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);
152 * Try to catch underflows by checking if we are seeing small
153 * mask-relative negative values.
155 if (unlikely((~delta & tkr->mask) < (tkr->mask >> 3)))
158 /* Cap delta value to the max_cycles values to avoid mult overflows */
159 if (unlikely(delta > tkr->clock->max_cycles))
160 delta = tkr->clock->max_cycles;
165 static inline void timekeeping_check_update(struct timekeeper *tk, cycle_t offset)
168 static inline cycle_t timekeeping_get_delta(struct tk_read_base *tkr)
170 cycle_t cycle_now, delta;
172 /* read clocksource */
173 cycle_now = tkr->read(tkr->clock);
175 /* calculate the delta since the last update_wall_time */
176 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
183 * tk_setup_internals - Set up internals to use clocksource clock.
185 * @tk: The target timekeeper to setup.
186 * @clock: Pointer to clocksource.
188 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
189 * pair and interval request.
191 * Unless you're the timekeeping code, you should not be using this!
193 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
196 u64 tmp, ntpinterval;
197 struct clocksource *old_clock;
199 old_clock = tk->tkr.clock;
200 tk->tkr.clock = clock;
201 tk->tkr.read = clock->read;
202 tk->tkr.mask = clock->mask;
203 tk->tkr.cycle_last = tk->tkr.read(clock);
205 /* Do the ns -> cycle conversion first, using original mult */
206 tmp = NTP_INTERVAL_LENGTH;
207 tmp <<= clock->shift;
209 tmp += clock->mult/2;
210 do_div(tmp, clock->mult);
214 interval = (cycle_t) tmp;
215 tk->cycle_interval = interval;
217 /* Go back from cycles -> shifted ns */
218 tk->xtime_interval = (u64) interval * clock->mult;
219 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
221 ((u64) interval * clock->mult) >> clock->shift;
223 /* if changing clocks, convert xtime_nsec shift units */
225 int shift_change = clock->shift - old_clock->shift;
226 if (shift_change < 0)
227 tk->tkr.xtime_nsec >>= -shift_change;
229 tk->tkr.xtime_nsec <<= shift_change;
231 tk->tkr.shift = clock->shift;
234 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
235 tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
238 * The timekeeper keeps its own mult values for the currently
239 * active clocksource. These value will be adjusted via NTP
240 * to counteract clock drifting.
242 tk->tkr.mult = clock->mult;
243 tk->ntp_err_mult = 0;
246 /* Timekeeper helper functions. */
248 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
249 static u32 default_arch_gettimeoffset(void) { return 0; }
250 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
252 static inline u32 arch_gettimeoffset(void) { return 0; }
255 static inline s64 timekeeping_get_ns(struct tk_read_base *tkr)
260 delta = timekeeping_get_delta(tkr);
262 nsec = delta * tkr->mult + tkr->xtime_nsec;
265 /* If arch requires, add in get_arch_timeoffset() */
266 return nsec + arch_gettimeoffset();
269 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
271 struct clocksource *clock = tk->tkr.clock;
275 delta = timekeeping_get_delta(&tk->tkr);
277 /* convert delta to nanoseconds. */
278 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
280 /* If arch requires, add in get_arch_timeoffset() */
281 return nsec + arch_gettimeoffset();
285 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
286 * @tkr: Timekeeping readout base from which we take the update
288 * We want to use this from any context including NMI and tracing /
289 * instrumenting the timekeeping code itself.
291 * So we handle this differently than the other timekeeping accessor
292 * functions which retry when the sequence count has changed. The
295 * smp_wmb(); <- Ensure that the last base[1] update is visible
297 * smp_wmb(); <- Ensure that the seqcount update is visible
298 * update(tkf->base[0], tkr);
299 * smp_wmb(); <- Ensure that the base[0] update is visible
301 * smp_wmb(); <- Ensure that the seqcount update is visible
302 * update(tkf->base[1], tkr);
304 * The reader side does:
310 * now = now(tkf->base[idx]);
312 * } while (seq != tkf->seq)
314 * As long as we update base[0] readers are forced off to
315 * base[1]. Once base[0] is updated readers are redirected to base[0]
316 * and the base[1] update takes place.
318 * So if a NMI hits the update of base[0] then it will use base[1]
319 * which is still consistent. In the worst case this can result is a
320 * slightly wrong timestamp (a few nanoseconds). See
321 * @ktime_get_mono_fast_ns.
