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 = (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;
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 tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
133 u64 tmp, ntpinterval;
134 struct clocksource *old_clock;
136 old_clock = tk->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 tk->cycle_interval = interval;
152 /* Go back from cycles -> shifted ns */
153 tk->xtime_interval = (u64) interval * clock->mult;
154 tk->xtime_remainder = ntpinterval - tk->xtime_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 tk->xtime_nsec >>= -shift_change;
164 tk->xtime_nsec <<= shift_change;
166 tk->shift = clock->shift;
169 tk->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 tk->mult = clock->mult;
179 /* Timekeeper helper functions. */
180 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
182 cycle_t cycle_now, cycle_delta;
183 struct clocksource *clock;
186 /* read clocksource: */
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 * tk->mult + tk->xtime_nsec;
196 /* If arch requires, add in gettimeoffset() */
197 return nsec + arch_gettimeoffset();
200 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
202 cycle_t cycle_now, cycle_delta;
203 struct clocksource *clock;
206 /* read clocksource: */
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(struct timekeeper *tk)
222 struct timespec tmp, *wtm = &tk->wall_to_monotonic;
224 set_normalized_timespec(&tmp, -wtm->tv_sec, -wtm->tv_nsec);
225 tk->offs_real = timespec_to_ktime(tmp);
228 /* must hold write on timekeeper.lock */
229 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
237 update_rt_offset(tk);
239 update_vsyscall(&xt, &tk->wall_to_monotonic, tk->clock, tk->mult);
243 * timekeeping_forward_now - update clock to the current time
245 * Forward the current clock to update its state since the last call to
246 * update_wall_time(). This is useful before significant clock changes,
247 * as it avoids having to deal with this time offset explicitly.
249 static void timekeeping_forward_now(struct timekeeper *tk)
251 cycle_t cycle_now, cycle_delta;
252 struct clocksource *clock;
256 cycle_now = clock->read(clock);
257 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
258 clock->cycle_last = cycle_now;
260 tk->xtime_nsec += cycle_delta * tk->mult;
262 /* If arch requires, add in gettimeoffset() */
263 tk->xtime_nsec += arch_gettimeoffset() << tk->shift;
265 tk_normalize_xtime(tk);
267 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
268 timespec_add_ns(&tk->raw_time, nsec);
272 * getnstimeofday - Returns the time of day in a timespec
273 * @ts: pointer to the timespec to be set
275 * Returns the time of day in a timespec.
277 void getnstimeofday(struct timespec *ts)
282 WARN_ON(timekeeping_suspended);
285 seq = read_seqbegin(&timekeeper.lock);
287 ts->tv_sec = timekeeper.xtime_sec;
288 ts->tv_nsec = timekeeping_get_ns(&timekeeper);
290 } while (read_seqretry(&timekeeper.lock, seq));
292 timespec_add_ns(ts, nsecs);
294 EXPORT_SYMBOL(getnstimeofday);
296 ktime_t ktime_get(void)
301 WARN_ON(timekeeping_suspended);
304 seq = read_seqbegin(&timekeeper.lock);
305 secs = timekeeper.xtime_sec +
306 timekeeper.wall_to_monotonic.tv_sec;
307 nsecs = timekeeping_get_ns(&timekeeper) +
308 timekeeper.wall_to_monotonic.tv_nsec;
310 } while (read_seqretry(&timekeeper.lock, seq));
312 * Use ktime_set/ktime_add_ns to create a proper ktime on
313 * 32-bit architectures without CONFIG_KTIME_SCALAR.
315 return ktime_add_ns(ktime_set(secs, 0), nsecs);
317 EXPORT_SYMBOL_GPL(ktime_get);
320 * ktime_get_ts - get the monotonic clock in timespec format
321 * @ts: pointer to timespec variable
323 * The function calculates the monotonic clock from the realtime
324 * clock and the wall_to_monotonic offset and stores the result
325 * in normalized timespec format in the variable pointed to by @ts.
