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>
25 #include <linux/compiler.h>
27 #include "tick-internal.h"
28 #include "ntp_internal.h"
29 #include "timekeeping_internal.h"
31 #define TK_CLEAR_NTP (1 << 0)
32 #define TK_MIRROR (1 << 1)
33 #define TK_CLOCK_WAS_SET (1 << 2)
36 * The most important data for readout fits into a single 64 byte
41 struct timekeeper timekeeper;
42 } tk_core ____cacheline_aligned;
44 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
45 static struct timekeeper shadow_timekeeper;
48 * struct tk_fast - NMI safe timekeeper
49 * @seq: Sequence counter for protecting updates. The lowest bit
50 * is the index for the tk_read_base array
51 * @base: tk_read_base array. Access is indexed by the lowest bit of
54 * See @update_fast_timekeeper() below.
58 struct tk_read_base base[2];
61 static struct tk_fast tk_fast_mono ____cacheline_aligned;
63 /* flag for if timekeeping is suspended */
64 int __read_mostly timekeeping_suspended;
66 /* Flag for if there is a persistent clock on this platform */
67 bool __read_mostly persistent_clock_exist = false;
69 static inline void tk_normalize_xtime(struct timekeeper *tk)
71 while (tk->tkr.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr.shift)) {
72 tk->tkr.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr.shift;
77 static inline struct timespec64 tk_xtime(struct timekeeper *tk)
81 ts.tv_sec = tk->xtime_sec;
82 ts.tv_nsec = (long)(tk->tkr.xtime_nsec >> tk->tkr.shift);
86 static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
88 tk->xtime_sec = ts->tv_sec;
89 tk->tkr.xtime_nsec = (u64)ts->tv_nsec << tk->tkr.shift;
92 static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
94 tk->xtime_sec += ts->tv_sec;
95 tk->tkr.xtime_nsec += (u64)ts->tv_nsec << tk->tkr.shift;
96 tk_normalize_xtime(tk);
99 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
101 struct timespec64 tmp;
104 * Verify consistency of: offset_real = -wall_to_monotonic
105 * before modifying anything
107 set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
108 -tk->wall_to_monotonic.tv_nsec);
109 WARN_ON_ONCE(tk->offs_real.tv64 != timespec64_to_ktime(tmp).tv64);
110 tk->wall_to_monotonic = wtm;
111 set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
112 tk->offs_real = timespec64_to_ktime(tmp);
113 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
116 static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
118 tk->offs_boot = ktime_add(tk->offs_boot, delta);
121 #ifdef CONFIG_DEBUG_TIMEKEEPING
122 static void timekeeping_check_update(struct timekeeper *tk, cycle_t offset)
125 cycle_t max_cycles = tk->tkr.clock->max_cycles;
126 const char *name = tk->tkr.clock->name;
128 if (offset > max_cycles) {
129 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
130 offset, name, max_cycles);
131 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
133 if (offset > (max_cycles >> 1)) {
134 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the the '%s' clock's 50%% safety margin (%lld)\n",
135 offset, name, max_cycles >> 1);
136 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
141 static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
143 cycle_t cycle_now, delta;
145 /* read clocksource */
146 cycle_now = tkr->read(tkr->clock);
148 /* calculate the delta since the last update_wall_time */
149 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
151 /* Cap delta value to the max_cycles values to avoid mult overflows */
152 if (unlikely(delta > tkr->clock->max_cycles))
153 delta = tkr->clock->max_cycles;
158 static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset)
161 static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
163 cycle_t cycle_now, delta;
165 /* read clocksource */
166 cycle_now = tkr->read(tkr->clock);
168 /* calculate the delta since the last update_wall_time */
169 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
176 * tk_setup_internals - Set up internals to use clocksource clock.
178 * @tk: The target timekeeper to setup.
179 * @clock: Pointer to clocksource.
181 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
182 * pair and interval request.
184 * Unless you're the timekeeping code, you should not be using this!
186 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
189 u64 tmp, ntpinterval;
190 struct clocksource *old_clock;
192 old_clock = tk->tkr.clock;
193 tk->tkr.clock = clock;
194 tk->tkr.read = clock->read;
195 tk->tkr.mask = clock->mask;
196 tk->tkr.cycle_last = tk->tkr.read(clock);
198 /* Do the ns -> cycle conversion first, using original mult */
199 tmp = NTP_INTERVAL_LENGTH;
200 tmp <<= clock->shift;
202 tmp += clock->mult/2;
203 do_div(tmp, clock->mult);
207 interval = (cycle_t) tmp;
208 tk->cycle_interval = interval;
210 /* Go back from cycles -> shifted ns */
211 tk->xtime_interval = (u64) interval * clock->mult;
212 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
214 ((u64) interval * clock->mult) >> clock->shift;
216 /* if changing clocks, convert xtime_nsec shift units */
218 int shift_change = clock->shift - old_clock->shift;
219 if (shift_change < 0)
220 tk->tkr.xtime_nsec >>= -shift_change;
222 tk->tkr.xtime_nsec <<= shift_change;
224 tk->tkr.shift = clock->shift;
227 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
228 tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
231 * The timekeeper keeps its own mult values for the currently
232 * active clocksource. These value will be adjusted via NTP
233 * to counteract clock drifting.
235 tk->tkr.mult = clock->mult;
236 tk->ntp_err_mult = 0;
239 /* Timekeeper helper functions. */
241 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
242 static u32 default_arch_gettimeoffset(void) { return 0; }
243 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
245 static inline u32 arch_gettimeoffset(void) { return 0; }
248 static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
253 delta = timekeeping_get_delta(tkr);
255 nsec = delta * tkr->mult + tkr->xtime_nsec;
258 /* If arch requires, add in get_arch_timeoffset() */
259 return nsec + arch_gettimeoffset();
262 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
264 struct clocksource *clock = tk->tkr.clock;
268 delta = timekeeping_get_delta(&tk->tkr);
270 /* convert delta to nanoseconds. */
271 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
273 /* If arch requires, add in get_arch_timeoffset() */
274 return nsec + arch_gettimeoffset();
278 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
279 * @tkr: Timekeeping readout base from which we take the update
281 * We want to use this from any context including NMI and tracing /
282 * instrumenting the timekeeping code itself.