323 static void update_fast_timekeeper(struct tk_read_base *tkr)
325 struct tk_read_base *base = tk_fast_mono.base;
327 /* Force readers off to base[1] */
328 raw_write_seqcount_latch(&tk_fast_mono.seq);
331 memcpy(base, tkr, sizeof(*base));
333 /* Force readers back to base[0] */
334 raw_write_seqcount_latch(&tk_fast_mono.seq);
337 memcpy(base + 1, base, sizeof(*base));
341 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
343 * This timestamp is not guaranteed to be monotonic across an update.
344 * The timestamp is calculated by:
346 * now = base_mono + clock_delta * slope
348 * So if the update lowers the slope, readers who are forced to the
349 * not yet updated second array are still using the old steeper slope.
358 * |12345678---> reader order
364 * So reader 6 will observe time going backwards versus reader 5.
366 * While other CPUs are likely to be able observe that, the only way
367 * for a CPU local observation is when an NMI hits in the middle of
368 * the update. Timestamps taken from that NMI context might be ahead
369 * of the following timestamps. Callers need to be aware of that and
372 u64 notrace ktime_get_mono_fast_ns(void)
374 struct tk_read_base *tkr;
379 seq = raw_read_seqcount(&tk_fast_mono.seq);
380 tkr = tk_fast_mono.base + (seq & 0x01);
381 now = ktime_to_ns(tkr->base_mono) + timekeeping_get_ns(tkr);
383 } while (read_seqcount_retry(&tk_fast_mono.seq, seq));
386 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
388 /* Suspend-time cycles value for halted fast timekeeper. */
389 static cycle_t cycles_at_suspend;
391 static cycle_t dummy_clock_read(struct clocksource *cs)
393 return cycles_at_suspend;
397 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
398 * @tk: Timekeeper to snapshot.
400 * It generally is unsafe to access the clocksource after timekeeping has been
401 * suspended, so take a snapshot of the readout base of @tk and use it as the
402 * fast timekeeper's readout base while suspended. It will return the same
403 * number of cycles every time until timekeeping is resumed at which time the
404 * proper readout base for the fast timekeeper will be restored automatically.
406 static void halt_fast_timekeeper(struct timekeeper *tk)
408 static struct tk_read_base tkr_dummy;
409 struct tk_read_base *tkr = &tk->tkr;
411 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
412 cycles_at_suspend = tkr->read(tkr->clock);
413 tkr_dummy.read = dummy_clock_read;
414 update_fast_timekeeper(&tkr_dummy);
417 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
419 static inline void update_vsyscall(struct timekeeper *tk)
421 struct timespec xt, wm;
423 xt = timespec64_to_timespec(tk_xtime(tk));
424 wm = timespec64_to_timespec(tk->wall_to_monotonic);
425 update_vsyscall_old(&xt, &wm, tk->tkr.clock, tk->tkr.mult,
429 static inline void old_vsyscall_fixup(struct timekeeper *tk)
434 * Store only full nanoseconds into xtime_nsec after rounding
435 * it up and add the remainder to the error difference.
436 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
437 * by truncating the remainder in vsyscalls. However, it causes
438 * additional work to be done in timekeeping_adjust(). Once
439 * the vsyscall implementations are converted to use xtime_nsec
440 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
441 * users are removed, this can be killed.
443 remainder = tk->tkr.xtime_nsec & ((1ULL << tk->tkr.shift) - 1);
444 tk->tkr.xtime_nsec -= remainder;
445 tk->tkr.xtime_nsec += 1ULL << tk->tkr.shift;
446 tk->ntp_error += remainder << tk->ntp_error_shift;
447 tk->ntp_error -= (1ULL << tk->tkr.shift) << tk->ntp_error_shift;
450 #define old_vsyscall_fixup(tk)
453 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
455 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
457 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
461 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
463 int pvclock_gtod_register_notifier(struct notifier_block *nb)
465 struct timekeeper *tk = &tk_core.timekeeper;
469 raw_spin_lock_irqsave(&timekeeper_lock, flags);
470 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
471 update_pvclock_gtod(tk, true);
472 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
476 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
479 * pvclock_gtod_unregister_notifier - unregister a pvclock
480 * timedata update listener
482 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
487 raw_spin_lock_irqsave(&timekeeper_lock, flags);
488 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
489 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
493 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
496 * Update the ktime_t based scalar nsec members of the timekeeper
498 static inline void tk_update_ktime_data(struct timekeeper *tk)
504 * The xtime based monotonic readout is:
505 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
506 * The ktime based monotonic readout is:
507 * nsec = base_mono + now();
508 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
510 seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
511 nsec = (u32) tk->wall_to_monotonic.tv_nsec;
512 tk->tkr.base_mono = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
514 /* Update the monotonic raw base */
515 tk->base_raw = timespec64_to_ktime(tk->raw_time);
518 * The sum of the nanoseconds portions of xtime and
519 * wall_to_monotonic can be greater/equal one second. Take
520 * this into account before updating tk->ktime_sec.
