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>
24 #include <linux/pvclock_gtod.h>
27 static struct timekeeper timekeeper;
29 /* flag for if timekeeping is suspended */
30 int __read_mostly timekeeping_suspended;
32 /* Flag for if there is a persistent clock on this platform */
33 bool __read_mostly persistent_clock_exist = false;
35 static inline void tk_normalize_xtime(struct timekeeper *tk)
37 while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
38 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
43 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
45 tk->xtime_sec = ts->tv_sec;
46 tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
49 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
51 tk->xtime_sec += ts->tv_sec;
52 tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
53 tk_normalize_xtime(tk);
56 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
61 * Verify consistency of: offset_real = -wall_to_monotonic
62 * before modifying anything
64 set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
65 -tk->wall_to_monotonic.tv_nsec);
66 WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
67 tk->wall_to_monotonic = wtm;
68 set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
69 tk->offs_real = timespec_to_ktime(tmp);
72 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
74 /* Verify consistency before modifying */
75 WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
77 tk->total_sleep_time = t;
78 tk->offs_boot = timespec_to_ktime(t);
82 * timekeeper_setup_internals - Set up internals to use clocksource clock.
84 * @clock: Pointer to clocksource.
86 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
87 * pair and interval request.
89 * Unless you're the timekeeping code, you should not be using this!
91 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
95 struct clocksource *old_clock;
97 old_clock = tk->clock;
99 clock->cycle_last = clock->read(clock);
101 /* Do the ns -> cycle conversion first, using original mult */
102 tmp = NTP_INTERVAL_LENGTH;
103 tmp <<= clock->shift;
105 tmp += clock->mult/2;
106 do_div(tmp, clock->mult);
110 interval = (cycle_t) tmp;
111 tk->cycle_interval = interval;
113 /* Go back from cycles -> shifted ns */
114 tk->xtime_interval = (u64) interval * clock->mult;
115 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
117 ((u64) interval * clock->mult) >> clock->shift;
119 /* if changing clocks, convert xtime_nsec shift units */
121 int shift_change = clock->shift - old_clock->shift;
122 if (shift_change < 0)
123 tk->xtime_nsec >>= -shift_change;
125 tk->xtime_nsec <<= shift_change;
127 tk->shift = clock->shift;
130 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
133 * The timekeeper keeps its own mult values for the currently
134 * active clocksource. These value will be adjusted via NTP
135 * to counteract clock drifting.
137 tk->mult = clock->mult;
140 /* Timekeeper helper functions. */
142 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
143 u32 (*arch_gettimeoffset)(void);
145 u32 get_arch_timeoffset(void)
147 if (likely(arch_gettimeoffset))
148 return arch_gettimeoffset();
152 static inline u32 get_arch_timeoffset(void) { return 0; }
155 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
157 cycle_t cycle_now, cycle_delta;
158 struct clocksource *clock;
161 /* read clocksource: */
163 cycle_now = clock->read(clock);
165 /* calculate the delta since the last update_wall_time: */
166 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
168 nsec = cycle_delta * tk->mult + tk->xtime_nsec;
171 /* If arch requires, add in get_arch_timeoffset() */
172 return nsec + get_arch_timeoffset();
175 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
177 cycle_t cycle_now, cycle_delta;
178 struct clocksource *clock;
181 /* read clocksource: */
183 cycle_now = clock->read(clock);
185 /* calculate the delta since the last update_wall_time: */
186 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
188 /* convert delta to nanoseconds. */
189 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
191 /* If arch requires, add in get_arch_timeoffset() */
192 return nsec + get_arch_timeoffset();
195 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
197 static void update_pvclock_gtod(struct timekeeper *tk)
199 raw_notifier_call_chain(&pvclock_gtod_chain, 0, tk);
203 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
205 * Must hold write on timekeeper.lock
207 int pvclock_gtod_register_notifier(struct notifier_block *nb)
209 struct timekeeper *tk = &timekeeper;
213 write_seqlock_irqsave(&tk->lock, flags);
214 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
215 /* update timekeeping data */
216 update_pvclock_gtod(tk);
217 write_sequnlock_irqrestore(&tk->lock, flags);
221 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
224 * pvclock_gtod_unregister_notifier - unregister a pvclock
225 * timedata update listener
227 * Must hold write on timekeeper.lock
229 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
231 struct timekeeper *tk = &timekeeper;
235 write_seqlock_irqsave(&tk->lock, flags);
236 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
237 write_sequnlock_irqrestore(&tk->lock, flags);
241 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
243 /* must hold write on timekeeper.lock */
244 static void timekeeping_update(struct timekeeper *tk, bool clearntp)
251 update_pvclock_gtod(tk);
255 * timekeeping_forward_now - update clock to the current time
257 * Forward the current clock to update its state since the last call to
258 * update_wall_time(). This is useful before significant clock changes,
259 * as it avoids having to deal with this time offset explicitly.
