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/timekeeper_internal.h>
12 #include <linux/module.h>
13 #include <linux/interrupt.h>
14 #include <linux/percpu.h>
15 #include <linux/init.h>
17 #include <linux/sched.h>
18 #include <linux/syscore_ops.h>
19 #include <linux/clocksource.h>
20 #include <linux/jiffies.h>
21 #include <linux/time.h>
22 #include <linux/tick.h>
23 #include <linux/stop_machine.h>
26 static struct timekeeper timekeeper;
29 * This read-write spinlock protects us from races in SMP while
32 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
34 /* flag for if timekeeping is suspended */
35 int __read_mostly timekeeping_suspended;
37 static inline void tk_normalize_xtime(struct timekeeper *tk)
39 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
40 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
45 static struct timespec tk_xtime(struct timekeeper *tk)
49 ts.tv_sec = tk->xtime_sec;
50 ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift);
54 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
56 tk->xtime_sec = ts->tv_sec;
57 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
60 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
62 tk->xtime_sec += ts->tv_sec;
63 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
64 tk_normalize_xtime(tk);
67 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
72 * Verify consistency of: offset_real = -wall_to_monotonic
73 * before modifying anything
75 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
76 -tk->wall_to_monotonic.tv_nsec);
77 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
78 tk->wall_to_monotonic = wtm;
79 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
80 tk->offs_real = timespec_to_ktime(tmp);
83 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
85 /* Verify consistency before modifying */
86 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
88 tk->total_sleep_time = t;
89 tk->offs_boot = timespec_to_ktime(t);
93 * timekeeper_setup_internals - Set up internals to use clocksource clock.
95 * @clock: Pointer to clocksource.
97 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
98 * pair and interval request.
100 * Unless you're the timekeeping code, you should not be using this!
102 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
105 u64 tmp, ntpinterval;
106 struct clocksource *old_clock;
108 old_clock = tk->clock;
110 clock->cycle_last = clock->read(clock);
112 /* Do the ns -> cycle conversion first, using original mult */
113 tmp = NTP_INTERVAL_LENGTH;
114 tmp <<= clock->shift;
116 tmp += clock->mult/2;
117 do_div(tmp, clock->mult);
121 interval = (cycle_t) tmp;
122 tk->cycle_interval = interval;
124 /* Go back from cycles -> shifted ns */
125 tk->xtime_interval = (u64) interval * clock->mult;
126 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
128 ((u64) interval * clock->mult) >> clock->shift;
130 /* if changing clocks, convert xtime_nsec shift units */
132 int shift_change = clock->shift - old_clock->shift;
133 if (shift_change < 0)
134 tk->xtime_nsec >>= -shift_change;
136 tk->xtime_nsec <<= shift_change;
138 tk->shift = clock->shift;
141 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
144 * The timekeeper keeps its own mult values for the currently
145 * active clocksource. These value will be adjusted via NTP
146 * to counteract clock drifting.
148 tk->mult = clock->mult;
151 /* Timekeeper helper functions. */
152 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
154 cycle_t cycle_now, cycle_delta;
155 struct clocksource *clock;
158 /* read clocksource: */
160 cycle_now = clock->read(clock);
162 /* calculate the delta since the last update_wall_time: */
163 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
165 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
168 /* If arch requires, add in gettimeoffset() */
169 return nsec + arch_gettimeoffset();
172 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
174 cycle_t cycle_now, cycle_delta;
175 struct clocksource *clock;
178 /* read clocksource: */
180 cycle_now = clock->read(clock);
182 /* calculate the delta since the last update_wall_time: */
183 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
185 /* convert delta to nanoseconds. */
186 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
188 /* If arch requires, add in gettimeoffset() */
189 return nsec + arch_gettimeoffset();
192 /* must hold write on timekeeper.lock */
193 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
202 update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->clock, tk->mult);
206 * timekeeping_forward_now - update clock to the current time
208 * Forward the current clock to update its state since the last call to
209 * update_wall_time(). This is useful before significant clock changes,
210 * as it avoids having to deal with this time offset explicitly.