284 * So we handle this differently than the other timekeeping accessor
285 * functions which retry when the sequence count has changed. The
288 * smp_wmb(); <- Ensure that the last base[1] update is visible
290 * smp_wmb(); <- Ensure that the seqcount update is visible
291 * update(tkf->base[0], tkr);
292 * smp_wmb(); <- Ensure that the base[0] update is visible
294 * smp_wmb(); <- Ensure that the seqcount update is visible
295 * update(tkf->base[1], tkr);
297 * The reader side does:
303 * now = now(tkf->base[idx]);
305 * } while (seq != tkf->seq)
307 * As long as we update base[0] readers are forced off to
308 * base[1]. Once base[0] is updated readers are redirected to base[0]
309 * and the base[1] update takes place.
311 * So if a NMI hits the update of base[0] then it will use base[1]
312 * which is still consistent. In the worst case this can result is a
313 * slightly wrong timestamp (a few nanoseconds). See
314 * @ktime_get_mono_fast_ns.
316 static void update_fast_timekeeper(struct tk_read_base *tkr)
318 struct tk_read_base *base = tk_fast_mono.base;
320 /* Force readers off to base[1] */
321 raw_write_seqcount_latch(&tk_fast_mono.seq);
324 memcpy(base, tkr, sizeof(*base));
326 /* Force readers back to base[0] */
327 raw_write_seqcount_latch(&tk_fast_mono.seq);
330 memcpy(base + 1, base, sizeof(*base));
334 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
336 * This timestamp is not guaranteed to be monotonic across an update.
337 * The timestamp is calculated by:
339 * now = base_mono + clock_delta * slope
341 * So if the update lowers the slope, readers who are forced to the
342 * not yet updated second array are still using the old steeper slope.
351 * |12345678---> reader order
357 * So reader 6 will observe time going backwards versus reader 5.
359 * While other CPUs are likely to be able observe that, the only way
360 * for a CPU local observation is when an NMI hits in the middle of
361 * the update. Timestamps taken from that NMI context might be ahead
362 * of the following timestamps. Callers need to be aware of that and
365 u64 notrace ktime_get_mono_fast_ns(void)
367 struct tk_read_base *tkr;
372 seq = raw_read_seqcount(&tk_fast_mono.seq);
373 tkr = tk_fast_mono.base + (seq & 0x01);
374 now = ktime_to_ns(tkr->base_mono) + timekeeping_get_ns(tkr);
376 } while (read_seqcount_retry(&tk_fast_mono.seq, seq));
379 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
381 /* Suspend-time cycles value for halted fast timekeeper. */
382 static cycle_t cycles_at_suspend;
384 static cycle_t dummy_clock_read(struct clocksource *cs)
386 return cycles_at_suspend;
390 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
391 * @tk: Timekeeper to snapshot.
393 * It generally is unsafe to access the clocksource after timekeeping has been
394 * suspended, so take a snapshot of the readout base of @tk and use it as the
395 * fast timekeeper's readout base while suspended. It will return the same
396 * number of cycles every time until timekeeping is resumed at which time the
397 * proper readout base for the fast timekeeper will be restored automatically.
399 static void halt_fast_timekeeper(struct timekeeper *tk)
401 static struct tk_read_base tkr_dummy;
402 struct tk_read_base *tkr = &tk->tkr;
404 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
405 cycles_at_suspend = tkr->read(tkr->clock);
406 tkr_dummy.read = dummy_clock_read;
407 update_fast_timekeeper(&tkr_dummy);
410 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
412 static inline void update_vsyscall(struct timekeeper *tk)
414 struct timespec xt, wm;
416 xt = timespec64_to_timespec(tk_xtime(tk));
417 wm = timespec64_to_timespec(tk->wall_to_monotonic);
418 update_vsyscall_old(&xt, &wm, tk->tkr.clock, tk->tkr.mult,
422 static inline void old_vsyscall_fixup(struct timekeeper *tk)
427 * Store only full nanoseconds into xtime_nsec after rounding
428 * it up and add the remainder to the error difference.
429 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
430 * by truncating the remainder in vsyscalls. However, it causes
431 * additional work to be done in timekeeping_adjust(). Once
432 * the vsyscall implementations are converted to use xtime_nsec
433 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
434 * users are removed, this can be killed.
436 remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1);
437 tk->tkr.xtime_nsec -= remainder;
438 tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift;
439 tk->ntp_error += remainder << tk->ntp_error_shift;
440 tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift;
443 #define old_vsyscall_fixup(tk)
446 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
448 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
450 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
454 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
456 int pvclock_gtod_register_notifier(struct notifier_block *nb)
458 struct timekeeper *tk = &tk_core.timekeeper;
462 raw_spin_lock_irqsave(&timekeeper_lock, flags);
463 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
464 update_pvclock_gtod(tk, true);
465 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
469 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
472 * pvclock_gtod_unregister_notifier - unregister a pvclock
473 * timedata update listener
475 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
480 raw_spin_lock_irqsave(&timekeeper_lock, flags);
481 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
482 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
486 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
489 * Update the ktime_t based scalar nsec members of the timekeeper
491 static inline void tk_update_ktime_data(struct timekeeper *tk)
497 * The xtime based monotonic readout is:
498 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
499 * The ktime based monotonic readout is:
500 * nsec = base_mono + now();
501 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
503 seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
504 nsec = (u32) tk->wall_to_monotonic.tv_nsec;
505 tk->tkr.base_mono = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
507 /* Update the monotonic raw base */
508 tk->base_raw = timespec64_to_ktime(tk->raw_time);
511 * The sum of the nanoseconds portions of xtime and
512 * wall_to_monotonic can be greater/equal one second. Take
513 * this into account before updating tk->ktime_sec.
515 nsec += (u32)(tk->tkr.xtime_nsec >> tk->tkr.shift);
516 if (nsec >= NSEC_PER_SEC)
518 tk->ktime_sec = seconds;
521 /* must hold timekeeper_lock */
522 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
524 if (action & TK_CLEAR_NTP) {
529 tk_update_ktime_data(tk);
532 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
534 if (action & TK_MIRROR)
535 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
536 sizeof(tk_core.timekeeper));
538 update_fast_timekeeper(&tk->tkr);
542 * timekeeping_forward_now - update clock to the current time
544 * Forward the current clock to update its state since the last call to
545 * update_wall_time(). This is useful before significant clock changes,
546 * as it avoids having to deal with this time offset explicitly.