261 static void timekeeping_forward_now(struct timekeeper *tk)
263 cycle_t cycle_now, cycle_delta;
264 struct clocksource *clock;
268 cycle_now = clock->read(clock);
269 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
270 clock->cycle_last = cycle_now;
272 tk->xtime_nsec += cycle_delta * tk->mult;
274 /* If arch requires, add in get_arch_timeoffset() */
275 tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
277 tk_normalize_xtime(tk);
279 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
280 timespec_add_ns(&tk->raw_time, nsec);
284 * __getnstimeofday - Returns the time of day in a timespec.
285 * @ts: pointer to the timespec to be set
287 * Updates the time of day in the timespec.
288 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
290 int __getnstimeofday(struct timespec *ts)
292 struct timekeeper *tk = &timekeeper;
297 seq = read_seqbegin(&tk->lock);
299 ts->tv_sec = tk->xtime_sec;
300 nsecs = timekeeping_get_ns(tk);
302 } while (read_seqretry(&tk->lock, seq));
305 timespec_add_ns(ts, nsecs);
308 * Do not bail out early, in case there were callers still using
309 * the value, even in the face of the WARN_ON.
311 if (unlikely(timekeeping_suspended))
315 EXPORT_SYMBOL(__getnstimeofday);
318 * getnstimeofday - Returns the time of day in a timespec.
319 * @ts: pointer to the timespec to be set
321 * Returns the time of day in a timespec (WARN if suspended).
323 void getnstimeofday(struct timespec *ts)
325 WARN_ON(__getnstimeofday(ts));
327 EXPORT_SYMBOL(getnstimeofday);
329 ktime_t ktime_get(void)
331 struct timekeeper *tk = &timekeeper;
335 WARN_ON(timekeeping_suspended);
338 seq = read_seqbegin(&tk->lock);
339 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
340 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
342 } while (read_seqretry(&tk->lock, seq));
344 * Use ktime_set/ktime_add_ns to create a proper ktime on
345 * 32-bit architectures without CONFIG_KTIME_SCALAR.
347 return ktime_add_ns(ktime_set(secs, 0), nsecs);
349 EXPORT_SYMBOL_GPL(ktime_get);
352 * ktime_get_ts - get the monotonic clock in timespec format
353 * @ts: pointer to timespec variable
355 * The function calculates the monotonic clock from the realtime
356 * clock and the wall_to_monotonic offset and stores the result
357 * in normalized timespec format in the variable pointed to by @ts.
359 void ktime_get_ts(struct timespec *ts)
361 struct timekeeper *tk = &timekeeper;
362 struct timespec tomono;
366 WARN_ON(timekeeping_suspended);
369 seq = read_seqbegin(&tk->lock);
370 ts->tv_sec = tk->xtime_sec;
371 nsec = timekeeping_get_ns(tk);
372 tomono = tk->wall_to_monotonic;
374 } while (read_seqretry(&tk->lock, seq));
376 ts->tv_sec += tomono.tv_sec;
378 timespec_add_ns(ts, nsec + tomono.tv_nsec);
380 EXPORT_SYMBOL_GPL(ktime_get_ts);
382 #ifdef CONFIG_NTP_PPS
385 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
386 * @ts_raw: pointer to the timespec to be set to raw monotonic time
387 * @ts_real: pointer to the timespec to be set to the time of day
389 * This function reads both the time of day and raw monotonic time at the
390 * same time atomically and stores the resulting timestamps in timespec
393 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
395 struct timekeeper *tk = &timekeeper;
397 s64 nsecs_raw, nsecs_real;
399 WARN_ON_ONCE(timekeeping_suspended);
402 seq = read_seqbegin(&tk->lock);
404 *ts_raw = tk->raw_time;
405 ts_real->tv_sec = tk->xtime_sec;
406 ts_real->tv_nsec = 0;
408 nsecs_raw = timekeeping_get_ns_raw(tk);
409 nsecs_real = timekeeping_get_ns(tk);
411 } while (read_seqretry(&tk->lock, seq));
413 timespec_add_ns(ts_raw, nsecs_raw);
414 timespec_add_ns(ts_real, nsecs_real);
416 EXPORT_SYMBOL(getnstime_raw_and_real);
418 #endif /* CONFIG_NTP_PPS */
421 * do_gettimeofday - Returns the time of day in a timeval
422 * @tv: pointer to the timeval to be set
424 * NOTE: Users should be converted to using getnstimeofday()
426 void do_gettimeofday(struct timeval *tv)
430 getnstimeofday(&now);
431 tv->tv_sec = now.tv_sec;
432 tv->tv_usec = now.tv_nsec/1000;
434 EXPORT_SYMBOL(do_gettimeofday);
437 * do_settimeofday - Sets the time of day
438 * @tv: pointer to the timespec variable containing the new time
440 * Sets the time of day to the new time and update NTP and notify hrtimers
442 int do_settimeofday(const struct timespec *tv)
444 struct timekeeper *tk = &timekeeper;
445 struct timespec ts_delta, xt;
448 if (!timespec_valid_strict(tv))
451 write_seqlock_irqsave(&tk->lock, flags);
453 timekeeping_forward_now(tk);
456 ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
457 ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
459 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
461 tk_set_xtime(tk, tv);
463 timekeeping_update(tk, true);
465 write_sequnlock_irqrestore(&tk->lock, flags);
467 /* signal hrtimers about time change */
472 EXPORT_SYMBOL(do_settimeofday);
475 * timekeeping_inject_offset - Adds or subtracts from the current time.