522 nsec += (u32)(tk->tkr.xtime_nsec >> tk->tkr.shift);
523 if (nsec >= NSEC_PER_SEC)
525 tk->ktime_sec = seconds;
528 /* must hold timekeeper_lock */
529 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
531 if (action & TK_CLEAR_NTP) {
536 tk_update_ktime_data(tk);
539 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
541 if (action & TK_MIRROR)
542 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
543 sizeof(tk_core.timekeeper));
545 update_fast_timekeeper(&tk->tkr);
549 * timekeeping_forward_now - update clock to the current time
551 * Forward the current clock to update its state since the last call to
552 * update_wall_time(). This is useful before significant clock changes,
553 * as it avoids having to deal with this time offset explicitly.
555 static void timekeeping_forward_now(struct timekeeper *tk)
557 struct clocksource *clock = tk->tkr.clock;
558 cycle_t cycle_now, delta;
561 cycle_now = tk->tkr.read(clock);
562 delta = clocksource_delta(cycle_now, tk->tkr.cycle_last, tk->tkr.mask);
563 tk->tkr.cycle_last = cycle_now;
565 tk->tkr.xtime_nsec += delta * tk->tkr.mult;
567 /* If arch requires, add in get_arch_timeoffset() */
568 tk->tkr.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr.shift;
570 tk_normalize_xtime(tk);
572 nsec = clocksource_cyc2ns(delta, clock->mult, clock->shift);
573 timespec64_add_ns(&tk->raw_time, nsec);
577 * __getnstimeofday64 - Returns the time of day in a timespec64.
578 * @ts: pointer to the timespec to be set
580 * Updates the time of day in the timespec.
581 * Returns 0 on success, or -ve when suspended (timespec will be undefined).
583 int __getnstimeofday64(struct timespec64 *ts)
585 struct timekeeper *tk = &tk_core.timekeeper;
590 seq = read_seqcount_begin(&tk_core.seq);
592 ts->tv_sec = tk->xtime_sec;
593 nsecs = timekeeping_get_ns(&tk->tkr);
595 } while (read_seqcount_retry(&tk_core.seq, seq));
598 timespec64_add_ns(ts, nsecs);
601 * Do not bail out early, in case there were callers still using
602 * the value, even in the face of the WARN_ON.
604 if (unlikely(timekeeping_suspended))
608 EXPORT_SYMBOL(__getnstimeofday64);
611 * getnstimeofday64 - Returns the time of day in a timespec64.
612 * @ts: pointer to the timespec64 to be set
614 * Returns the time of day in a timespec64 (WARN if suspended).
616 void getnstimeofday64(struct timespec64 *ts)
618 WARN_ON(__getnstimeofday64(ts));
620 EXPORT_SYMBOL(getnstimeofday64);
622 ktime_t ktime_get(void)
624 struct timekeeper *tk = &tk_core.timekeeper;
629 WARN_ON(timekeeping_suspended);
632 seq = read_seqcount_begin(&tk_core.seq);
633 base = tk->tkr.base_mono;
634 nsecs = timekeeping_get_ns(&tk->tkr);
636 } while (read_seqcount_retry(&tk_core.seq, seq));
638 return ktime_add_ns(base, nsecs);
640 EXPORT_SYMBOL_GPL(ktime_get);
642 static ktime_t *offsets[TK_OFFS_MAX] = {
643 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
644 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
645 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
648 ktime_t ktime_get_with_offset(enum tk_offsets offs)
650 struct timekeeper *tk = &tk_core.timekeeper;
652 ktime_t base, *offset = offsets[offs];
655 WARN_ON(timekeeping_suspended);
658 seq = read_seqcount_begin(&tk_core.seq);
659 base = ktime_add(tk->tkr.base_mono, *offset);
660 nsecs = timekeeping_get_ns(&tk->tkr);
662 } while (read_seqcount_retry(&tk_core.seq, seq));
664 return ktime_add_ns(base, nsecs);
667 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
670 * ktime_mono_to_any() - convert mononotic time to any other time
671 * @tmono: time to convert.