548 static void timekeeping_forward_now(struct timekeeper *tk)
550 struct clocksource *clock = tk->tkr.clock;
551 cycle_t cycle_now, delta;
554 cycle_now = tk->tkr.read(clock);
555 delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask);
556 tk->tkr.cycle_last = cycle_now;
558 tk->tkr.xtime_nsec += delta * tk->tkr.mult;
560 /* If arch requires, add in get_arch_timeoffset() */
561 tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift;
563 tk_normalize_xtime(tk);
565 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
566 timespec64_add_ns(&tk->raw_time, nsec);
570 * __getnstimeofday64 - Returns the time of day in a timespec64.
571 * @ts: pointer to the timespec to be set
573 * Updates the time of day in the timespec.
574 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
576 int __getnstimeofday64(struct timespec64 *ts)
578 struct timekeeper *tk = &tk_core.timekeeper;
583 seq = read_seqcount_begin(&tk_core.seq);
585 ts->tv_sec = tk->xtime_sec;
586 nsecs = timekeeping_get_ns(&tk->tkr);
588 } while (read_seqcount_retry(&tk_core.seq, seq));
591 timespec64_add_ns(ts, nsecs);
594 * Do not bail out early, in case there were callers still using
595 * the value, even in the face of the WARN_ON.
597 if (unlikely(timekeeping_suspended))
601 EXPORT_SYMBOL(__getnstimeofday64);
604 * getnstimeofday64 - Returns the time of day in a timespec64.
605 * @ts: pointer to the timespec64 to be set
607 * Returns the time of day in a timespec64 (WARN if suspended).
609 void getnstimeofday64(struct timespec64 *ts)
611 WARN_ON(__getnstimeofday64(ts));
613 EXPORT_SYMBOL(getnstimeofday64);
615 ktime_t ktime_get(void)
617 struct timekeeper *tk = &tk_core.timekeeper;
622 WARN_ON(timekeeping_suspended);
625 seq = read_seqcount_begin(&tk_core.seq);
626 base = tk->tkr.base_mono;
627 nsecs = timekeeping_get_ns(&tk->tkr);
629 } while (read_seqcount_retry(&tk_core.seq, seq));
631 return ktime_add_ns(base, nsecs);
633 EXPORT_SYMBOL_GPL(ktime_get);
635 static ktime_t *offsets[TK_OFFS_MAX] = {
636 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
637 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
638 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
641 ktime_t ktime_get_with_offset(enum tk_offsets offs)
643 struct timekeeper *tk = &tk_core.timekeeper;
645 ktime_t base, *offset = offsets[offs];
648 WARN_ON(timekeeping_suspended);
651 seq = read_seqcount_begin(&tk_core.seq);
652 base = ktime_add(tk->tkr.base_mono, *offset);
653 nsecs = timekeeping_get_ns(&tk->tkr);
655 } while (read_seqcount_retry(&tk_core.seq, seq));
657 return ktime_add_ns(base, nsecs);
660 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
663 * ktime_mono_to_any() - convert mononotic time to any other time
664 * @tmono: time to convert.
665 * @offs: which offset to use
667 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
669 ktime_t *offset = offsets[offs];
674 seq = read_seqcount_begin(&tk_core.seq);
675 tconv = ktime_add(tmono, *offset);
676 } while (read_seqcount_retry(&tk_core.seq, seq));
680 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
683 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
685 ktime_t ktime_get_raw(void)
687 struct timekeeper *tk = &tk_core.timekeeper;
693 seq = read_seqcount_begin(&tk_core.seq);
695 nsecs = timekeeping_get_ns_raw(tk);
697 } while (read_seqcount_retry(&tk_core.seq, seq));
699 return ktime_add_ns(base, nsecs);
701 EXPORT_SYMBOL_GPL(ktime_get_raw);
704 * ktime_get_ts64 - get the monotonic clock in timespec64 format
705 * @ts: pointer to timespec variable
707 * The function calculates the monotonic clock from the realtime
708 * clock and the wall_to_monotonic offset and stores the result
709 * in normalized timespec64 format in the variable pointed to by @ts.
711 void ktime_get_ts64(struct timespec64 *ts)
713 struct timekeeper *tk = &tk_core.timekeeper;
714 struct timespec64 tomono;
718 WARN_ON(timekeeping_suspended);
721 seq = read_seqcount_begin(&tk_core.seq);
722 ts->tv_sec = tk->xtime_sec;
723 nsec = timekeeping_get_ns(&tk->tkr);
724 tomono = tk->wall_to_monotonic;
726 } while (read_seqcount_retry(&tk_core.seq, seq));
728 ts->tv_sec += tomono.tv_sec;
730 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
732 EXPORT_SYMBOL_GPL(ktime_get_ts64);
735 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
737 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
738 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
739 * works on both 32 and 64 bit systems. On 32 bit systems the readout
740 * covers ~136 years of uptime which should be enough to prevent
741 * premature wrap arounds.
743 time64_t ktime_get_seconds(void)
745 struct timekeeper *tk = &tk_core.timekeeper;
747 WARN_ON(timekeeping_suspended);
748 return tk->ktime_sec;
750 EXPORT_SYMBOL_GPL(ktime_get_seconds);
753 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
755 * Returns the wall clock seconds since 1970. This replaces the
756 * get_seconds() interface which is not y2038 safe on 32bit systems.