476 * @tv: pointer to the timespec variable containing the offset
478 * Adds or subtracts an offset value from the current time.
480 int timekeeping_inject_offset(struct timespec *ts)
482 struct timekeeper *tk = &timekeeper;
487 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
490 write_seqlock_irqsave(&tk->lock, flags);
492 timekeeping_forward_now(tk);
494 /* Make sure the proposed value is valid */
495 tmp = timespec_add(tk_xtime(tk), *ts);
496 if (!timespec_valid_strict(&tmp)) {
501 tk_xtime_add(tk, ts);
502 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
504 error: /* even if we error out, we forwarded the time, so call update */
505 timekeeping_update(tk, true);
507 write_sequnlock_irqrestore(&tk->lock, flags);
509 /* signal hrtimers about time change */
514 EXPORT_SYMBOL(timekeeping_inject_offset);
518 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
521 s32 timekeeping_get_tai_offset(void)
523 struct timekeeper *tk = &timekeeper;
528 seq = read_seqbegin(&tk->lock);
529 ret = tk->tai_offset;
530 } while (read_seqretry(&tk->lock, seq));
536 * __timekeeping_set_tai_offset - Lock free worker function
539 void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
541 tk->tai_offset = tai_offset;
545 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
548 void timekeeping_set_tai_offset(s32 tai_offset)
550 struct timekeeper *tk = &timekeeper;
553 write_seqlock_irqsave(&tk->lock, flags);
554 __timekeeping_set_tai_offset(tk, tai_offset);
555 write_sequnlock_irqrestore(&tk->lock, flags);
559 * change_clocksource - Swaps clocksources if a new one is available
561 * Accumulates current time interval and initializes new clocksource
563 static int change_clocksource(void *data)
565 struct timekeeper *tk = &timekeeper;
566 struct clocksource *new, *old;
569 new = (struct clocksource *) data;
571 write_seqlock_irqsave(&tk->lock, flags);
573 timekeeping_forward_now(tk);
574 if (!new->enable || new->enable(new) == 0) {
576 tk_setup_internals(tk, new);
580 timekeeping_update(tk, true);
582 write_sequnlock_irqrestore(&tk->lock, flags);
588 * timekeeping_notify - Install a new clock source
589 * @clock: pointer to the clock source
591 * This function is called from clocksource.c after a new, better clock
592 * source has been registered. The caller holds the clocksource_mutex.
594 void timekeeping_notify(struct clocksource *clock)
596 struct timekeeper *tk = &timekeeper;
598 if (tk->clock == clock)
600 stop_machine(change_clocksource, clock, NULL);
605 * ktime_get_real - get the real (wall-) time in ktime_t format
607 * returns the time in ktime_t format
609 ktime_t ktime_get_real(void)
613 getnstimeofday(&now);
615 return timespec_to_ktime(now);
617 EXPORT_SYMBOL_GPL(ktime_get_real);
620 * getrawmonotonic - Returns the raw monotonic time in a timespec
621 * @ts: pointer to the timespec to be set
623 * Returns the raw monotonic time (completely un-modified by ntp)
625 void getrawmonotonic(struct timespec *ts)
627 struct timekeeper *tk = &timekeeper;
632 seq = read_seqbegin(&tk->lock);
633 nsecs = timekeeping_get_ns_raw(tk);
636 } while (read_seqretry(&tk->lock, seq));
638 timespec_add_ns(ts, nsecs);
640 EXPORT_SYMBOL(getrawmonotonic);
643 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
645 int timekeeping_valid_for_hres(void)
647 struct timekeeper *tk = &timekeeper;
652 seq = read_seqbegin(&tk->lock);
654 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
656 } while (read_seqretry(&tk->lock, seq));
662 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
664 u64 timekeeping_max_deferment(void)
666 struct timekeeper *tk = &timekeeper;
671 seq = read_seqbegin(&tk->lock);
673 ret = tk->clock->max_idle_ns;
675 } while (read_seqretry(&tk->lock, seq));
681 * read_persistent_clock - Return time from the persistent clock.
