cycle_t logarithmic_accumulation(cycle_t offset, u32 shift)
1049 /* If the offset is smaller than a shifted interval, do nothing */
1050 if (offset < timekeeper.cycle_interval<<shift)
1053 /* Accumulate one shifted interval */
1054 offset -= timekeeper.cycle_interval << shift;
1055 timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift;
1057 timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
1059 accumulate_nsecs_to_secs(&timekeeper);
1061 /* Accumulate raw time */
1062 raw_nsecs = timekeeper.raw_interval << shift;
1063 raw_nsecs += timekeeper.raw_time.tv_nsec;
1064 if (raw_nsecs >= NSEC_PER_SEC) {
1065 u64 raw_secs = raw_nsecs;
1066 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1067 timekeeper.raw_time.tv_sec += raw_secs;
1069 timekeeper.raw_time.tv_nsec = raw_nsecs;
1071 /* Accumulate error between NTP and clock interval */
1072 timekeeper.ntp_error += ntp_tick_length() << shift;
1073 timekeeper.ntp_error -=
1074 (timekeeper.xtime_interval + timekeeper.xtime_remainder) <<
1075 (timekeeper.ntp_error_shift + shift);
1082 * update_wall_time - Uses the current clocksource to increment the wall time
1085 static void update_wall_time(void)
1087 struct clocksource *clock;
1089 int shift = 0, maxshift;
1090 unsigned long flags;
1093 write_seqlock_irqsave(&timekeeper.lock, flags);
1095 /* Make sure we're fully resumed: */
1096 if (unlikely(timekeeping_suspended))
1099 clock = timekeeper.clock;
1101 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1102 offset = timekeeper.cycle_interval;
1104 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1108 * With NO_HZ we may have to accumulate many cycle_intervals
1109 * (think "ticks") worth of time at once. To do this efficiently,
1110 * we calculate the largest doubling multiple of cycle_intervals
1111 * that is smaller than the offset. We then accumulate that
1112 * chunk in one go, and then try to consume the next smaller
1115 shift = ilog2(offset) - ilog2(timekeeper.cycle_interval);
1116 shift = max(0, shift);
1117 /* Bound shift to one less than what overflows tick_length */
1118 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1119 shift = min(shift, maxshift);
1120 while (offset >= timekeeper.cycle_interval) {
1121 offset = logarithmic_accumulation(offset, shift);
1122 if(offset < timekeeper.cycle_interval<<shift)
1126 /* correct the clock when NTP error is too big */
1127 timekeeping_adjust(offset);
1131 * Store only full nanoseconds into xtime_nsec after rounding
1132 * it up and add the remainder to the error difference.
1133 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1134 * by truncating the remainder in vsyscalls. However, it causes
1135 * additional work to be done in timekeeping_adjust(). Once
1136 * the vsyscall implementations are converted to use xtime_nsec
1137 * (shifted nanoseconds), this can be killed.
1139 remainder = timekeeper.xtime_nsec & ((1 << timekeeper.shift) - 1);
1140 timekeeper.xtime_nsec -= remainder;
1141 timekeeper.xtime_nsec += 1 << timekeeper.shift;
1142 timekeeper.ntp_error += remainder << timekeeper.ntp_error_shift;
1145 * Finally, make sure that after the rounding
1146 * xtime_nsec isn't larger than NSEC_PER_SEC
1148 accumulate_nsecs_to_secs(&timekeeper);
1150 timekeeping_update(false);
1153 write_sequnlock_irqrestore(&timekeeper.lock, flags);
1158 * getboottime - Return the real time of system boot.
1159 * @ts: pointer to the timespec to be set
1161 * Returns the wall-time of boot in a timespec.
1163 * This is based on the wall_to_monotonic offset and the total suspend
1164 * time. Calls to settimeofday will affect the value returned (which
1165 * basically means that however wrong your real time clock is at boot time,
1166 * you get the right time here).
1168 void getboottime(struct timespec *ts)
1170 struct timespec boottime = {
1171 .tv_sec = timekeeper.wall_to_monotonic.tv_sec +
1172 timekeeper.total_sleep_time.tv_sec,
1173 .tv_nsec = timekeeper.wall_to_monotonic.tv_nsec +
1174 timekeeper.total_sleep_time.tv_nsec
1177 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1179 EXPORT_SYMBOL_GPL(getboottime);
1183 * get_monotonic_boottime - Returns monotonic time since boot
1184 * @ts: pointer to the timespec to be set
1186 * Returns the monotonic time since boot in a timespec.