212 static void timekeeping_forward_now(struct timekeeper *tk)
214 cycle_t cycle_now, cycle_delta;
215 struct clocksource *clock;
219 cycle_now = clock->read(clock);
220 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
221 clock->cycle_last = cycle_now;
223 tk->xtime_nsec += cycle_delta * tk->mult;
225 /* If arch requires, add in gettimeoffset() */
226 tk->xtime_nsec += (u64)arch_gettimeoffset() << tk->shift;
228 tk_normalize_xtime(tk);
230 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
231 timespec_add_ns(&tk->raw_time, nsec);
235 * getnstimeofday - Returns the time of day in a timespec
236 * @ts: pointer to the timespec to be set
238 * Returns the time of day in a timespec.
240 void getnstimeofday(struct timespec *ts)
242 struct timekeeper *tk = &timekeeper;
246 WARN_ON(timekeeping_suspended);
249 seq = read_seqbegin(&tk->lock);
251 ts->tv_sec = tk->xtime_sec;
252 nsecs = timekeeping_get_ns(tk);
254 } while (read_seqretry(&tk->lock, seq));
257 timespec_add_ns(ts, nsecs);
259 EXPORT_SYMBOL(getnstimeofday);
261 ktime_t ktime_get(void)
263 struct timekeeper *tk = &timekeeper;
267 WARN_ON(timekeeping_suspended);
270 seq = read_seqbegin(&tk->lock);
271 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
272 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
274 } while (read_seqretry(&tk->lock, seq));
276 * Use ktime_set/ktime_add_ns to create a proper ktime on
277 * 32-bit architectures without CONFIG_KTIME_SCALAR.
279 return ktime_add_ns(ktime_set(secs, 0), nsecs);
281 EXPORT_SYMBOL_GPL(ktime_get);
284 * ktime_get_ts - get the monotonic clock in timespec format
285 * @ts: pointer to timespec variable
287 * The function calculates the monotonic clock from the realtime
288 * clock and the wall_to_monotonic offset and stores the result
289 * in normalized timespec format in the variable pointed to by @ts.
291 void ktime_get_ts(struct timespec *ts)
293 struct timekeeper *tk = &timekeeper;
294 struct timespec tomono;
298 WARN_ON(timekeeping_suspended);
301 seq = read_seqbegin(&tk->lock);
302 ts->tv_sec = tk->xtime_sec;
303 nsec = timekeeping_get_ns(tk);
304 tomono = tk->wall_to_monotonic;
306 } while (read_seqretry(&tk->lock, seq));
308 ts->tv_sec += tomono.tv_sec;
310 timespec_add_ns(ts, nsec + tomono.tv_nsec);
312 EXPORT_SYMBOL_GPL(ktime_get_ts);
314 #ifdef CONFIG_NTP_PPS
317 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
318 * @ts_raw: pointer to the timespec to be set to raw monotonic time
319 * @ts_real: pointer to the timespec to be set to the time of day
321 * This function reads both the time of day and raw monotonic time at the
322 * same time atomically and stores the resulting timestamps in timespec
325 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
327 struct timekeeper *tk = &timekeeper;
329 s64 nsecs_raw, nsecs_real;
331 WARN_ON_ONCE(timekeeping_suspended);
334 seq = read_seqbegin(&tk->lock);
336 *ts_raw = tk->raw_time;
337 ts_real->tv_sec = tk->xtime_sec;
338 ts_real->tv_nsec = 0;
340 nsecs_raw = timekeeping_get_ns_raw(tk);
341 nsecs_real = timekeeping_get_ns(tk);
343 } while (read_seqretry(&tk->lock, seq));
345 timespec_add_ns(ts_raw, nsecs_raw);
346 timespec_add_ns(ts_real, nsecs_real);
348 EXPORT_SYMBOL(getnstime_raw_and_real);
350 #endif /* CONFIG_NTP_PPS */
353 * do_gettimeofday - Returns the time of day in a timeval
354 * @tv: pointer to the timeval to be set
356 * NOTE: Users should be converted to using getnstimeofday()
358 void do_gettimeofday(struct timeval *tv)
362 getnstimeofday(&now);
363 tv->tv_sec = now.tv_sec;
364 tv->tv_usec = now.tv_nsec/1000;
366 EXPORT_SYMBOL(do_gettimeofday);
369 * do_settimeofday - Sets the time of day
370 * @tv: pointer to the timespec variable containing the new time
372 * Sets the time of day to the new time and update NTP and notify hrtimers
374 int do_settimeofday(const struct timespec *tv)
376 struct timekeeper *tk = &timekeeper;
377 struct timespec ts_delta, xt;
380 if (!timespec_valid_strict(tv))
383 write_seqlock_irqsave(&tk->lock, flags);
385 timekeeping_forward_now(tk);
388 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
389 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
391 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
393 tk_set_xtime(tk, tv);
395 timekeeping_update(tk, true);
397 write_sequnlock_irqrestore(&tk->lock, flags);
399 /* signal hrtimers about time change */
404 EXPORT_SYMBOL(do_settimeofday);
407 * timekeeping_inject_offset - Adds or subtracts from the current time.