683 * Weak dummy function for arches that do not yet support it.
684 * Reads the time from the battery backed persistent clock.
685 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
687 * XXX - Do be sure to remove it once all arches implement it.
689 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
696 * read_boot_clock - Return time of the system start.
698 * Weak dummy function for arches that do not yet support it.
699 * Function to read the exact time the system has been started.
700 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
702 * XXX - Do be sure to remove it once all arches implement it.
704 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
711 * timekeeping_init - Initializes the clocksource and common timekeeping values
713 void __init timekeeping_init(void)
715 struct timekeeper *tk = &timekeeper;
716 struct clocksource *clock;
718 struct timespec now, boot, tmp;
720 read_persistent_clock(&now);
722 if (!timespec_valid_strict(&now)) {
723 pr_warn("WARNING: Persistent clock returned invalid value!\n"
724 " Check your CMOS/BIOS settings.\n");
727 } else if (now.tv_sec || now.tv_nsec)
728 persistent_clock_exist = true;
730 read_boot_clock(&boot);
731 if (!timespec_valid_strict(&boot)) {
732 pr_warn("WARNING: Boot clock returned invalid value!\n"
733 " Check your CMOS/BIOS settings.\n");
738 seqlock_init(&tk->lock);
742 write_seqlock_irqsave(&tk->lock, flags);
743 clock = clocksource_default_clock();
745 clock->enable(clock);
746 tk_setup_internals(tk, clock);
748 tk_set_xtime(tk, &now);
749 tk->raw_time.tv_sec = 0;
750 tk->raw_time.tv_nsec = 0;
751 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
754 set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
755 tk_set_wall_to_mono(tk, tmp);
759 tk_set_sleep_time(tk, tmp);
761 write_sequnlock_irqrestore(&tk->lock, flags);
764 /* time in seconds when suspend began */
765 static struct timespec timekeeping_suspend_time;
768 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
769 * @delta: pointer to a timespec delta value
771 * Takes a timespec offset measuring a suspend interval and properly
772 * adds the sleep offset to the timekeeping variables.
774 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
775 struct timespec *delta)
777 if (!timespec_valid_strict(delta)) {
778 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
779 "sleep delta value!\n");
782 tk_xtime_add(tk, delta);
783 tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
784 tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
788 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
789 * @delta: pointer to a timespec delta value
791 * This hook is for architectures that cannot support read_persistent_clock
792 * because their RTC/persistent clock is only accessible when irqs are enabled.
794 * This function should only be called by rtc_resume(), and allows
795 * a suspend offset to be injected into the timekeeping values.
797 void timekeeping_inject_sleeptime(struct timespec *delta)
799 struct timekeeper *tk = &timekeeper;
803 * Make sure we don't set the clock twice, as timekeeping_resume()
806 if (has_persistent_clock())
809 write_seqlock_irqsave(&tk->lock, flags);
811 timekeeping_forward_now(tk);
813 __timekeeping_inject_sleeptime(tk, delta);
815 timekeeping_update(tk, true);
817 write_sequnlock_irqrestore(&tk->lock, flags);
819 /* signal hrtimers about time change */
824 * timekeeping_resume - Resumes the generic timekeeping subsystem.
826 * This is for the generic clocksource timekeeping.
827 * xtime/wall_to_monotonic/jiffies/etc are
828 * still managed by arch specific suspend/resume code.
830 static void timekeeping_resume(void)
832 struct timekeeper *tk = &timekeeper;
833 struct clocksource *clock = tk->clock;
835 struct timespec ts_new, ts_delta;
836 cycle_t cycle_now, cycle_delta;
837 bool suspendtime_found = false;
839 read_persistent_clock(&ts_new);
841 clockevents_resume();
842 clocksource_resume();
844 write_seqlock_irqsave(&tk->lock, flags);
847 * After system resumes, we need to calculate the suspended time and
848 * compensate it for the OS time. There are 3 sources that could be
849 * used: Nonstop clocksource during suspend, persistent clock and rtc
852 * One specific platform may have 1 or 2 or all of them, and the
853 * preference will be:
854 * suspend-nonstop clocksource -> persistent clock -> rtc
855 * The less preferred source will only be tried if there is no better
856 * usable source. The rtc part is handled separately in rtc core code.