672 * @offs: which offset to use
674 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
676 ktime_t *offset = offsets[offs];
681 seq = read_seqcount_begin(&tk_core.seq);
682 tconv = ktime_add(tmono, *offset);
683 } while (read_seqcount_retry(&tk_core.seq, seq));
687 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
690 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
692 ktime_t ktime_get_raw(void)
694 struct timekeeper *tk = &tk_core.timekeeper;
700 seq = read_seqcount_begin(&tk_core.seq);
702 nsecs = timekeeping_get_ns_raw(tk);
704 } while (read_seqcount_retry(&tk_core.seq, seq));
706 return ktime_add_ns(base, nsecs);
708 EXPORT_SYMBOL_GPL(ktime_get_raw);
711 * ktime_get_ts64 - get the monotonic clock in timespec64 format
712 * @ts: pointer to timespec variable
714 * The function calculates the monotonic clock from the realtime
715 * clock and the wall_to_monotonic offset and stores the result
716 * in normalized timespec64 format in the variable pointed to by @ts.
718 void ktime_get_ts64(struct timespec64 *ts)
720 struct timekeeper *tk = &tk_core.timekeeper;
721 struct timespec64 tomono;
725 WARN_ON(timekeeping_suspended);
728 seq = read_seqcount_begin(&tk_core.seq);
729 ts->tv_sec = tk->xtime_sec;
730 nsec = timekeeping_get_ns(&tk->tkr);
731 tomono = tk->wall_to_monotonic;
733 } while (read_seqcount_retry(&tk_core.seq, seq));
735 ts->tv_sec += tomono.tv_sec;
737 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
739 EXPORT_SYMBOL_GPL(ktime_get_ts64);
742 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
744 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
745 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
746 * works on both 32 and 64 bit systems. On 32 bit systems the readout
747 * covers ~136 years of uptime which should be enough to prevent
748 * premature wrap arounds.
750 time64_t ktime_get_seconds(void)
752 struct timekeeper *tk = &tk_core.timekeeper;
754 WARN_ON(timekeeping_suspended);
755 return tk->ktime_sec;
757 EXPORT_SYMBOL_GPL(ktime_get_seconds);
760 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
762 * Returns the wall clock seconds since 1970. This replaces the
763 * get_seconds() interface which is not y2038 safe on 32bit systems.
765 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
766 * 32bit systems the access must be protected with the sequence
767 * counter to provide "atomic" access to the 64bit tk->xtime_sec
770 time64_t ktime_get_real_seconds(void)
772 struct timekeeper *tk = &tk_core.timekeeper;
776 if (IS_ENABLED(CONFIG_64BIT))
777 return tk->xtime_sec;
780 seq = read_seqcount_begin(&tk_core.seq);
781 seconds = tk->xtime_sec;
783 } while (read_seqcount_retry(&tk_core.seq, seq));
787 EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
789 #ifdef CONFIG_NTP_PPS
792 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
793 * @ts_raw: pointer to the timespec to be set to raw monotonic time
794 * @ts_real: pointer to the timespec to be set to the time of day
796 * This function reads both the time of day and raw monotonic time at the
797 * same time atomically and stores the resulting timestamps in timespec
800 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
802 struct timekeeper *tk = &tk_core.timekeeper;
804 s64 nsecs_raw, nsecs_real;
806 WARN_ON_ONCE(timekeeping_suspended);
809 seq = read_seqcount_begin(&tk_core.seq);
811 *ts_raw = timespec64_to_timespec(tk->raw_time);
812 ts_real->tv_sec = tk->xtime_sec;
813 ts_real->tv_nsec = 0;
815 nsecs_raw = timekeeping_get_ns_raw(tk);
816 nsecs_real = timekeeping_get_ns(&tk->tkr);
818 } while (read_seqcount_retry(&tk_core.seq, seq));
820 timespec_add_ns(ts_raw, nsecs_raw);
821 timespec_add_ns(ts_real, nsecs_real);
823 EXPORT_SYMBOL(getnstime_raw_and_real);
825 #endif /* CONFIG_NTP_PPS */
828 * do_gettimeofday - Returns the time of day in a timeval
829 * @tv: pointer to the timeval to be set
831 * NOTE: Users should be converted to using getnstimeofday()
833 void do_gettimeofday(struct timeval *tv)
835 struct timespec64 now;
837 getnstimeofday64(&now);
838 tv->tv_sec = now.tv_sec;
839 tv->tv_usec = now.tv_nsec/1000;
841 EXPORT_SYMBOL(do_gettimeofday);
844 * do_settimeofday64 - Sets the time of day.