408 * @tv: pointer to the timespec variable containing the offset
410 * Adds or subtracts an offset value from the current time.
412 int timekeeping_inject_offset(struct timespec *ts)
414 struct timekeeper *tk = &timekeeper;
419 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
422 write_seqlock_irqsave(&tk->lock, flags);
424 timekeeping_forward_now(tk);
426 /* Make sure the proposed value is valid */
427 tmp = timespec_add(tk_xtime(tk), *ts);
428 if (!timespec_valid_strict(&tmp)) {
433 tk_xtime_add(tk, ts);
434 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
436 error: /* even if we error out, we forwarded the time, so call update */
437 timekeeping_update(tk, true);
439 write_sequnlock_irqrestore(&tk->lock, flags);
441 /* signal hrtimers about time change */
446 EXPORT_SYMBOL(timekeeping_inject_offset);
449 * change_clocksource - Swaps clocksources if a new one is available
451 * Accumulates current time interval and initializes new clocksource
453 static int change_clocksource(void *data)
455 struct timekeeper *tk = &timekeeper;
456 struct clocksource *new, *old;
459 new = (struct clocksource *) data;
461 write_seqlock_irqsave(&tk->lock, flags);
463 timekeeping_forward_now(tk);
464 if (!new->enable || new->enable(new) == 0) {
466 tk_setup_internals(tk, new);
470 timekeeping_update(tk, true);
472 write_sequnlock_irqrestore(&tk->lock, flags);
478 * timekeeping_notify - Install a new clock source
479 * @clock: pointer to the clock source
481 * This function is called from clocksource.c after a new, better clock
482 * source has been registered. The caller holds the clocksource_mutex.
484 void timekeeping_notify(struct clocksource *clock)
486 struct timekeeper *tk = &timekeeper;
488 if (tk->clock == clock)
490 stop_machine(change_clocksource, clock, NULL);
495 * ktime_get_real - get the real (wall-) time in ktime_t format
497 * returns the time in ktime_t format
499 ktime_t ktime_get_real(void)
503 getnstimeofday(&now);
505 return timespec_to_ktime(now);
507 EXPORT_SYMBOL_GPL(ktime_get_real);
510 * getrawmonotonic - Returns the raw monotonic time in a timespec
511 * @ts: pointer to the timespec to be set
513 * Returns the raw monotonic time (completely un-modified by ntp)
515 void getrawmonotonic(struct timespec *ts)
517 struct timekeeper *tk = &timekeeper;
522 seq = read_seqbegin(&tk->lock);
523 nsecs = timekeeping_get_ns_raw(tk);
526 } while (read_seqretry(&tk->lock, seq));
528 timespec_add_ns(ts, nsecs);
530 EXPORT_SYMBOL(getrawmonotonic);
533 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
535 int timekeeping_valid_for_hres(void)
537 struct timekeeper *tk = &timekeeper;
542 seq = read_seqbegin(&tk->lock);
544 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
546 } while (read_seqretry(&tk->lock, seq));
552 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
554 u64 timekeeping_max_deferment(void)
556 struct timekeeper *tk = &timekeeper;
561 seq = read_seqbegin(&tk->lock);
563 ret = tk->clock->max_idle_ns;
565 } while (read_seqretry(&tk->lock, seq));
571 * read_persistent_clock - Return time from the persistent clock.