858 cycle_now = clock->read(clock);
859 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
860 cycle_now > clock->cycle_last) {
861 u64 num, max = ULLONG_MAX;
862 u32 mult = clock->mult;
863 u32 shift = clock->shift;
866 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
869 * "cycle_delta * mutl" may cause 64 bits overflow, if the
870 * suspended time is too long. In that case we need do the
871 * 64 bits math carefully
874 if (cycle_delta > max) {
875 num = div64_u64(cycle_delta, max);
876 nsec = (((u64) max * mult) >> shift) * num;
877 cycle_delta -= num * max;
879 nsec += ((u64) cycle_delta * mult) >> shift;
881 ts_delta = ns_to_timespec(nsec);
882 suspendtime_found = true;
883 } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
884 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
885 suspendtime_found = true;
888 if (suspendtime_found)
889 __timekeeping_inject_sleeptime(tk, &ts_delta);
891 /* Re-base the last cycle value */
892 clock->cycle_last = cycle_now;
894 timekeeping_suspended = 0;
895 timekeeping_update(tk, false);
896 write_sequnlock_irqrestore(&tk->lock, flags);
898 touch_softlockup_watchdog();
900 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
902 /* Resume hrtimers */
906 static int timekeeping_suspend(void)
908 struct timekeeper *tk = &timekeeper;
910 struct timespec delta, delta_delta;
911 static struct timespec old_delta;
913 read_persistent_clock(&timekeeping_suspend_time);
915 write_seqlock_irqsave(&tk->lock, flags);
916 timekeeping_forward_now(tk);
917 timekeeping_suspended = 1;
920 * To avoid drift caused by repeated suspend/resumes,
921 * which each can add ~1 second drift error,
922 * try to compensate so the difference in system time
923 * and persistent_clock time stays close to constant.
925 delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
926 delta_delta = timespec_sub(delta, old_delta);
927 if (abs(delta_delta.tv_sec) >= 2) {
929 * if delta_delta is too large, assume time correction
930 * has occured and set old_delta to the current delta.
934 /* Otherwise try to adjust old_system to compensate */
935 timekeeping_suspend_time =
936 timespec_add(timekeeping_suspend_time, delta_delta);
938 write_sequnlock_irqrestore(&tk->lock, flags);
940 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
941 clocksource_suspend();
942 clockevents_suspend();
947 /* sysfs resume/suspend bits for timekeeping */
948 static struct syscore_ops timekeeping_syscore_ops = {
949 .resume = timekeeping_resume,
950 .suspend = timekeeping_suspend,
953 static int __init timekeeping_init_ops(void)
955 register_syscore_ops(&timekeeping_syscore_ops);
959 device_initcall(timekeeping_init_ops);
962 * If the error is already larger, we look ahead even further
963 * to compensate for late or lost adjustments.
965 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
966 s64 error, s64 *interval,
974 * Use the current error value to determine how much to look ahead.
975 * The larger the error the slower we adjust for it to avoid problems
976 * with losing too many ticks, otherwise we would overadjust and
977 * produce an even larger error. The smaller the adjustment the
978 * faster we try to adjust for it, as lost ticks can do less harm
979 * here. This is tuned so that an error of about 1 msec is adjusted
980 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
982 error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
983 error2 = abs(error2);
984 for (look_ahead = 0; error2 > 0; look_ahead++)
988 * Now calculate the error in (1 << look_ahead) ticks, but first
989 * remove the single look ahead already included in the error.
991 tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
992 tick_error -= tk->xtime_interval >> 1;
993 error = ((error - tick_error) >> look_ahead) + tick_error;
995 /* Finally calculate the adjustment shift value. */
1000 *interval = -*interval;
1004 for (adj = 0; error > i; adj++)
1013 * Adjust the multiplier to reduce the error value,
1014 * this is optimized for the most common adjustments of -1,0,1,
1015 * for other values we can do a bit more work.
1017 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1019 s64 error, interval = tk->cycle_interval;
1023 * The point of this is to check if the error is greater than half
1026 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1028 * Note we subtract one in the shift, so that error is really error*2.
1029 * This "saves" dividing(shifting) interval twice, but keeps the
1030 * (error > interval) comparison as still measuring if error is
1031 * larger than half an interval.
1033 * Note: It does not "save" on aggravation when reading the code.