758 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
759 * 32bit systems the access must be protected with the sequence
760 * counter to provide "atomic" access to the 64bit tk->xtime_sec
763 time64_t ktime_get_real_seconds(void)
765 struct timekeeper *tk = &tk_core.timekeeper;
769 if (IS_ENABLED(CONFIG_64BIT))
770 return tk->xtime_sec;
773 seq = read_seqcount_begin(&tk_core.seq);
774 seconds = tk->xtime_sec;
776 } while (read_seqcount_retry(&tk_core.seq, seq));
780 EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
782 #ifdef CONFIG_NTP_PPS
785 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
786 * @ts_raw: pointer to the timespec to be set to raw monotonic time
787 * @ts_real: pointer to the timespec to be set to the time of day
789 * This function reads both the time of day and raw monotonic time at the
790 * same time atomically and stores the resulting timestamps in timespec
793 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
795 struct timekeeper *tk = &tk_core.timekeeper;
797 s64 nsecs_raw, nsecs_real;
799 WARN_ON_ONCE(timekeeping_suspended);
802 seq = read_seqcount_begin(&tk_core.seq);
804 *ts_raw = timespec64_to_timespec(tk->raw_time);
805 ts_real->tv_sec = tk->xtime_sec;
806 ts_real->tv_nsec = 0;
808 nsecs_raw = timekeeping_get_ns_raw(tk);
809 nsecs_real = timekeeping_get_ns(&tk->tkr);
811 } while (read_seqcount_retry(&tk_core.seq, seq));
813 timespec_add_ns(ts_raw, nsecs_raw);
814 timespec_add_ns(ts_real, nsecs_real);
816 EXPORT_SYMBOL(getnstime_raw_and_real);
818 #endif /* CONFIG_NTP_PPS */
821 * do_gettimeofday - Returns the time of day in a timeval
822 * @tv: pointer to the timeval to be set
824 * NOTE: Users should be converted to using getnstimeofday()
826 void do_gettimeofday(struct timeval *tv)
828 struct timespec64 now;
830 getnstimeofday64(&now);
831 tv->tv_sec = now.tv_sec;
832 tv->tv_usec = now.tv_nsec/1000;
834 EXPORT_SYMBOL(do_gettimeofday);
837 * do_settimeofday64 - Sets the time of day.
838 * @ts: pointer to the timespec64 variable containing the new time
840 * Sets the time of day to the new time and update NTP and notify hrtimers
842 int do_settimeofday64(const struct timespec64 *ts)
844 struct timekeeper *tk = &tk_core.timekeeper;
845 struct timespec64 ts_delta, xt;
848 if (!timespec64_valid_strict(ts))
851 raw_spin_lock_irqsave(&timekeeper_lock, flags);
852 write_seqcount_begin(&tk_core.seq);
854 timekeeping_forward_now(tk);
857 ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
858 ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
860 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
862 tk_set_xtime(tk, ts);
864 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
866 write_seqcount_end(&tk_core.seq);
867 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
869 /* signal hrtimers about time change */
874 EXPORT_SYMBOL(do_settimeofday64);
877 * timekeeping_inject_offset - Adds or subtracts from the current time.
878 * @tv: pointer to the timespec variable containing the offset
880 * Adds or subtracts an offset value from the current time.
882 int timekeeping_inject_offset(struct timespec *ts)
884 struct timekeeper *tk = &tk_core.timekeeper;
886 struct timespec64 ts64, tmp;
889 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
892 ts64 = timespec_to_timespec64(*ts);
894 raw_spin_lock_irqsave(&timekeeper_lock, flags);
895 write_seqcount_begin(&tk_core.seq);
897 timekeeping_forward_now(tk);
899 /* Make sure the proposed value is valid */
900 tmp = timespec64_add(tk_xtime(tk), ts64);
901 if (!timespec64_valid_strict(&tmp)) {
906 tk_xtime_add(tk, &ts64);
907 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64));
909 error: /* even if we error out, we forwarded the time, so call update */
910 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
912 write_seqcount_end(&tk_core.seq);
913 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
915 /* signal hrtimers about time change */
920 EXPORT_SYMBOL(timekeeping_inject_offset);
924 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
927 s32 timekeeping_get_tai_offset(void)
929 struct timekeeper *tk = &tk_core.timekeeper;
934 seq = read_seqcount_begin(&tk_core.seq);
935 ret = tk->tai_offset;
936 } while (read_seqcount_retry(&tk_core.seq, seq));
942 * __timekeeping_set_tai_offset - Lock free worker function
945 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
947 tk->tai_offset = tai_offset;
948 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
952 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
955 void timekeeping_set_tai_offset(s32 tai_offset)
957 struct timekeeper *tk = &tk_core.timekeeper;
960 raw_spin_lock_irqsave(&timekeeper_lock, flags);
961 write_seqcount_begin(&tk_core.seq);
962 __timekeeping_set_tai_offset(tk, tai_offset);
963 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
964 write_seqcount_end(&tk_core.seq);
965 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
970 * change_clocksource - Swaps clocksources if a new one is available
972 * Accumulates current time interval and initializes new clocksource
974 static int change_clocksource(void *data)
976 struct timekeeper *tk = &tk_core.timekeeper;
977 struct clocksource *new, *old;
980 new = (struct clocksource *) data;
982 raw_spin_lock_irqsave(&timekeeper_lock, flags);
983 write_seqcount_begin(&tk_core.seq);
985 timekeeping_forward_now(tk);
987 * If the cs is in module, get a module reference. Succeeds
988 * for built-in code (owner == NULL) as well.
990 if (try_module_get(new->owner)) {
991 if (!new->enable || new->enable(new) == 0) {
993 tk_setup_internals(tk, new);
996 module_put(old->owner);
998 module_put(new->owner);
1001 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1003 write_seqcount_end(&tk_core.seq);
1004 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1010 * timekeeping_notify - Install a new clock source
1011 * @clock: pointer to the clock source
1013 * This function is called from clocksource.c after a new, better clock
1014 * source has been registered. The caller holds the clocksource_mutex.