327 void ktime_get_ts(struct timespec *ts)
329 struct timespec tomono;
332 WARN_ON(timekeeping_suspended);
335 seq = read_seqbegin(&timekeeper.lock);
336 ts->tv_sec = timekeeper.xtime_sec;
337 ts->tv_nsec = timekeeping_get_ns(&timekeeper);
338 tomono = timekeeper.wall_to_monotonic;
340 } while (read_seqretry(&timekeeper.lock, seq));
342 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
343 ts->tv_nsec + tomono.tv_nsec);
345 EXPORT_SYMBOL_GPL(ktime_get_ts);
347 #ifdef CONFIG_NTP_PPS
350 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
351 * @ts_raw: pointer to the timespec to be set to raw monotonic time
352 * @ts_real: pointer to the timespec to be set to the time of day
354 * This function reads both the time of day and raw monotonic time at the
355 * same time atomically and stores the resulting timestamps in timespec
358 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
361 s64 nsecs_raw, nsecs_real;
363 WARN_ON_ONCE(timekeeping_suspended);
366 seq = read_seqbegin(&timekeeper.lock);
368 *ts_raw = timekeeper.raw_time;
369 ts_real->tv_sec = timekeeper.xtime_sec;
370 ts_real->tv_nsec = 0;
372 nsecs_raw = timekeeping_get_ns_raw(&timekeeper);
373 nsecs_real = timekeeping_get_ns(&timekeeper);
375 } while (read_seqretry(&timekeeper.lock, seq));
377 timespec_add_ns(ts_raw, nsecs_raw);
378 timespec_add_ns(ts_real, nsecs_real);
380 EXPORT_SYMBOL(getnstime_raw_and_real);
382 #endif /* CONFIG_NTP_PPS */
385 * do_gettimeofday - Returns the time of day in a timeval
386 * @tv: pointer to the timeval to be set
388 * NOTE: Users should be converted to using getnstimeofday()
390 void do_gettimeofday(struct timeval *tv)
394 getnstimeofday(&now);
395 tv->tv_sec = now.tv_sec;
396 tv->tv_usec = now.tv_nsec/1000;
398 EXPORT_SYMBOL(do_gettimeofday);
401 * do_settimeofday - Sets the time of day
402 * @tv: pointer to the timespec variable containing the new time
404 * Sets the time of day to the new time and update NTP and notify hrtimers
406 int do_settimeofday(const struct timespec *tv)
408 struct timespec ts_delta, xt;
411 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
414 write_seqlock_irqsave(&timekeeper.lock, flags);
416 timekeeping_forward_now(&timekeeper);
418 xt = tk_xtime(&timekeeper);
419 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
420 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
422 timekeeper.wall_to_monotonic =
423 timespec_sub(timekeeper.wall_to_monotonic, ts_delta);
425 tk_set_xtime(&timekeeper, tv);
427 timekeeping_update(&timekeeper, true);
429 write_sequnlock_irqrestore(&timekeeper.lock, flags);
431 /* signal hrtimers about time change */
436 EXPORT_SYMBOL(do_settimeofday);
439 * timekeeping_inject_offset - Adds or subtracts from the current time.
440 * @tv: pointer to the timespec variable containing the offset
442 * Adds or subtracts an offset value from the current time.