573 * Weak dummy function for arches that do not yet support it.
574 * Reads the time from the battery backed persistent clock.
575 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
577 * XXX - Do be sure to remove it once all arches implement it.
579 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
586 * read_boot_clock - Return time of the system start.
588 * Weak dummy function for arches that do not yet support it.
589 * Function to read the exact time the system has been started.
590 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
592 * XXX - Do be sure to remove it once all arches implement it.
594 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
601 * timekeeping_init - Initializes the clocksource and common timekeeping values
603 void __init timekeeping_init(void)
605 struct timekeeper *tk = &timekeeper;
606 struct clocksource *clock;
608 struct timespec now, boot, tmp;
610 read_persistent_clock(&now);
611 if (!timespec_valid_strict(&now)) {
612 pr_warn("WARNING: Persistent clock returned invalid value!\n"
613 " Check your CMOS/BIOS settings.\n");
618 read_boot_clock(&boot);
619 if (!timespec_valid_strict(&boot)) {
620 pr_warn("WARNING: Boot clock returned invalid value!\n"
621 " Check your CMOS/BIOS settings.\n");
626 seqlock_init(&tk->lock);
630 write_seqlock_irqsave(&tk->lock, flags);
631 clock = clocksource_default_clock();
633 clock->enable(clock);
634 tk_setup_internals(tk, clock);
636 tk_set_xtime(tk, &now);
637 tk->raw_time.tv_sec = 0;
638 tk->raw_time.tv_nsec = 0;
639 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
642 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
643 tk_set_wall_to_mono(tk, tmp);
647 tk_set_sleep_time(tk, tmp);
649 write_sequnlock_irqrestore(&tk->lock, flags);
652 /* time in seconds when suspend began */
653 static struct timespec timekeeping_suspend_time;
656 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
657 * @delta: pointer to a timespec delta value
659 * Takes a timespec offset measuring a suspend interval and properly
660 * adds the sleep offset to the timekeeping variables.
662 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
663 struct timespec *delta)
665 if (!timespec_valid_strict(delta)) {
666 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
667 "sleep delta value!\n");
670 tk_xtime_add(tk, delta);
671 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
672 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
676 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
677 * @delta: pointer to a timespec delta value
679 * This hook is for architectures that cannot support read_persistent_clock
680 * because their RTC/persistent clock is only accessible when irqs are enabled.
682 * This function should only be called by rtc_resume(), and allows
683 * a suspend offset to be injected into the timekeeping values.
685 void timekeeping_inject_sleeptime(struct timespec *delta)
687 struct timekeeper *tk = &timekeeper;
691 /* Make sure we don't set the clock twice */
692 read_persistent_clock(&ts);
693 if (!(ts.tv_sec == 0 && ts.tv_nsec == 0))
696 write_seqlock_irqsave(&tk->lock, flags);
698 timekeeping_forward_now(tk);
700 __timekeeping_inject_sleeptime(tk, delta);
702 timekeeping_update(tk, true);
704 write_sequnlock_irqrestore(&tk->lock, flags);
706 /* signal hrtimers about time change */
711 * timekeeping_resume - Resumes the generic timekeeping subsystem.
713 * This is for the generic clocksource timekeeping.
714 * xtime/wall_to_monotonic/jiffies/etc are
715 * still managed by arch specific suspend/resume code.
717 static void timekeeping_resume(void)
719 struct timekeeper *tk = &timekeeper;
723 read_persistent_clock(&ts);
725 clocksource_resume();
727 write_seqlock_irqsave(&tk->lock, flags);
729 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
730 ts = timespec_sub(ts, timekeeping_suspend_time);
731 __timekeeping_inject_sleeptime(tk, &ts);
733 /* re-base the last cycle value */
734 tk->clock->cycle_last = tk->clock->read(tk->clock);
736 timekeeping_suspended = 0;
737 timekeeping_update(tk, false);
738 write_sequnlock_irqrestore(&tk->lock, flags);
740 touch_softlockup_watchdog();
742 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
744 /* Resume hrtimers */
748 static int timekeeping_suspend(void)
750 struct timekeeper *tk = &timekeeper;
752 struct timespec delta, delta_delta;
753 static struct timespec old_delta;
755 read_persistent_clock(&timekeeping_suspend_time);
757 write_seqlock_irqsave(&tk->lock, flags);
758 timekeeping_forward_now(tk);
759 timekeeping_suspended = 1;
762 * To avoid drift caused by repeated suspend/resumes,
763 * which each can add ~1 second drift error,
764 * try to compensate so the difference in system time
765 * and persistent_clock time stays close to constant.