1016 int timekeeping_notify(struct clocksource *clock)
1018 struct timekeeper *tk = &tk_core.timekeeper;
1020 if (tk->tkr.clock == clock)
1022 stop_machine(change_clocksource, clock, NULL);
1023 tick_clock_notify();
1024 return tk->tkr.clock == clock ? 0 : -1;
1028 * getrawmonotonic64 - Returns the raw monotonic time in a timespec
1029 * @ts: pointer to the timespec64 to be set
1031 * Returns the raw monotonic time (completely un-modified by ntp)
1033 void getrawmonotonic64(struct timespec64 *ts)
1035 struct timekeeper *tk = &tk_core.timekeeper;
1036 struct timespec64 ts64;
1041 seq = read_seqcount_begin(&tk_core.seq);
1042 nsecs = timekeeping_get_ns_raw(tk);
1043 ts64 = tk->raw_time;
1045 } while (read_seqcount_retry(&tk_core.seq, seq));
1047 timespec64_add_ns(&ts64, nsecs);
1050 EXPORT_SYMBOL(getrawmonotonic64);
1054 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1056 int timekeeping_valid_for_hres(void)
1058 struct timekeeper *tk = &tk_core.timekeeper;
1063 seq = read_seqcount_begin(&tk_core.seq);
1065 ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1067 } while (read_seqcount_retry(&tk_core.seq, seq));
1073 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1075 u64 timekeeping_max_deferment(void)
1077 struct timekeeper *tk = &tk_core.timekeeper;
1082 seq = read_seqcount_begin(&tk_core.seq);
1084 ret = tk->tkr.clock->max_idle_ns;
1086 } while (read_seqcount_retry(&tk_core.seq, seq));
1092 * read_persistent_clock - Return time from the persistent clock.
1094 * Weak dummy function for arches that do not yet support it.
1095 * Reads the time from the battery backed persistent clock.
1096 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1098 * XXX - Do be sure to remove it once all arches implement it.
1100 void __weak read_persistent_clock(struct timespec *ts)
1107 * read_boot_clock - Return time of the system start.
1109 * Weak dummy function for arches that do not yet support it.
1110 * Function to read the exact time the system has been started.
1111 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1113 * XXX - Do be sure to remove it once all arches implement it.
1115 void __weak read_boot_clock(struct timespec *ts)
1122 * timekeeping_init - Initializes the clocksource and common timekeeping values
1124 void __init timekeeping_init(void)
1126 struct timekeeper *tk = &tk_core.timekeeper;
1127 struct clocksource *clock;
1128 unsigned long flags;
1129 struct timespec64 now, boot, tmp;
1132 read_persistent_clock(&ts);
1133 now = timespec_to_timespec64(ts);
1134 if (!timespec64_valid_strict(&now)) {
1135 pr_warn("WARNING: Persistent clock returned invalid value!\n"
1136 " Check your CMOS/BIOS settings.\n");
1139 } else if (now.tv_sec || now.tv_nsec)
1140 persistent_clock_exist = true;
1142 read_boot_clock(&ts);
1143 boot = timespec_to_timespec64(ts);
1144 if (!timespec64_valid_strict(&boot)) {
1145 pr_warn("WARNING: Boot clock returned invalid value!\n"
1146 " Check your CMOS/BIOS settings.\n");
1151 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1152 write_seqcount_begin(&tk_core.seq);
1155 clock = clocksource_default_clock();
1157 clock->enable(clock);
1158 tk_setup_internals(tk, clock);
1160 tk_set_xtime(tk, &now);
1161 tk->raw_time.tv_sec = 0;
1162 tk->raw_time.tv_nsec = 0;
1163 tk->base_raw.tv64 = 0;
1164 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
1165 boot = tk_xtime(tk);
1167 set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec);
1168 tk_set_wall_to_mono(tk, tmp);
1170 timekeeping_update(tk, TK_MIRROR);
1172 write_seqcount_end(&tk_core.seq);
1173 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1176 /* time in seconds when suspend began */
1177 static struct timespec64 timekeeping_suspend_time;
1180 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1181 * @delta: pointer to a timespec delta value
1183 * Takes a timespec offset measuring a suspend interval and properly
1184 * adds the sleep offset to the timekeeping variables.
1186 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1187 struct timespec64 *delta)
1189 if (!timespec64_valid_strict(delta)) {
1190 printk_deferred(KERN_WARNING
1191 "__timekeeping_inject_sleeptime: Invalid "
1192 "sleep delta value!\n");
1195 tk_xtime_add(tk, delta);
1196 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1197 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1198 tk_debug_account_sleep_time(delta);
1202 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1203 * @delta: pointer to a timespec64 delta value
1205 * This hook is for architectures that cannot support read_persistent_clock
1206 * because their RTC/persistent clock is only accessible when irqs are enabled.
1208 * This function should only be called by rtc_resume(), and allows
1209 * a suspend offset to be injected into the timekeeping values.
1211 void timekeeping_inject_sleeptime64(struct timespec64 *delta)
1213 struct timekeeper *tk = &tk_core.timekeeper;
1214 unsigned long flags;
1217 * Make sure we don't set the clock twice, as timekeeping_resume()
1220 if (has_persistent_clock())
1223 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1224 write_seqcount_begin(&tk_core.seq);
1226 timekeeping_forward_now(tk);
1228 __timekeeping_inject_sleeptime(tk, delta);
1230 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1232 write_seqcount_end(&tk_core.seq);
1233 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1235 /* signal hrtimers about time change */
1240 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1242 * This is for the generic clocksource timekeeping.
1243 * xtime/wall_to_monotonic/jiffies/etc are
1244 * still managed by arch specific suspend/resume code.
1246 void timekeeping_resume(void)
1248 struct timekeeper *tk = &tk_core.timekeeper;
1249 struct clocksource *clock = tk->tkr.clock;
1250 unsigned long flags;
1251 struct timespec64 ts_new, ts_delta;
1252 struct timespec tmp;
1253 cycle_t cycle_now, cycle_delta;
1254 bool suspendtime_found = false;
1256 read_persistent_clock(&tmp);
1257 ts_new = timespec_to_timespec64(tmp);
1259 clockevents_resume();
1260 clocksource_resume();
1262 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1263 write_seqcount_begin(&tk_core.seq);
1266 * After system resumes, we need to calculate the suspended time and
1267 * compensate it for the OS time. There are 3 sources that could be
1268 * used: Nonstop clocksource during suspend, persistent clock and rtc
1271 * One specific platform may have 1 or 2 or all of them, and the
1272 * preference will be:
1273 * suspend-nonstop clocksource -> persistent clock -> rtc
1274 * The less preferred source will only be tried if there is no better
1275 * usable source. The rtc part is handled separately in rtc core code.