444 int timekeeping_inject_offset(struct timespec *ts)
448 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
451 write_seqlock_irqsave(&timekeeper.lock, flags);
453 timekeeping_forward_now(&timekeeper);
456 tk_xtime_add(&timekeeper, ts);
457 timekeeper.wall_to_monotonic =
458 timespec_sub(timekeeper.wall_to_monotonic, *ts);
460 timekeeping_update(&timekeeper, true);
462 write_sequnlock_irqrestore(&timekeeper.lock, flags);
464 /* signal hrtimers about time change */
469 EXPORT_SYMBOL(timekeeping_inject_offset);
472 * change_clocksource - Swaps clocksources if a new one is available
474 * Accumulates current time interval and initializes new clocksource
476 static int change_clocksource(void *data)
478 struct clocksource *new, *old;
481 new = (struct clocksource *) data;
483 write_seqlock_irqsave(&timekeeper.lock, flags);
485 timekeeping_forward_now(&timekeeper);
486 if (!new->enable || new->enable(new) == 0) {
487 old = timekeeper.clock;
488 tk_setup_internals(&timekeeper, new);
492 timekeeping_update(&timekeeper, true);
494 write_sequnlock_irqrestore(&timekeeper.lock, flags);
500 * timekeeping_notify - Install a new clock source
501 * @clock: pointer to the clock source
503 * This function is called from clocksource.c after a new, better clock
504 * source has been registered. The caller holds the clocksource_mutex.
506 void timekeeping_notify(struct clocksource *clock)
508 if (timekeeper.clock == clock)
510 stop_machine(change_clocksource, clock, NULL);
515 * ktime_get_real - get the real (wall-) time in ktime_t format
517 * returns the time in ktime_t format
519 ktime_t ktime_get_real(void)
523 getnstimeofday(&now);
525 return timespec_to_ktime(now);
527 EXPORT_SYMBOL_GPL(ktime_get_real);
530 * getrawmonotonic - Returns the raw monotonic time in a timespec
531 * @ts: pointer to the timespec to be set
533 * Returns the raw monotonic time (completely un-modified by ntp)
535 void getrawmonotonic(struct timespec *ts)
541 seq = read_seqbegin(&timekeeper.lock);
542 nsecs = timekeeping_get_ns_raw(&timekeeper);
543 *ts = timekeeper.raw_time;
545 } while (read_seqretry(&timekeeper.lock, seq));
547 timespec_add_ns(ts, nsecs);
549 EXPORT_SYMBOL(getrawmonotonic);
552 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
554 int timekeeping_valid_for_hres(void)
560 seq = read_seqbegin(&timekeeper.lock);
562 ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
564 } while (read_seqretry(&timekeeper.lock, seq));
570 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
572 u64 timekeeping_max_deferment(void)
578 seq = read_seqbegin(&timekeeper.lock);
580 ret = timekeeper.clock->max_idle_ns;
582 } while (read_seqretry(&timekeeper.lock, seq));
588 * read_persistent_clock - Return time from the persistent clock.
590 * Weak dummy function for arches that do not yet support it.
591 * Reads the time from the battery backed persistent clock.
592 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
594 * XXX - Do be sure to remove it once all arches implement it.
596 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
603 * read_boot_clock - Return time of the system start.
605 * Weak dummy function for arches that do not yet support it.
606 * Function to read the exact time the system has been started.
607 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
609 * XXX - Do be sure to remove it once all arches implement it.
611 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
618 * timekeeping_init - Initializes the clocksource and common timekeeping values
620 void __init timekeeping_init(void)
622 struct clocksource *clock;
624 struct timespec now, boot;
626 read_persistent_clock(&now);
627 read_boot_clock(&boot);
629 seqlock_init(&timekeeper.lock);
633 write_seqlock_irqsave(&timekeeper.lock, flags);
634 clock = clocksource_default_clock();
636 clock->enable(clock);
637 tk_setup_internals(&timekeeper, clock);
639 tk_set_xtime(&timekeeper, &now);
640 timekeeper.raw_time.tv_sec = 0;
641 timekeeper.raw_time.tv_nsec = 0;
642 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
643 boot = tk_xtime(&timekeeper);
645 set_normalized_timespec(&timekeeper.wall_to_monotonic,
646 -boot.tv_sec, -boot.tv_nsec);
647 update_rt_offset(&timekeeper);
648 timekeeper.total_sleep_time.tv_sec = 0;
649 timekeeper.total_sleep_time.tv_nsec = 0;
650 write_sequnlock_irqrestore(&timekeeper.lock, flags);
653 /* time in seconds when suspend began */
654 static struct timespec timekeeping_suspend_time;
656 static void update_sleep_time(struct timespec t)
658 timekeeper.total_sleep_time = t;
659 timekeeper.offs_boot = timespec_to_ktime(t);
663 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
664 * @delta: pointer to a timespec delta value
666 * Takes a timespec offset measuring a suspend interval and properly
667 * adds the sleep offset to the timekeeping variables.