1035 error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1036 if (error > interval) {
1038 * We now divide error by 4(via shift), which checks if
1039 * the error is greater than twice the interval.
1040 * If it is greater, we need a bigadjust, if its smaller,
1041 * we can adjust by 1.
1045 * XXX - In update_wall_time, we round up to the next
1046 * nanosecond, and store the amount rounded up into
1047 * the error. This causes the likely below to be unlikely.
1049 * The proper fix is to avoid rounding up by using
1050 * the high precision tk->xtime_nsec instead of
1051 * xtime.tv_nsec everywhere. Fixing this will take some
1054 if (likely(error <= interval))
1057 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1059 if (error < -interval) {
1060 /* See comment above, this is just switched for the negative */
1062 if (likely(error >= -interval)) {
1064 interval = -interval;
1067 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1074 if (unlikely(tk->clock->maxadj &&
1075 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1076 printk_once(KERN_WARNING
1077 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1078 tk->clock->name, (long)tk->mult + adj,
1079 (long)tk->clock->mult + tk->clock->maxadj);
1082 * So the following can be confusing.
1084 * To keep things simple, lets assume adj == 1 for now.
1086 * When adj != 1, remember that the interval and offset values
1087 * have been appropriately scaled so the math is the same.
1089 * The basic idea here is that we're increasing the multiplier
1090 * by one, this causes the xtime_interval to be incremented by
1091 * one cycle_interval. This is because:
1092 * xtime_interval = cycle_interval * mult
1093 * So if mult is being incremented by one:
1094 * xtime_interval = cycle_interval * (mult + 1)
1096 * xtime_interval = (cycle_interval * mult) + cycle_interval
1097 * Which can be shortened to:
1098 * xtime_interval += cycle_interval
1100 * So offset stores the non-accumulated cycles. Thus the current
1101 * time (in shifted nanoseconds) is:
1102 * now = (offset * adj) + xtime_nsec
1103 * Now, even though we're adjusting the clock frequency, we have
1104 * to keep time consistent. In other words, we can't jump back
1105 * in time, and we also want to avoid jumping forward in time.
1107 * So given the same offset value, we need the time to be the same
1108 * both before and after the freq adjustment.
1109 * now = (offset * adj_1) + xtime_nsec_1
1110 * now = (offset * adj_2) + xtime_nsec_2
1112 * (offset * adj_1) + xtime_nsec_1 =
1113 * (offset * adj_2) + xtime_nsec_2
1117 * (offset * adj_1) + xtime_nsec_1 =
1118 * (offset * (adj_1+1)) + xtime_nsec_2
1119 * (offset * adj_1) + xtime_nsec_1 =
1120 * (offset * adj_1) + offset + xtime_nsec_2
1121 * Canceling the sides:
1122 * xtime_nsec_1 = offset + xtime_nsec_2
1124 * xtime_nsec_2 = xtime_nsec_1 - offset
1125 * Which simplfies to:
1126 * xtime_nsec -= offset
1128 * XXX - TODO: Doc ntp_error calculation.
1131 tk->xtime_interval += interval;
1132 tk->xtime_nsec -= offset;
1133 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1137 * It may be possible that when we entered this function, xtime_nsec
1138 * was very small. Further, if we're slightly speeding the clocksource
1139 * in the code above, its possible the required corrective factor to
1140 * xtime_nsec could cause it to underflow.
1142 * Now, since we already accumulated the second, cannot simply roll
1143 * the accumulated second back, since the NTP subsystem has been
1144 * notified via second_overflow. So instead we push xtime_nsec forward
1145 * by the amount we underflowed, and add that amount into the error.
1147 * We'll correct this error next time through this function, when
1148 * xtime_nsec is not as small.
1150 if (unlikely((s64)tk->xtime_nsec < 0)) {
1151 s64 neg = -(s64)tk->xtime_nsec;
1153 tk->ntp_error += neg << tk->ntp_error_shift;
1159 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1161 * Helper function that accumulates a the nsecs greater then a second
1162 * from the xtime_nsec field to the xtime_secs field.
1163 * It also calls into the NTP code to handle leapsecond processing.
1166 static inline void accumulate_nsecs_to_secs(struct timekeeper *tk)
1168 u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1170 while (tk->xtime_nsec >= nsecps) {
1173 tk->xtime_nsec -= nsecps;
1176 /* Figure out if its a leap sec and apply if needed */
1177 leap = second_overflow(tk->xtime_sec);
1178 if (unlikely(leap)) {
1181 tk->xtime_sec += leap;
1185 tk_set_wall_to_mono(tk,
1186 timespec_sub(tk->wall_to_monotonic, ts));
1188 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1190 clock_was_set_delayed();
1196 * logarithmic_accumulation - shifted accumulation of cycles
1198 * This functions accumulates a shifted interval of cycles into
1199 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1202 * Returns the unconsumed cycles.