767 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
768 delta_delta = timespec_sub(delta, old_delta);
769 if (abs(delta_delta.tv_sec) >= 2) {
771 * if delta_delta is too large, assume time correction
772 * has occured and set old_delta to the current delta.
776 /* Otherwise try to adjust old_system to compensate */
777 timekeeping_suspend_time =
778 timespec_add(timekeeping_suspend_time, delta_delta);
780 write_sequnlock_irqrestore(&tk->lock, flags);
782 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
783 clocksource_suspend();
788 /* sysfs resume/suspend bits for timekeeping */
789 static struct syscore_ops timekeeping_syscore_ops = {
790 .resume = timekeeping_resume,
791 .suspend = timekeeping_suspend,
794 static int __init timekeeping_init_ops(void)
796 register_syscore_ops(&timekeeping_syscore_ops);
800 device_initcall(timekeeping_init_ops);
803 * If the error is already larger, we look ahead even further
804 * to compensate for late or lost adjustments.
806 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
807 s64 error, s64 *interval,
815 * Use the current error value to determine how much to look ahead.
816 * The larger the error the slower we adjust for it to avoid problems
817 * with losing too many ticks, otherwise we would overadjust and
818 * produce an even larger error. The smaller the adjustment the
819 * faster we try to adjust for it, as lost ticks can do less harm
820 * here. This is tuned so that an error of about 1 msec is adjusted
821 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
823 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
824 error2 = abs(error2);
825 for (look_ahead = 0; error2 > 0; look_ahead++)
829 * Now calculate the error in (1 << look_ahead) ticks, but first
830 * remove the single look ahead already included in the error.
832 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
833 tick_error -= tk->xtime_interval >> 1;
834 error = ((error - tick_error) >> look_ahead) + tick_error;
836 /* Finally calculate the adjustment shift value. */
841 *interval = -*interval;
845 for (adj = 0; error > i; adj++)
854 * Adjust the multiplier to reduce the error value,
855 * this is optimized for the most common adjustments of -1,0,1,
856 * for other values we can do a bit more work.
858 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
860 s64 error, interval = tk->cycle_interval;
864 * The point of this is to check if the error is greater than half
867 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
869 * Note we subtract one in the shift, so that error is really error*2.
870 * This "saves" dividing(shifting) interval twice, but keeps the
871 * (error > interval) comparison as still measuring if error is
872 * larger than half an interval.
874 * Note: It does not "save" on aggravation when reading the code.
876 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
877 if (error > interval) {
879 * We now divide error by 4(via shift), which checks if
880 * the error is greater than twice the interval.
881 * If it is greater, we need a bigadjust, if its smaller,
882 * we can adjust by 1.
886 * XXX - In update_wall_time, we round up to the next
887 * nanosecond, and store the amount rounded up into
888 * the error. This causes the likely below to be unlikely.
890 * The proper fix is to avoid rounding up by using
891 * the high precision tk->xtime_nsec instead of
892 * xtime.tv_nsec everywhere. Fixing this will take some
895 if (likely(error <= interval))
898 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
900 if (error < -interval) {
901 /* See comment above, this is just switched for the negative */
903 if (likely(error >= -interval)) {
905 interval = -interval;
908 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
915 if (unlikely(tk->clock->maxadj &&
916 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
917 printk_once(KERN_WARNING
918 "Adjusting %s more than 11%% (%ld vs %ld)\n",
919 tk->clock->name, (long)tk->mult + adj,
920 (long)tk->clock->mult + tk->clock->maxadj);
923 * So the following can be confusing.