669 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
670 struct timespec *delta)
672 if (!timespec_valid(delta)) {
673 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
674 "sleep delta value!\n");
678 tk_xtime_add(tk, delta);
679 tk->wall_to_monotonic = timespec_sub(tk->wall_to_monotonic, *delta);
680 update_sleep_time(timespec_add(tk->total_sleep_time, *delta));
684 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
685 * @delta: pointer to a timespec delta value
687 * This hook is for architectures that cannot support read_persistent_clock
688 * because their RTC/persistent clock is only accessible when irqs are enabled.
690 * This function should only be called by rtc_resume(), and allows
691 * a suspend offset to be injected into the timekeeping values.
693 void timekeeping_inject_sleeptime(struct timespec *delta)
698 /* Make sure we don't set the clock twice */
699 read_persistent_clock(&ts);
700 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
703 write_seqlock_irqsave(&timekeeper.lock, flags);
705 timekeeping_forward_now(&timekeeper);
707 __timekeeping_inject_sleeptime(&timekeeper, delta);
709 timekeeping_update(&timekeeper, true);
711 write_sequnlock_irqrestore(&timekeeper.lock, flags);
713 /* signal hrtimers about time change */
718 * timekeeping_resume - Resumes the generic timekeeping subsystem.
720 * This is for the generic clocksource timekeeping.
721 * xtime/wall_to_monotonic/jiffies/etc are
722 * still managed by arch specific suspend/resume code.
724 static void timekeeping_resume(void)
729 read_persistent_clock(&ts);
731 clocksource_resume();
733 write_seqlock_irqsave(&timekeeper.lock, flags);
735 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
736 ts = timespec_sub(ts, timekeeping_suspend_time);
737 __timekeeping_inject_sleeptime(&timekeeper, &ts);
739 /* re-base the last cycle value */
740 timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock);
741 timekeeper.ntp_error = 0;
742 timekeeping_suspended = 0;
743 timekeeping_update(&timekeeper, false);
744 write_sequnlock_irqrestore(&timekeeper.lock, flags);
746 touch_softlockup_watchdog();
748 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
750 /* Resume hrtimers */
754 static int timekeeping_suspend(void)
757 struct timespec delta, delta_delta;
758 static struct timespec old_delta;
760 read_persistent_clock(&timekeeping_suspend_time);
762 write_seqlock_irqsave(&timekeeper.lock, flags);
763 timekeeping_forward_now(&timekeeper);
764 timekeeping_suspended = 1;
767 * To avoid drift caused by repeated suspend/resumes,
768 * which each can add ~1 second drift error,
769 * try to compensate so the difference in system time
770 * and persistent_clock time stays close to constant.
772 delta = timespec_sub(tk_xtime(&timekeeper), timekeeping_suspend_time);
773 delta_delta = timespec_sub(delta, old_delta);
774 if (abs(delta_delta.tv_sec) >= 2) {
776 * if delta_delta is too large, assume time correction
777 * has occured and set old_delta to the current delta.