1277 cycle_now = tk->tkr.read(clock);
1278 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
1279 cycle_now > tk->tkr.cycle_last) {
1280 u64 num, max = ULLONG_MAX;
1281 u32 mult = clock->mult;
1282 u32 shift = clock->shift;
1285 cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last,
1289 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1290 * suspended time is too long. In that case we need do the
1291 * 64 bits math carefully
1294 if (cycle_delta > max) {
1295 num = div64_u64(cycle_delta, max);
1296 nsec = (((u64) max * mult) >> shift) * num;
1297 cycle_delta -= num * max;
1299 nsec += ((u64) cycle_delta * mult) >> shift;
1301 ts_delta = ns_to_timespec64(nsec);
1302 suspendtime_found = true;
1303 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1304 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1305 suspendtime_found = true;
1308 if (suspendtime_found)
1309 __timekeeping_inject_sleeptime(tk, &ts_delta);
1311 /* Re-base the last cycle value */
1312 tk->tkr.cycle_last = cycle_now;
1314 timekeeping_suspended = 0;
1315 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1316 write_seqcount_end(&tk_core.seq);
1317 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1319 touch_softlockup_watchdog();
1321 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
1323 /* Resume hrtimers */
1327 int timekeeping_suspend(void)
1329 struct timekeeper *tk = &tk_core.timekeeper;
1330 unsigned long flags;
1331 struct timespec64 delta, delta_delta;
1332 static struct timespec64 old_delta;
1333 struct timespec tmp;
1335 read_persistent_clock(&tmp);
1336 timekeeping_suspend_time = timespec_to_timespec64(tmp);
1339 * On some systems the persistent_clock can not be detected at
1340 * timekeeping_init by its return value, so if we see a valid
1341 * value returned, update the persistent_clock_exists flag.
1343 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1344 persistent_clock_exist = true;
1346 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1347 write_seqcount_begin(&tk_core.seq);
1348 timekeeping_forward_now(tk);
1349 timekeeping_suspended = 1;
1352 * To avoid drift caused by repeated suspend/resumes,
1353 * which each can add ~1 second drift error,
1354 * try to compensate so the difference in system time
1355 * and persistent_clock time stays close to constant.
1357 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1358 delta_delta = timespec64_sub(delta, old_delta);
1359 if (abs(delta_delta.tv_sec) >= 2) {
1361 * if delta_delta is too large, assume time correction
1362 * has occured and set old_delta to the current delta.
1366 /* Otherwise try to adjust old_system to compensate */
1367 timekeeping_suspend_time =
1368 timespec64_add(timekeeping_suspend_time, delta_delta);
1371 timekeeping_update(tk, TK_MIRROR);
1372 halt_fast_timekeeper(tk);
1373 write_seqcount_end(&tk_core.seq);
1374 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1376 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1377 clocksource_suspend();
1378 clockevents_suspend();
1383 /* sysfs resume/suspend bits for timekeeping */
1384 static struct syscore_ops timekeeping_syscore_ops = {
1385 .resume = timekeeping_resume,
1386 .suspend = timekeeping_suspend,
1389 static int __init timekeeping_init_ops(void)
1391 register_syscore_ops(&timekeeping_syscore_ops);
1394 device_initcall(timekeeping_init_ops);
1397 * Apply a multiplier adjustment to the timekeeper
1399 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1404 s64 interval = tk->cycle_interval;
1408 mult_adj = -mult_adj;
1409 interval = -interval;
1412 mult_adj <<= adj_scale;
1413 interval <<= adj_scale;
1414 offset <<= adj_scale;
1417 * So the following can be confusing.
1419 * To keep things simple, lets assume mult_adj == 1 for now.
1421 * When mult_adj != 1, remember that the interval and offset values
1422 * have been appropriately scaled so the math is the same.
1424 * The basic idea here is that we're increasing the multiplier
1425 * by one, this causes the xtime_interval to be incremented by
1426 * one cycle_interval. This is because:
1427 * xtime_interval = cycle_interval * mult
1428 * So if mult is being incremented by one:
1429 * xtime_interval = cycle_interval * (mult + 1)
1431 * xtime_interval = (cycle_interval * mult) + cycle_interval
1432 * Which can be shortened to:
1433 * xtime_interval += cycle_interval
1435 * So offset stores the non-accumulated cycles. Thus the current
1436 * time (in shifted nanoseconds) is:
1437 * now = (offset * adj) + xtime_nsec
1438 * Now, even though we're adjusting the clock frequency, we have
1439 * to keep time consistent. In other words, we can't jump back
1440 * in time, and we also want to avoid jumping forward in time.
1442 * So given the same offset value, we need the time to be the same
1443 * both before and after the freq adjustment.
1444 * now = (offset * adj_1) + xtime_nsec_1
1445 * now = (offset * adj_2) + xtime_nsec_2
1447 * (offset * adj_1) + xtime_nsec_1 =
1448 * (offset * adj_2) + xtime_nsec_2
1452 * (offset * adj_1) + xtime_nsec_1 =
1453 * (offset * (adj_1+1)) + xtime_nsec_2
1454 * (offset * adj_1) + xtime_nsec_1 =
1455 * (offset * adj_1) + offset + xtime_nsec_2
1456 * Canceling the sides:
1457 * xtime_nsec_1 = offset + xtime_nsec_2
1459 * xtime_nsec_2 = xtime_nsec_1 - offset
1460 * Which simplfies to:
1461 * xtime_nsec -= offset
1463 * XXX - TODO: Doc ntp_error calculation.
1465 if ((mult_adj > 0) && (tk->tkr.mult + mult_adj < mult_adj)) {
1466 /* NTP adjustment caused clocksource mult overflow */
1471 tk->tkr.mult += mult_adj;
1472 tk->xtime_interval += interval;
1473 tk->tkr.xtime_nsec -= offset;
1474 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1478 * Calculate the multiplier adjustment needed to match the frequency
1481 static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
1484 s64 interval = tk->cycle_interval;
1485 s64 xinterval = tk->xtime_interval;
1490 /* Remove any current error adj from freq calculation */
1491 if (tk->ntp_err_mult)
1492 xinterval -= tk->cycle_interval;
1494 tk->ntp_tick = ntp_tick_length();
1496 /* Calculate current error per tick */
1497 tick_error = ntp_tick_length() >> tk->ntp_error_shift;
1498 tick_error -= (xinterval + tk->xtime_remainder);
1500 /* Don't worry about correcting it if its small */
1501 if (likely((tick_error >= 0) && (tick_error <= interval)))
1504 /* preserve the direction of correction */
1505 negative = (tick_error < 0);
1507 /* Sort out the magnitude of the correction */
1508 tick_error = abs(tick_error);
1509 for (adj = 0; tick_error > interval; adj++)
1512 /* scale the corrections */
1513 timekeeping_apply_adjustment(tk, offset, negative, adj);
1517 * Adjust the timekeeper's multiplier to the correct frequency
1518 * and also to reduce the accumulated error value.