845 * @ts: pointer to the timespec64 variable containing the new time
847 * Sets the time of day to the new time and update NTP and notify hrtimers
849 int do_settimeofday64(const struct timespec64 *ts)
851 struct timekeeper *tk = &tk_core.timekeeper;
852 struct timespec64 ts_delta, xt;
855 if (!timespec64_valid_strict(ts))
858 raw_spin_lock_irqsave(&timekeeper_lock, flags);
859 write_seqcount_begin(&tk_core.seq);
861 timekeeping_forward_now(tk);
864 ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
865 ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
867 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
869 tk_set_xtime(tk, ts);
871 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
873 write_seqcount_end(&tk_core.seq);
874 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
876 /* signal hrtimers about time change */
881 EXPORT_SYMBOL(do_settimeofday64);
884 * timekeeping_inject_offset - Adds or subtracts from the current time.
885 * @tv: pointer to the timespec variable containing the offset
887 * Adds or subtracts an offset value from the current time.
889 int timekeeping_inject_offset(struct timespec *ts)
891 struct timekeeper *tk = &tk_core.timekeeper;
893 struct timespec64 ts64, tmp;
896 if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
899 ts64 = timespec_to_timespec64(*ts);
901 raw_spin_lock_irqsave(&timekeeper_lock, flags);
902 write_seqcount_begin(&tk_core.seq);
904 timekeeping_forward_now(tk);
906 /* Make sure the proposed value is valid */
907 tmp = timespec64_add(tk_xtime(tk), ts64);
908 if (!timespec64_valid_strict(&tmp)) {
913 tk_xtime_add(tk, &ts64);
914 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts64));
916 error: /* even if we error out, we forwarded the time, so call update */
917 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
919 write_seqcount_end(&tk_core.seq);
920 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
922 /* signal hrtimers about time change */
927 EXPORT_SYMBOL(timekeeping_inject_offset);
931 * timekeeping_get_tai_offset - Returns current TAI offset from UTC
934 s32 timekeeping_get_tai_offset(void)
936 struct timekeeper *tk = &tk_core.timekeeper;
941 seq = read_seqcount_begin(&tk_core.seq);
942 ret = tk->tai_offset;
943 } while (read_seqcount_retry(&tk_core.seq, seq));
949 * __timekeeping_set_tai_offset - Lock free worker function
952 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
954 tk->tai_offset = tai_offset;
955 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
959 * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
962 void timekeeping_set_tai_offset(s32 tai_offset)
964 struct timekeeper *tk = &tk_core.timekeeper;
967 raw_spin_lock_irqsave(&timekeeper_lock, flags);
968 write_seqcount_begin(&tk_core.seq);
969 __timekeeping_set_tai_offset(tk, tai_offset);
970 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
971 write_seqcount_end(&tk_core.seq);
972 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
977 * change_clocksource - Swaps clocksources if a new one is available
979 * Accumulates current time interval and initializes new clocksource
981 static int change_clocksource(void *data)
983 struct timekeeper *tk = &tk_core.timekeeper;
984 struct clocksource *new, *old;
987 new = (struct clocksource *) data;
989 raw_spin_lock_irqsave(&timekeeper_lock, flags);
990 write_seqcount_begin(&tk_core.seq);
992 timekeeping_forward_now(tk);
994 * If the cs is in module, get a module reference. Succeeds
995 * for built-in code (owner == NULL) as well.
997 if (try_module_get(new->owner)) {
998 if (!new->enable || new->enable(new) == 0) {
1000 tk_setup_internals(tk, new);
1003 module_put(old->owner);
1005 module_put(new->owner);
1008 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1010 write_seqcount_end(&tk_core.seq);
1011 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1017 * timekeeping_notify - Install a new clock source
1018 * @clock: pointer to the clock source
1020 * This function is called from clocksource.c after a new, better clock
1021 * source has been registered. The caller holds the clocksource_mutex.