781 /* Otherwise try to adjust old_system to compensate */
782 timekeeping_suspend_time =
783 timespec_add(timekeeping_suspend_time, delta_delta);
785 write_sequnlock_irqrestore(&timekeeper.lock, flags);
787 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
788 clocksource_suspend();
793 /* sysfs resume/suspend bits for timekeeping */
794 static struct syscore_ops timekeeping_syscore_ops = {
795 .resume = timekeeping_resume,
796 .suspend = timekeeping_suspend,
799 static int __init timekeeping_init_ops(void)
801 register_syscore_ops(&timekeeping_syscore_ops);
805 device_initcall(timekeeping_init_ops);
808 * If the error is already larger, we look ahead even further
809 * to compensate for late or lost adjustments.
811 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
812 s64 error, s64 *interval,
820 * Use the current error value to determine how much to look ahead.
821 * The larger the error the slower we adjust for it to avoid problems
822 * with losing too many ticks, otherwise we would overadjust and
823 * produce an even larger error. The smaller the adjustment the
824 * faster we try to adjust for it, as lost ticks can do less harm
825 * here. This is tuned so that an error of about 1 msec is adjusted
826 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
828 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
829 error2 = abs(error2);
830 for (look_ahead = 0; error2 > 0; look_ahead++)
834 * Now calculate the error in (1 << look_ahead) ticks, but first
835 * remove the single look ahead already included in the error.
837 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
838 tick_error -= tk->xtime_interval >> 1;
839 error = ((error - tick_error) >> look_ahead) + tick_error;
841 /* Finally calculate the adjustment shift value. */
846 *interval = -*interval;
850 for (adj = 0; error > i; adj++)
859 * Adjust the multiplier to reduce the error value,
860 * this is optimized for the most common adjustments of -1,0,1,
861 * for other values we can do a bit more work.
863 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
865 s64 error, interval = tk->cycle_interval;
869 * The point of this is to check if the error is greater than half
872 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
874 * Note we subtract one in the shift, so that error is really error*2.
875 * This "saves" dividing(shifting) interval twice, but keeps the
876 * (error > interval) comparison as still measuring if error is
877 * larger than half an interval.
879 * Note: It does not "save" on aggravation when reading the code.
881 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
882 if (error > interval) {
884 * We now divide error by 4(via shift), which checks if
885 * the error is greater than twice the interval.
886 * If it is greater, we need a bigadjust, if its smaller,
887 * we can adjust by 1.
891 * XXX - In update_wall_time, we round up to the next
892 * nanosecond, and store the amount rounded up into
893 * the error. This causes the likely below to be unlikely.
895 * The proper fix is to avoid rounding up by using
896 * the high precision timekeeper.xtime_nsec instead of
897 * xtime.tv_nsec everywhere. Fixing this will take some
900 if (likely(error <= interval))
903 adj = timekeeping_bigadjust(tk, error, &interval,
905 } else if (error < -interval) {
906 /* See comment above, this is just switched for the negative */
908 if (likely(error >= -interval)) {
910 interval = -interval;
913 adj = timekeeping_bigadjust(tk, error, &interval,
918 if (unlikely(tk->clock->maxadj &&
919 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
920 printk_once(KERN_WARNING
921 "Adjusting %s more than 11%% (%ld vs %ld)\n",
922 tk->clock->name, (long)tk->mult + adj,
923 (long)tk->clock->mult + tk->clock->maxadj);
926 * So the following can be confusing.
928 * To keep things simple, lets assume adj == 1 for now.
930 * When adj != 1, remember that the interval and offset values
931 * have been appropriately scaled so the math is the same.
933 * The basic idea here is that we're increasing the multiplier
934 * by one, this causes the xtime_interval to be incremented by
935 * one cycle_interval. This is because:
936 * xtime_interval = cycle_interval * mult
937 * So if mult is being incremented by one:
938 * xtime_interval = cycle_interval * (mult + 1)
940 * xtime_interval = (cycle_interval * mult) + cycle_interval
941 * Which can be shortened to:
942 * xtime_interval += cycle_interval
944 * So offset stores the non-accumulated cycles. Thus the current
945 * time (in shifted nanoseconds) is:
946 * now = (offset * adj) + xtime_nsec
947 * Now, even though we're adjusting the clock frequency, we have
948 * to keep time consistent. In other words, we can't jump back
949 * in time, and we also want to avoid jumping forward in time.