925 * To keep things simple, lets assume adj == 1 for now.
927 * When adj != 1, remember that the interval and offset values
928 * have been appropriately scaled so the math is the same.
930 * The basic idea here is that we're increasing the multiplier
931 * by one, this causes the xtime_interval to be incremented by
932 * one cycle_interval. This is because:
933 * xtime_interval = cycle_interval * mult
934 * So if mult is being incremented by one:
935 * xtime_interval = cycle_interval * (mult + 1)
937 * xtime_interval = (cycle_interval * mult) + cycle_interval
938 * Which can be shortened to:
939 * xtime_interval += cycle_interval
941 * So offset stores the non-accumulated cycles. Thus the current
942 * time (in shifted nanoseconds) is:
943 * now = (offset * adj) + xtime_nsec
944 * Now, even though we're adjusting the clock frequency, we have
945 * to keep time consistent. In other words, we can't jump back
946 * in time, and we also want to avoid jumping forward in time.
948 * So given the same offset value, we need the time to be the same
949 * both before and after the freq adjustment.
950 * now = (offset * adj_1) + xtime_nsec_1
951 * now = (offset * adj_2) + xtime_nsec_2
953 * (offset * adj_1) + xtime_nsec_1 =
954 * (offset * adj_2) + xtime_nsec_2
958 * (offset * adj_1) + xtime_nsec_1 =
959 * (offset * (adj_1+1)) + xtime_nsec_2
960 * (offset * adj_1) + xtime_nsec_1 =
961 * (offset * adj_1) + offset + xtime_nsec_2
962 * Canceling the sides:
963 * xtime_nsec_1 = offset + xtime_nsec_2
965 * xtime_nsec_2 = xtime_nsec_1 - offset
966 * Which simplfies to:
967 * xtime_nsec -= offset
969 * XXX - TODO: Doc ntp_error calculation.
972 tk->xtime_interval += interval;
973 tk->xtime_nsec -= offset;
974 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
978 * It may be possible that when we entered this function, xtime_nsec
979 * was very small. Further, if we're slightly speeding the clocksource
980 * in the code above, its possible the required corrective factor to
981 * xtime_nsec could cause it to underflow.
983 * Now, since we already accumulated the second, cannot simply roll
984 * the accumulated second back, since the NTP subsystem has been
985 * notified via second_overflow. So instead we push xtime_nsec forward
986 * by the amount we underflowed, and add that amount into the error.
988 * We'll correct this error next time through this function, when
989 * xtime_nsec is not as small.
991 if (unlikely((s64)tk->xtime_nsec < 0)) {
992 s64 neg = -(s64)tk->xtime_nsec;
994 tk->ntp_error += neg << tk->ntp_error_shift;
1000 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1002 * Helper function that accumulates a the nsecs greater then a second
1003 * from the xtime_nsec field to the xtime_secs field.
1004 * It also calls into the NTP code to handle leapsecond processing.
1007 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1009 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1011 while (tk->xtime_nsec >= nsecps) {
1014 tk->xtime_nsec -= nsecps;
1017 /* Figure out if its a leap sec and apply if needed */
1018 leap = second_overflow(tk->xtime_sec);
1019 if (unlikely(leap)) {
1022 tk->xtime_sec += leap;
1026 tk_set_wall_to_mono(tk,
1027 timespec_sub(tk->wall_to_monotonic, ts));
1029 clock_was_set_delayed();
1035 * logarithmic_accumulation - shifted accumulation of cycles
1037 * This functions accumulates a shifted interval of cycles into
1038 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1041 * Returns the unconsumed cycles.