1520 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1522 /* Correct for the current frequency error */
1523 timekeeping_freqadjust(tk, offset);
1525 /* Next make a small adjustment to fix any cumulative error */
1526 if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
1527 tk->ntp_err_mult = 1;
1528 timekeeping_apply_adjustment(tk, offset, 0, 0);
1529 } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
1530 /* Undo any existing error adjustment */
1531 timekeeping_apply_adjustment(tk, offset, 1, 0);
1532 tk->ntp_err_mult = 0;
1535 if (unlikely(tk->tkr.clock->maxadj &&
1536 (abs(tk->tkr.mult - tk->tkr.clock->mult)
1537 > tk->tkr.clock->maxadj))) {
1538 printk_once(KERN_WARNING
1539 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1540 tk->tkr.clock->name, (long)tk->tkr.mult,
1541 (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
1545 * It may be possible that when we entered this function, xtime_nsec
1546 * was very small. Further, if we're slightly speeding the clocksource
1547 * in the code above, its possible the required corrective factor to
1548 * xtime_nsec could cause it to underflow.
1550 * Now, since we already accumulated the second, cannot simply roll
1551 * the accumulated second back, since the NTP subsystem has been
1552 * notified via second_overflow. So instead we push xtime_nsec forward
1553 * by the amount we underflowed, and add that amount into the error.
1555 * We'll correct this error next time through this function, when
1556 * xtime_nsec is not as small.
1558 if (unlikely((s64)tk->tkr.xtime_nsec < 0)) {
1559 s64 neg = -(s64)tk->tkr.xtime_nsec;
1560 tk->tkr.xtime_nsec = 0;
1561 tk->ntp_error += neg << tk->ntp_error_shift;
1566 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1568 * Helper function that accumulates a the nsecs greater then a second
1569 * from the xtime_nsec field to the xtime_secs field.
1570 * It also calls into the NTP code to handle leapsecond processing.
1573 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1575 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift;
1576 unsigned int clock_set = 0;
1578 while (tk->tkr.xtime_nsec >= nsecps) {
1581 tk->tkr.xtime_nsec -= nsecps;
1584 /* Figure out if its a leap sec and apply if needed */
1585 leap = second_overflow(tk->xtime_sec);
1586 if (unlikely(leap)) {
1587 struct timespec64 ts;
1589 tk->xtime_sec += leap;
1593 tk_set_wall_to_mono(tk,
1594 timespec64_sub(tk->wall_to_monotonic, ts));
1596 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1598 clock_set = TK_CLOCK_WAS_SET;
1605 * logarithmic_accumulation - shifted accumulation of cycles
1607 * This functions accumulates a shifted interval of cycles into
1608 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1611 * Returns the unconsumed cycles.
1613 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1615 unsigned int *clock_set)
1617 cycle_t interval = tk->cycle_interval << shift;
1620 /* If the offset is smaller then a shifted interval, do nothing */
1621 if (offset < interval)
1624 /* Accumulate one shifted interval */
1626 tk->tkr.cycle_last += interval;
1628 tk->tkr.xtime_nsec += tk->xtime_interval << shift;
1629 *clock_set |= accumulate_nsecs_to_secs(tk);
1631 /* Accumulate raw time */
1632 raw_nsecs = (u64)tk->raw_interval << shift;
1633 raw_nsecs += tk->raw_time.tv_nsec;
1634 if (raw_nsecs >= NSEC_PER_SEC) {
1635 u64 raw_secs = raw_nsecs;
1636 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1637 tk->raw_time.tv_sec += raw_secs;
1639 tk->raw_time.tv_nsec = raw_nsecs;
1641 /* Accumulate error between NTP and clock interval */
1642 tk->ntp_error += tk->ntp_tick << shift;
1643 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1644 (tk->ntp_error_shift + shift);
1650 * update_wall_time - Uses the current clocksource to increment the wall time
1653 void update_wall_time(void)
1655 struct timekeeper *real_tk = &tk_core.timekeeper;
1656 struct timekeeper *tk = &shadow_timekeeper;
1658 int shift = 0, maxshift;
1659 unsigned int clock_set = 0;
1660 unsigned long flags;
1662 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1664 /* Make sure we're fully resumed: */
1665 if (unlikely(timekeeping_suspended))
1668 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1669 offset = real_tk->cycle_interval;
1671 offset = clocksource_delta(tk->tkr.read(tk->tkr.clock),
1672 tk->tkr.cycle_last, tk->tkr.mask);
1675 /* Check if there's really nothing to do */
1676 if (offset < real_tk->cycle_interval)
1679 /* Do some additional sanity checking */
1680 timekeeping_check_update(real_tk, offset);
1683 * With NO_HZ we may have to accumulate many cycle_intervals
1684 * (think "ticks") worth of time at once. To do this efficiently,
1685 * we calculate the largest doubling multiple of cycle_intervals
1686 * that is smaller than the offset. We then accumulate that
1687 * chunk in one go, and then try to consume the next smaller
1690 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1691 shift = max(0, shift);
1692 /* Bound shift to one less than what overflows tick_length */
1693 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1694 shift = min(shift, maxshift);
1695 while (offset >= tk->cycle_interval) {
1696 offset = logarithmic_accumulation(tk, offset, shift,
1698 if (offset < tk->cycle_interval<<shift)
1702 /* correct the clock when NTP error is too big */
1703 timekeeping_adjust(tk, offset);
1706 * XXX This can be killed once everyone converts
1707 * to the new update_vsyscall.
1709 old_vsyscall_fixup(tk);
1712 * Finally, make sure that after the rounding
1713 * xtime_nsec isn't larger than NSEC_PER_SEC
1715 clock_set |= accumulate_nsecs_to_secs(tk);
1717 write_seqcount_begin(&tk_core.seq);
1719 * Update the real timekeeper.