951 * So given the same offset value, we need the time to be the same
952 * both before and after the freq adjustment.
953 * now = (offset * adj_1) + xtime_nsec_1
954 * now = (offset * adj_2) + xtime_nsec_2
956 * (offset * adj_1) + xtime_nsec_1 =
957 * (offset * adj_2) + xtime_nsec_2
961 * (offset * adj_1) + xtime_nsec_1 =
962 * (offset * (adj_1+1)) + xtime_nsec_2
963 * (offset * adj_1) + xtime_nsec_1 =
964 * (offset * adj_1) + offset + xtime_nsec_2
965 * Canceling the sides:
966 * xtime_nsec_1 = offset + xtime_nsec_2
968 * xtime_nsec_2 = xtime_nsec_1 - offset
969 * Which simplfies to:
970 * xtime_nsec -= offset
972 * XXX - TODO: Doc ntp_error calculation.
975 tk->xtime_interval += interval;
976 tk->xtime_nsec -= offset;
977 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
980 * It may be possible that when we entered this function, xtime_nsec
981 * was very small. Further, if we're slightly speeding the clocksource
982 * in the code above, its possible the required corrective factor to
983 * xtime_nsec could cause it to underflow.
985 * Now, since we already accumulated the second, cannot simply roll
986 * the accumulated second back, since the NTP subsystem has been
987 * notified via second_overflow. So instead we push xtime_nsec forward
988 * by the amount we underflowed, and add that amount into the error.
990 * We'll correct this error next time through this function, when
991 * xtime_nsec is not as small.
993 if (unlikely((s64)tk->xtime_nsec < 0)) {
994 s64 neg = -(s64)tk->xtime_nsec;
996 tk->ntp_error += neg << tk->ntp_error_shift;
1002 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1004 * Helper function that accumulates a the nsecs greater then a second
1005 * from the xtime_nsec field to the xtime_secs field.
1006 * It also calls into the NTP code to handle leapsecond processing.
1009 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1011 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1013 while (tk->xtime_nsec >= nsecps) {
1016 tk->xtime_nsec -= nsecps;
1019 /* Figure out if its a leap sec and apply if needed */
1020 leap = second_overflow(tk->xtime_sec);
1021 tk->xtime_sec += leap;
1022 tk->wall_to_monotonic.tv_sec -= leap;
1024 clock_was_set_delayed();
1030 * logarithmic_accumulation - shifted accumulation of cycles
1032 * This functions accumulates a shifted interval of cycles into
1033 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1036 * Returns the unconsumed cycles.