1188 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1189 * includes the time spent in suspend.
1191 void get_monotonic_boottime(struct timespec *ts)
1193 struct timespec tomono, sleep;
1196 WARN_ON(timekeeping_suspended);
1199 seq = read_seqbegin(&timekeeper.lock);
1200 ts->tv_sec = timekeeper.xtime_sec;
1201 ts->tv_nsec = timekeeping_get_ns();
1202 tomono = timekeeper.wall_to_monotonic;
1203 sleep = timekeeper.total_sleep_time;
1205 } while (read_seqretry(&timekeeper.lock, seq));
1207 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec,
1208 ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec);
1210 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1213 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1215 * Returns the monotonic time since boot in a ktime
1217 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1218 * includes the time spent in suspend.
1220 ktime_t ktime_get_boottime(void)
1224 get_monotonic_boottime(&ts);
1225 return timespec_to_ktime(ts);
1227 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1230 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1231 * @ts: pointer to the timespec to be converted
1233 void monotonic_to_bootbased(struct timespec *ts)
1235 *ts = timespec_add(*ts, timekeeper.total_sleep_time);
1237 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1239 unsigned long get_seconds(void)
1241 return timekeeper.xtime_sec;
1243 EXPORT_SYMBOL(get_seconds);
1245 struct timespec __current_kernel_time(void)
1247 return tk_xtime(&timekeeper);
1250 struct timespec current_kernel_time(void)
1252 struct timespec now;
1256 seq = read_seqbegin(&timekeeper.lock);
1258 now = tk_xtime(&timekeeper);
1259 } while (read_seqretry(&timekeeper.lock, seq));
1263 EXPORT_SYMBOL(current_kernel_time);
1265 struct timespec get_monotonic_coarse(void)
1267 struct timespec now, mono;
1271 seq = read_seqbegin(&timekeeper.lock);
1273 now = tk_xtime(&timekeeper);
1274 mono = timekeeper.wall_to_monotonic;
1275 } while (read_seqretry(&timekeeper.lock, seq));
1277 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1278 now.tv_nsec + mono.tv_nsec);
1283 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1284 * without sampling the sequence number in xtime_lock.
1285 * jiffies is defined in the linker script...
1287 void do_timer(unsigned long ticks)
1289 jiffies_64 += ticks;
1291 calc_global_load(ticks);
1295 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1296 * and sleep offsets.
1297 * @xtim: pointer to timespec to be set with xtime
1298 * @wtom: pointer to timespec to be set with wall_to_monotonic
1299 * @sleep: pointer to timespec to be set with time in suspend
1301 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1302 struct timespec *wtom, struct timespec *sleep)
1307 seq = read_seqbegin(&timekeeper.lock);
1308 *xtim = tk_xtime(&timekeeper);
1309 *wtom = timekeeper.wall_to_monotonic;
1310 *sleep = timekeeper.total_sleep_time;
1311 } while (read_seqretry(&timekeeper.lock, seq));
1314 #ifdef CONFIG_HIGH_RES_TIMERS
1316 * ktime_get_update_offsets - hrtimer helper
1317 * @offs_real: pointer to storage for monotonic -> realtime offset
1318 * @offs_boot: pointer to storage for monotonic -> boottime offset
1320 * Returns current monotonic time and updates the offsets
1321 * Called from hrtimer_interupt() or retrigger_next_event()
1323 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1330 seq = read_seqbegin(&timekeeper.lock);
1332 secs = timekeeper.xtime_sec;
1333 nsecs = timekeeping_get_ns();
1335 *offs_real = timekeeper.offs_real;
1336 *offs_boot = timekeeper.offs_boot;
1337 } while (read_seqretry(&timekeeper.lock, seq));
1339 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1340 now = ktime_sub(now, *offs_real);
1346 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1348 ktime_t ktime_get_monotonic_offset(void)
1351 struct timespec wtom;
1354 seq = read_seqbegin(&timekeeper.lock);
1355 wtom = timekeeper.wall_to_monotonic;
1356 } while (read_seqretry(&timekeeper.lock, seq));
1358 return timespec_to_ktime(wtom);
1360 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1364 * xtime_update() - advances the timekeeping infrastructure
1365 * @ticks: number of ticks, that have elapsed since the last call.
1367 * Must be called with interrupts disabled.
1369 void xtime_update(unsigned long ticks)
1371 write_seqlock(&xtime_lock);
1373 write_sequnlock(&xtime_lock);