1204 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1209 /* If the offset is smaller then a shifted interval, do nothing */
1210 if (offset < tk->cycle_interval<<shift)
1213 /* Accumulate one shifted interval */
1214 offset -= tk->cycle_interval << shift;
1215 tk->clock->cycle_last += tk->cycle_interval << shift;
1217 tk->xtime_nsec += tk->xtime_interval << shift;
1218 accumulate_nsecs_to_secs(tk);
1220 /* Accumulate raw time */
1221 raw_nsecs = (u64)tk->raw_interval << shift;
1222 raw_nsecs += tk->raw_time.tv_nsec;
1223 if (raw_nsecs >= NSEC_PER_SEC) {
1224 u64 raw_secs = raw_nsecs;
1225 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1226 tk->raw_time.tv_sec += raw_secs;
1228 tk->raw_time.tv_nsec = raw_nsecs;
1230 /* Accumulate error between NTP and clock interval */
1231 tk->ntp_error += ntp_tick_length() << shift;
1232 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1233 (tk->ntp_error_shift + shift);
1238 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1239 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1244 * Store only full nanoseconds into xtime_nsec after rounding
1245 * it up and add the remainder to the error difference.
1246 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1247 * by truncating the remainder in vsyscalls. However, it causes
1248 * additional work to be done in timekeeping_adjust(). Once
1249 * the vsyscall implementations are converted to use xtime_nsec
1250 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1251 * users are removed, this can be killed.
1253 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1254 tk->xtime_nsec -= remainder;
1255 tk->xtime_nsec += 1ULL << tk->shift;
1256 tk->ntp_error += remainder << tk->ntp_error_shift;
1260 #define old_vsyscall_fixup(tk)
1266 * update_wall_time - Uses the current clocksource to increment the wall time
1269 static void update_wall_time(void)
1271 struct clocksource *clock;
1272 struct timekeeper *tk = &timekeeper;
1274 int shift = 0, maxshift;
1275 unsigned long flags;
1277 write_seqlock_irqsave(&tk->lock, flags);
1279 /* Make sure we're fully resumed: */
1280 if (unlikely(timekeeping_suspended))
1285 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1286 offset = tk->cycle_interval;
1288 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1291 /* Check if there's really nothing to do */
1292 if (offset < tk->cycle_interval)
1296 * With NO_HZ we may have to accumulate many cycle_intervals
1297 * (think "ticks") worth of time at once. To do this efficiently,
1298 * we calculate the largest doubling multiple of cycle_intervals
1299 * that is smaller than the offset. We then accumulate that
1300 * chunk in one go, and then try to consume the next smaller
1303 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1304 shift = max(0, shift);
1305 /* Bound shift to one less than what overflows tick_length */
1306 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1307 shift = min(shift, maxshift);
1308 while (offset >= tk->cycle_interval) {
1309 offset = logarithmic_accumulation(tk, offset, shift);
1310 if (offset < tk->cycle_interval<<shift)
1314 /* correct the clock when NTP error is too big */
1315 timekeeping_adjust(tk, offset);
1318 * XXX This can be killed once everyone converts
1319 * to the new update_vsyscall.
1321 old_vsyscall_fixup(tk);
1324 * Finally, make sure that after the rounding
1325 * xtime_nsec isn't larger than NSEC_PER_SEC
1327 accumulate_nsecs_to_secs(tk);
1329 timekeeping_update(tk, false);
1332 write_sequnlock_irqrestore(&tk->lock, flags);
1337 * getboottime - Return the real time of system boot.
1338 * @ts: pointer to the timespec to be set
1340 * Returns the wall-time of boot in a timespec.
1342 * This is based on the wall_to_monotonic offset and the total suspend
1343 * time. Calls to settimeofday will affect the value returned (which
1344 * basically means that however wrong your real time clock is at boot time,
1345 * you get the right time here).
1347 void getboottime(struct timespec *ts)
1349 struct timekeeper *tk = &timekeeper;
1350 struct timespec boottime = {
1351 .tv_sec = tk->wall_to_monotonic.tv_sec +
1352 tk->total_sleep_time.tv_sec,
1353 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1354 tk->total_sleep_time.tv_nsec
1357 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1359 EXPORT_SYMBOL_GPL(getboottime);
1362 * get_monotonic_boottime - Returns monotonic time since boot
1363 * @ts: pointer to the timespec to be set
1365 * Returns the monotonic time since boot in a timespec.