1038 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1043 /* If the offset is smaller then a shifted interval, do nothing */
1044 if (offset < tk->cycle_interval<<shift)
1047 /* Accumulate one shifted interval */
1048 offset -= tk->cycle_interval << shift;
1049 tk->clock->cycle_last += tk->cycle_interval << shift;
1051 tk->xtime_nsec += tk->xtime_interval << shift;
1052 accumulate_nsecs_to_secs(tk);
1054 /* Accumulate raw time */
1055 raw_nsecs = tk->raw_interval << shift;
1056 raw_nsecs += tk->raw_time.tv_nsec;
1057 if (raw_nsecs >= NSEC_PER_SEC) {
1058 u64 raw_secs = raw_nsecs;
1059 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1060 tk->raw_time.tv_sec += raw_secs;
1062 tk->raw_time.tv_nsec = raw_nsecs;
1064 /* Accumulate error between NTP and clock interval */
1065 tk->ntp_error += ntp_tick_length() << shift;
1066 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1067 (tk->ntp_error_shift + shift);
1073 * update_wall_time - Uses the current clocksource to increment the wall time
1076 static void update_wall_time(void)
1078 struct clocksource *clock;
1080 int shift = 0, maxshift;
1081 unsigned long flags;
1084 write_seqlock_irqsave(&timekeeper.lock, flags);
1086 /* Make sure we're fully resumed: */
1087 if (unlikely(timekeeping_suspended))
1090 clock = timekeeper.clock;
1092 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1093 offset = timekeeper.cycle_interval;
1095 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1099 * With NO_HZ we may have to accumulate many cycle_intervals
1100 * (think "ticks") worth of time at once. To do this efficiently,
1101 * we calculate the largest doubling multiple of cycle_intervals
1102 * that is smaller than the offset. We then accumulate that
1103 * chunk in one go, and then try to consume the next smaller
1106 shift = ilog2(offset) - ilog2(timekeeper.cycle_interval);
1107 shift = max(0, shift);
1108 /* Bound shift to one less than what overflows tick_length */
1109 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1110 shift = min(shift, maxshift);
1111 while (offset >= timekeeper.cycle_interval) {
1112 offset = logarithmic_accumulation(&timekeeper, offset, shift);
1113 if(offset < timekeeper.cycle_interval<<shift)
1117 /* correct the clock when NTP error is too big */
1118 timekeeping_adjust(&timekeeper, offset);
1122 * Store only full nanoseconds into xtime_nsec after rounding
1123 * it up and add the remainder to the error difference.
1124 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1125 * by truncating the remainder in vsyscalls. However, it causes
1126 * additional work to be done in timekeeping_adjust(). Once
1127 * the vsyscall implementations are converted to use xtime_nsec
1128 * (shifted nanoseconds), this can be killed.
1130 remainder = timekeeper.xtime_nsec & ((1 << timekeeper.shift) - 1);
1131 timekeeper.xtime_nsec -= remainder;
1132 timekeeper.xtime_nsec += 1 << timekeeper.shift;
1133 timekeeper.ntp_error += remainder << timekeeper.ntp_error_shift;
1136 * Finally, make sure that after the rounding
1137 * xtime_nsec isn't larger than NSEC_PER_SEC
1139 accumulate_nsecs_to_secs(&timekeeper);
1141 timekeeping_update(&timekeeper, false);
1144 write_sequnlock_irqrestore(&timekeeper.lock, flags);
1149 * getboottime - Return the real time of system boot.
1150 * @ts: pointer to the timespec to be set
1152 * Returns the wall-time of boot in a timespec.
1154 * This is based on the wall_to_monotonic offset and the total suspend
1155 * time. Calls to settimeofday will affect the value returned (which
1156 * basically means that however wrong your real time clock is at boot time,
1157 * you get the right time here).
1159 void getboottime(struct timespec *ts)
1161 struct timespec boottime = {
1162 .tv_sec = timekeeper.wall_to_monotonic.tv_sec +
1163 timekeeper.total_sleep_time.tv_sec,
1164 .tv_nsec = timekeeper.wall_to_monotonic.tv_nsec +
1165 timekeeper.total_sleep_time.tv_nsec
1168 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1170 EXPORT_SYMBOL_GPL(getboottime);
1173 * get_monotonic_boottime - Returns monotonic time since boot
1174 * @ts: pointer to the timespec to be set
1176 * Returns the monotonic time since boot in a timespec.
1178 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1179 * includes the time spent in suspend.
1181 void get_monotonic_boottime(struct timespec *ts)
1183 struct timespec tomono, sleep;
1186 WARN_ON(timekeeping_suspended);
1189 seq = read_seqbegin(&timekeeper.lock);
1190 ts->tv_sec = timekeeper.xtime_sec;
1191 ts->tv_nsec = timekeeping_get_ns(&timekeeper);
1192 tomono = timekeeper.wall_to_monotonic;
1193 sleep = timekeeper.total_sleep_time;
1195 } while (read_seqretry(&timekeeper.lock, seq));
1197 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec,
1198 ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec);
1200 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1203 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1205 * Returns the monotonic time since boot in a ktime
1207 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1208 * includes the time spent in suspend.