1043 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1048 /* If the offset is smaller then a shifted interval, do nothing */
1049 if (offset < tk->cycle_interval<<shift)
1052 /* Accumulate one shifted interval */
1053 offset -= tk->cycle_interval << shift;
1054 tk->clock->cycle_last += tk->cycle_interval << shift;
1056 tk->xtime_nsec += tk->xtime_interval << shift;
1057 accumulate_nsecs_to_secs(tk);
1059 /* Accumulate raw time */
1060 raw_nsecs = tk->raw_interval << shift;
1061 raw_nsecs += tk->raw_time.tv_nsec;
1062 if (raw_nsecs >= NSEC_PER_SEC) {
1063 u64 raw_secs = raw_nsecs;
1064 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1065 tk->raw_time.tv_sec += raw_secs;
1067 tk->raw_time.tv_nsec = raw_nsecs;
1069 /* Accumulate error between NTP and clock interval */
1070 tk->ntp_error += ntp_tick_length() << shift;
1071 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1072 (tk->ntp_error_shift + shift);
1078 * update_wall_time - Uses the current clocksource to increment the wall time
1081 static void update_wall_time(void)
1083 struct clocksource *clock;
1084 struct timekeeper *tk = &timekeeper;
1086 int shift = 0, maxshift;
1087 unsigned long flags;
1090 write_seqlock_irqsave(&tk->lock, flags);
1092 /* Make sure we're fully resumed: */
1093 if (unlikely(timekeeping_suspended))
1098 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1099 offset = tk->cycle_interval;
1101 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1104 /* Check if there's really nothing to do */
1105 if (offset < tk->cycle_interval)
1109 * With NO_HZ we may have to accumulate many cycle_intervals
1110 * (think "ticks") worth of time at once. To do this efficiently,
1111 * we calculate the largest doubling multiple of cycle_intervals
1112 * that is smaller than the offset. We then accumulate that
1113 * chunk in one go, and then try to consume the next smaller
1116 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1117 shift = max(0, shift);
1118 /* Bound shift to one less than what overflows tick_length */
1119 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1120 shift = min(shift, maxshift);
1121 while (offset >= tk->cycle_interval) {
1122 offset = logarithmic_accumulation(tk, offset, shift);
1123 if (offset < tk->cycle_interval<<shift)
1127 /* correct the clock when NTP error is too big */
1128 timekeeping_adjust(tk, offset);
1132 * Store only full nanoseconds into xtime_nsec after rounding
1133 * it up and add the remainder to the error difference.
1134 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1135 * by truncating the remainder in vsyscalls. However, it causes
1136 * additional work to be done in timekeeping_adjust(). Once
1137 * the vsyscall implementations are converted to use xtime_nsec
1138 * (shifted nanoseconds), this can be killed.
1140 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1141 tk->xtime_nsec -= remainder;
1142 tk->xtime_nsec += 1ULL << tk->shift;
1143 tk->ntp_error += remainder << tk->ntp_error_shift;
1146 * Finally, make sure that after the rounding
1147 * xtime_nsec isn't larger than NSEC_PER_SEC
1149 accumulate_nsecs_to_secs(tk);
1151 timekeeping_update(tk, false);
1154 write_sequnlock_irqrestore(&tk->lock, flags);
1159 * getboottime - Return the real time of system boot.
1160 * @ts: pointer to the timespec to be set
1162 * Returns the wall-time of boot in a timespec.
1164 * This is based on the wall_to_monotonic offset and the total suspend
1165 * time. Calls to settimeofday will affect the value returned (which
1166 * basically means that however wrong your real time clock is at boot time,
1167 * you get the right time here).
1169 void getboottime(struct timespec *ts)
1171 struct timekeeper *tk = &timekeeper;
1172 struct timespec boottime = {
1173 .tv_sec = tk->wall_to_monotonic.tv_sec +
1174 tk->total_sleep_time.tv_sec,
1175 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1176 tk->total_sleep_time.tv_nsec
1179 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1181 EXPORT_SYMBOL_GPL(getboottime);
1184 * get_monotonic_boottime - Returns monotonic time since boot
1185 * @ts: pointer to the timespec to be set
1187 * Returns the monotonic time since boot in a timespec.
1189 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1190 * includes the time spent in suspend.
1192 void get_monotonic_boottime(struct timespec *ts)
1194 struct timekeeper *tk = &timekeeper;
1195 struct timespec tomono, sleep;
1199 WARN_ON(timekeeping_suspended);
1202 seq = read_seqbegin(&tk->lock);
1203 ts->tv_sec = tk->xtime_sec;
1204 nsec = timekeeping_get_ns(tk);
1205 tomono = tk->wall_to_monotonic;
1206 sleep = tk->total_sleep_time;
1208 } while (read_seqretry(&tk->lock, seq));
1210 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1212 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1214 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1217 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1219 * Returns the monotonic time since boot in a ktime
1221 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1222 * includes the time spent in suspend.