1721 * We could avoid this memcpy by switching pointers, but that
1722 * requires changes to all other timekeeper usage sites as
1723 * well, i.e. move the timekeeper pointer getter into the
1724 * spinlocked/seqcount protected sections. And we trade this
1725 * memcpy under the tk_core.seq against one before we start
1728 memcpy(real_tk, tk, sizeof(*tk));
1729 timekeeping_update(real_tk, clock_set);
1730 write_seqcount_end(&tk_core.seq);
1732 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1734 /* Have to call _delayed version, since in irq context*/
1735 clock_was_set_delayed();
1739 * getboottime64 - Return the real time of system boot.
1740 * @ts: pointer to the timespec64 to be set
1742 * Returns the wall-time of boot in a timespec64.
1744 * This is based on the wall_to_monotonic offset and the total suspend
1745 * time. Calls to settimeofday will affect the value returned (which
1746 * basically means that however wrong your real time clock is at boot time,
1747 * you get the right time here).
1749 void getboottime64(struct timespec64 *ts)
1751 struct timekeeper *tk = &tk_core.timekeeper;
1752 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
1754 *ts = ktime_to_timespec64(t);
1756 EXPORT_SYMBOL_GPL(getboottime64);
1758 unsigned long get_seconds(void)
1760 struct timekeeper *tk = &tk_core.timekeeper;
1762 return tk->xtime_sec;
1764 EXPORT_SYMBOL(get_seconds);
1766 struct timespec __current_kernel_time(void)
1768 struct timekeeper *tk = &tk_core.timekeeper;
1770 return timespec64_to_timespec(tk_xtime(tk));
1773 struct timespec current_kernel_time(void)
1775 struct timekeeper *tk = &tk_core.timekeeper;
1776 struct timespec64 now;
1780 seq = read_seqcount_begin(&tk_core.seq);
1783 } while (read_seqcount_retry(&tk_core.seq, seq));
1785 return timespec64_to_timespec(now);
1787 EXPORT_SYMBOL(current_kernel_time);
1789 struct timespec64 get_monotonic_coarse64(void)
1791 struct timekeeper *tk = &tk_core.timekeeper;
1792 struct timespec64 now, mono;
1796 seq = read_seqcount_begin(&tk_core.seq);
1799 mono = tk->wall_to_monotonic;
1800 } while (read_seqcount_retry(&tk_core.seq, seq));
1802 set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec,
1803 now.tv_nsec + mono.tv_nsec);
1809 * Must hold jiffies_lock
1811 void do_timer(unsigned long ticks)
1813 jiffies_64 += ticks;
1814 calc_global_load(ticks);
1818 * ktime_get_update_offsets_tick - hrtimer helper
1819 * @offs_real: pointer to storage for monotonic -> realtime offset
1820 * @offs_boot: pointer to storage for monotonic -> boottime offset
1821 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1823 * Returns monotonic time at last tick and various offsets
1825 ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot,
1828 struct timekeeper *tk = &tk_core.timekeeper;
1834 seq = read_seqcount_begin(&tk_core.seq);
1836 base = tk->tkr.base_mono;
1837 nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift;
1839 *offs_real = tk->offs_real;
1840 *offs_boot = tk->offs_boot;
1841 *offs_tai = tk->offs_tai;
1842 } while (read_seqcount_retry(&tk_core.seq, seq));
1844 return ktime_add_ns(base, nsecs);
1847 #ifdef CONFIG_HIGH_RES_TIMERS
1849 * ktime_get_update_offsets_now - hrtimer helper
1850 * @offs_real: pointer to storage for monotonic -> realtime offset
1851 * @offs_boot: pointer to storage for monotonic -> boottime offset
1852 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1854 * Returns current monotonic time and updates the offsets
1855 * Called from hrtimer_interrupt() or retrigger_next_event()
1857 ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot,
1860 struct timekeeper *tk = &tk_core.timekeeper;
1866 seq = read_seqcount_begin(&tk_core.seq);
1868 base = tk->tkr.base_mono;
1869 nsecs = timekeeping_get_ns(&tk->tkr);
1871 *offs_real = tk->offs_real;
1872 *offs_boot = tk->offs_boot;
1873 *offs_tai = tk->offs_tai;
1874 } while (read_seqcount_retry(&tk_core.seq, seq));
1876 return ktime_add_ns(base, nsecs);
1881 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1883 int do_adjtimex(struct timex *txc)
1885 struct timekeeper *tk = &tk_core.timekeeper;
1886 unsigned long flags;
1887 struct timespec64 ts;
1891 /* Validate the data before disabling interrupts */
1892 ret = ntp_validate_timex(txc);
1896 if (txc->modes & ADJ_SETOFFSET) {
1897 struct timespec delta;
1898 delta.tv_sec = txc->time.tv_sec;
1899 delta.tv_nsec = txc->time.tv_usec;
1900 if (!(txc->modes & ADJ_NANO))
1901 delta.tv_nsec *= 1000;
1902 ret = timekeeping_inject_offset(&delta);
1907 getnstimeofday64(&ts);
1909 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1910 write_seqcount_begin(&tk_core.seq);
1912 orig_tai = tai = tk->tai_offset;
1913 ret = __do_adjtimex(txc, &ts, &tai);
1915 if (tai != orig_tai) {
1916 __timekeeping_set_tai_offset(tk, tai);
1917 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1919 write_seqcount_end(&tk_core.seq);
1920 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1922 if (tai != orig_tai)
1925 ntp_notify_cmos_timer();
1930 #ifdef CONFIG_NTP_PPS
1932 * hardpps() - Accessor function to NTP __hardpps function
1934 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1936 unsigned long flags;
1938 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1939 write_seqcount_begin(&tk_core.seq);
1941 __hardpps(phase_ts, raw_ts);
1943 write_seqcount_end(&tk_core.seq);
1944 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1946 EXPORT_SYMBOL(hardpps);
1950 * xtime_update() - advances the timekeeping infrastructure
1951 * @ticks: number of ticks, that have elapsed since the last call.
1953 * Must be called with interrupts disabled.
1955 void xtime_update(unsigned long ticks)
1957 write_seqlock(&jiffies_lock);
1959 write_sequnlock(&jiffies_lock);