1210 ktime_t ktime_get_boottime(void)
1214 get_monotonic_boottime(&ts);
1215 return timespec_to_ktime(ts);
1217 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1220 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1221 * @ts: pointer to the timespec to be converted
1223 void monotonic_to_bootbased(struct timespec *ts)
1225 *ts = timespec_add(*ts, timekeeper.total_sleep_time);
1227 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1229 unsigned long get_seconds(void)
1231 return timekeeper.xtime_sec;
1233 EXPORT_SYMBOL(get_seconds);
1235 struct timespec __current_kernel_time(void)
1237 return tk_xtime(&timekeeper);
1240 struct timespec current_kernel_time(void)
1242 struct timespec now;
1246 seq = read_seqbegin(&timekeeper.lock);
1248 now = tk_xtime(&timekeeper);
1249 } while (read_seqretry(&timekeeper.lock, seq));
1253 EXPORT_SYMBOL(current_kernel_time);
1255 struct timespec get_monotonic_coarse(void)
1257 struct timespec now, mono;
1261 seq = read_seqbegin(&timekeeper.lock);
1263 now = tk_xtime(&timekeeper);
1264 mono = timekeeper.wall_to_monotonic;
1265 } while (read_seqretry(&timekeeper.lock, seq));
1267 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1268 now.tv_nsec + mono.tv_nsec);
1273 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1274 * without sampling the sequence number in xtime_lock.
1275 * jiffies is defined in the linker script...
1277 void do_timer(unsigned long ticks)
1279 jiffies_64 += ticks;
1281 calc_global_load(ticks);
1285 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1286 * and sleep offsets.
1287 * @xtim: pointer to timespec to be set with xtime
1288 * @wtom: pointer to timespec to be set with wall_to_monotonic
1289 * @sleep: pointer to timespec to be set with time in suspend
1291 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1292 struct timespec *wtom, struct timespec *sleep)
1297 seq = read_seqbegin(&timekeeper.lock);
1298 *xtim = tk_xtime(&timekeeper);
1299 *wtom = timekeeper.wall_to_monotonic;
1300 *sleep = timekeeper.total_sleep_time;
1301 } while (read_seqretry(&timekeeper.lock, seq));
1304 #ifdef CONFIG_HIGH_RES_TIMERS
1306 * ktime_get_update_offsets - hrtimer helper
1307 * @offs_real: pointer to storage for monotonic -> realtime offset
1308 * @offs_boot: pointer to storage for monotonic -> boottime offset
1310 * Returns current monotonic time and updates the offsets
1311 * Called from hrtimer_interupt() or retrigger_next_event()
1313 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1320 seq = read_seqbegin(&timekeeper.lock);
1322 secs = timekeeper.xtime_sec;
1323 nsecs = timekeeping_get_ns(&timekeeper);
1325 *offs_real = timekeeper.offs_real;
1326 *offs_boot = timekeeper.offs_boot;
1327 } while (read_seqretry(&timekeeper.lock, seq));
1329 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1330 now = ktime_sub(now, *offs_real);
1336 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1338 ktime_t ktime_get_monotonic_offset(void)
1341 struct timespec wtom;
1344 seq = read_seqbegin(&timekeeper.lock);
1345 wtom = timekeeper.wall_to_monotonic;
1346 } while (read_seqretry(&timekeeper.lock, seq));
1348 return timespec_to_ktime(wtom);
1350 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1353 * xtime_update() - advances the timekeeping infrastructure
1354 * @ticks: number of ticks, that have elapsed since the last call.
1356 * Must be called with interrupts disabled.
1358 void xtime_update(unsigned long ticks)
1360 write_seqlock(&xtime_lock);
1362 write_sequnlock(&xtime_lock);