1224 ktime_t ktime_get_boottime(void)
1228 get_monotonic_boottime(&ts);
1229 return timespec_to_ktime(ts);
1231 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1234 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1235 * @ts: pointer to the timespec to be converted
1237 void monotonic_to_bootbased(struct timespec *ts)
1239 struct timekeeper *tk = &timekeeper;
1241 *ts = timespec_add(*ts, tk->total_sleep_time);
1243 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1245 unsigned long get_seconds(void)
1247 struct timekeeper *tk = &timekeeper;
1249 return tk->xtime_sec;
1251 EXPORT_SYMBOL(get_seconds);
1253 struct timespec __current_kernel_time(void)
1255 struct timekeeper *tk = &timekeeper;
1257 return tk_xtime(tk);
1260 struct timespec current_kernel_time(void)
1262 struct timekeeper *tk = &timekeeper;
1263 struct timespec now;
1267 seq = read_seqbegin(&tk->lock);
1270 } while (read_seqretry(&tk->lock, seq));
1274 EXPORT_SYMBOL(current_kernel_time);
1276 struct timespec get_monotonic_coarse(void)
1278 struct timekeeper *tk = &timekeeper;
1279 struct timespec now, mono;
1283 seq = read_seqbegin(&tk->lock);
1286 mono = tk->wall_to_monotonic;
1287 } while (read_seqretry(&tk->lock, seq));
1289 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1290 now.tv_nsec + mono.tv_nsec);
1295 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1296 * without sampling the sequence number in xtime_lock.
1297 * jiffies is defined in the linker script...
1299 void do_timer(unsigned long ticks)
1301 jiffies_64 += ticks;
1303 calc_global_load(ticks);
1307 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1308 * and sleep offsets.
1309 * @xtim: pointer to timespec to be set with xtime
1310 * @wtom: pointer to timespec to be set with wall_to_monotonic
1311 * @sleep: pointer to timespec to be set with time in suspend
1313 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1314 struct timespec *wtom, struct timespec *sleep)
1316 struct timekeeper *tk = &timekeeper;
1320 seq = read_seqbegin(&tk->lock);
1321 *xtim = tk_xtime(tk);
1322 *wtom = tk->wall_to_monotonic;
1323 *sleep = tk->total_sleep_time;
1324 } while (read_seqretry(&tk->lock, seq));
1327 #ifdef CONFIG_HIGH_RES_TIMERS
1329 * ktime_get_update_offsets - hrtimer helper
1330 * @offs_real: pointer to storage for monotonic -> realtime offset
1331 * @offs_boot: pointer to storage for monotonic -> boottime offset
1333 * Returns current monotonic time and updates the offsets
1334 * Called from hrtimer_interupt() or retrigger_next_event()
1336 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1338 struct timekeeper *tk = &timekeeper;
1344 seq = read_seqbegin(&tk->lock);
1346 secs = tk->xtime_sec;
1347 nsecs = timekeeping_get_ns(tk);
1349 *offs_real = tk->offs_real;
1350 *offs_boot = tk->offs_boot;
1351 } while (read_seqretry(&tk->lock, seq));
1353 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1354 now = ktime_sub(now, *offs_real);
1360 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1362 ktime_t ktime_get_monotonic_offset(void)
1364 struct timekeeper *tk = &timekeeper;
1366 struct timespec wtom;
1369 seq = read_seqbegin(&tk->lock);
1370 wtom = tk->wall_to_monotonic;
1371 } while (read_seqretry(&tk->lock, seq));
1373 return timespec_to_ktime(wtom);
1375 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1378 * xtime_update() - advances the timekeeping infrastructure
1379 * @ticks: number of ticks, that have elapsed since the last call.
1381 * Must be called with interrupts disabled.
1383 void xtime_update(unsigned long ticks)
1385 write_seqlock(&xtime_lock);
1387 write_sequnlock(&xtime_lock);