1367 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1368 * includes the time spent in suspend.
1370 void get_monotonic_boottime(struct timespec *ts)
1372 struct timekeeper *tk = &timekeeper;
1373 struct timespec tomono, sleep;
1377 WARN_ON(timekeeping_suspended);
1380 seq = read_seqbegin(&tk->lock);
1381 ts->tv_sec = tk->xtime_sec;
1382 nsec = timekeeping_get_ns(tk);
1383 tomono = tk->wall_to_monotonic;
1384 sleep = tk->total_sleep_time;
1386 } while (read_seqretry(&tk->lock, seq));
1388 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1390 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1392 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1395 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1397 * Returns the monotonic time since boot in a ktime
1399 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1400 * includes the time spent in suspend.
1402 ktime_t ktime_get_boottime(void)
1406 get_monotonic_boottime(&ts);
1407 return timespec_to_ktime(ts);
1409 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1412 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1413 * @ts: pointer to the timespec to be converted
1415 void monotonic_to_bootbased(struct timespec *ts)
1417 struct timekeeper *tk = &timekeeper;
1419 *ts = timespec_add(*ts, tk->total_sleep_time);
1421 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1423 unsigned long get_seconds(void)
1425 struct timekeeper *tk = &timekeeper;
1427 return tk->xtime_sec;
1429 EXPORT_SYMBOL(get_seconds);
1431 struct timespec __current_kernel_time(void)
1433 struct timekeeper *tk = &timekeeper;
1435 return tk_xtime(tk);
1438 struct timespec current_kernel_time(void)
1440 struct timekeeper *tk = &timekeeper;
1441 struct timespec now;
1445 seq = read_seqbegin(&tk->lock);
1448 } while (read_seqretry(&tk->lock, seq));
1452 EXPORT_SYMBOL(current_kernel_time);
1454 struct timespec get_monotonic_coarse(void)
1456 struct timekeeper *tk = &timekeeper;
1457 struct timespec now, mono;
1461 seq = read_seqbegin(&tk->lock);
1464 mono = tk->wall_to_monotonic;
1465 } while (read_seqretry(&tk->lock, seq));
1467 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1468 now.tv_nsec + mono.tv_nsec);
1473 * Must hold jiffies_lock
1475 void do_timer(unsigned long ticks)
1477 jiffies_64 += ticks;
1479 calc_global_load(ticks);
1483 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1484 * and sleep offsets.
1485 * @xtim: pointer to timespec to be set with xtime
1486 * @wtom: pointer to timespec to be set with wall_to_monotonic
1487 * @sleep: pointer to timespec to be set with time in suspend
1489 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1490 struct timespec *wtom, struct timespec *sleep)
1492 struct timekeeper *tk = &timekeeper;
1496 seq = read_seqbegin(&tk->lock);
1497 *xtim = tk_xtime(tk);
1498 *wtom = tk->wall_to_monotonic;
1499 *sleep = tk->total_sleep_time;
1500 } while (read_seqretry(&tk->lock, seq));
1503 #ifdef CONFIG_HIGH_RES_TIMERS
1505 * ktime_get_update_offsets - hrtimer helper
1506 * @offs_real: pointer to storage for monotonic -> realtime offset
1507 * @offs_boot: pointer to storage for monotonic -> boottime offset
1509 * Returns current monotonic time and updates the offsets
1510 * Called from hrtimer_interupt() or retrigger_next_event()
1512 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot)
1514 struct timekeeper *tk = &timekeeper;
1520 seq = read_seqbegin(&tk->lock);
1522 secs = tk->xtime_sec;
1523 nsecs = timekeeping_get_ns(tk);
1525 *offs_real = tk->offs_real;
1526 *offs_boot = tk->offs_boot;
1527 } while (read_seqretry(&tk->lock, seq));
1529 now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1530 now = ktime_sub(now, *offs_real);
1536 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1538 ktime_t ktime_get_monotonic_offset(void)
1540 struct timekeeper *tk = &timekeeper;
1542 struct timespec wtom;
1545 seq = read_seqbegin(&tk->lock);
1546 wtom = tk->wall_to_monotonic;
1547 } while (read_seqretry(&tk->lock, seq));
1549 return timespec_to_ktime(wtom);
1551 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1554 * xtime_update() - advances the timekeeping infrastructure
1555 * @ticks: number of ticks, that have elapsed since the last call.
1557 * Must be called with interrupts disabled.
1559 void xtime_update(unsigned long ticks)
1561 write_seqlock(&jiffies_lock);
1563 write_sequnlock(&jiffies_lock);