1023 int timekeeping_notify(struct clocksource *clock)
1025 struct timekeeper *tk = &tk_core.timekeeper;
1027 if (tk->tkr.clock == clock)
1029 stop_machine(change_clocksource, clock, NULL);
1030 tick_clock_notify();
1031 return tk->tkr.clock == clock ? 0 : -1;
1035 * getrawmonotonic64 - Returns the raw monotonic time in a timespec
1036 * @ts: pointer to the timespec64 to be set
1038 * Returns the raw monotonic time (completely un-modified by ntp)
1040 void getrawmonotonic64(struct timespec64 *ts)
1042 struct timekeeper *tk = &tk_core.timekeeper;
1043 struct timespec64 ts64;
1048 seq = read_seqcount_begin(&tk_core.seq);
1049 nsecs = timekeeping_get_ns_raw(tk);
1050 ts64 = tk->raw_time;
1052 } while (read_seqcount_retry(&tk_core.seq, seq));
1054 timespec64_add_ns(&ts64, nsecs);
1057 EXPORT_SYMBOL(getrawmonotonic64);
1061 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1063 int timekeeping_valid_for_hres(void)
1065 struct timekeeper *tk = &tk_core.timekeeper;
1070 seq = read_seqcount_begin(&tk_core.seq);
1072 ret = tk->tkr.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1074 } while (read_seqcount_retry(&tk_core.seq, seq));
1080 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1082 u64 timekeeping_max_deferment(void)
1084 struct timekeeper *tk = &tk_core.timekeeper;
1089 seq = read_seqcount_begin(&tk_core.seq);
1091 ret = tk->tkr.clock->max_idle_ns;
1093 } while (read_seqcount_retry(&tk_core.seq, seq));
1099 * read_persistent_clock - Return time from the persistent clock.
1101 * Weak dummy function for arches that do not yet support it.
1102 * Reads the time from the battery backed persistent clock.
1103 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1105 * XXX - Do be sure to remove it once all arches implement it.
1107 void __weak read_persistent_clock(struct timespec *ts)
1114 * read_boot_clock - Return time of the system start.
1116 * Weak dummy function for arches that do not yet support it.
1117 * Function to read the exact time the system has been started.
1118 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1120 * XXX - Do be sure to remove it once all arches implement it.
1122 void __weak read_boot_clock(struct timespec *ts)
1129 * timekeeping_init - Initializes the clocksource and common timekeeping values
1131 void __init timekeeping_init(void)
1133 struct timekeeper *tk = &tk_core.timekeeper;
1134 struct clocksource *clock;
1135 unsigned long flags;
1136 struct timespec64 now, boot, tmp;
1139 read_persistent_clock(&ts);
1140 now = timespec_to_timespec64(ts);
1141 if (!timespec64_valid_strict(&now)) {
1142 pr_warn("WARNING: Persistent clock returned invalid value!\n"
1143 " Check your CMOS/BIOS settings.\n");
1146 } else if (now.tv_sec || now.tv_nsec)
1147 persistent_clock_exist = true;
1149 read_boot_clock(&ts);
1150 boot = timespec_to_timespec64(ts);
1151 if (!timespec64_valid_strict(&boot)) {
1152 pr_warn("WARNING: Boot clock returned invalid value!\n"
1153 " Check your CMOS/BIOS settings.\n");
1158 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1159 write_seqcount_begin(&tk_core.seq);
1162 clock = clocksource_default_clock();
1164 clock->enable(clock);
1165 tk_setup_internals(tk, clock);
1167 tk_set_xtime(tk, &now);
1168 tk->raw_time.tv_sec = 0;
1169 tk->raw_time.tv_nsec = 0;
1170 tk->base_raw.tv64 = 0;
1171 if (boot.tv_sec == 0 && boot.tv_nsec == 0)
1172 boot = tk_xtime(tk);
1174 set_normalized_timespec64(&tmp, -boot.tv_sec, -boot.tv_nsec);
1175 tk_set_wall_to_mono(tk, tmp);
1177 timekeeping_update(tk, TK_MIRROR);
1179 write_seqcount_end(&tk_core.seq);
1180 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1183 /* time in seconds when suspend began */
1184 static struct timespec64 timekeeping_suspend_time;
1187 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1188 * @delta: pointer to a timespec delta value
1190 * Takes a timespec offset measuring a suspend interval and properly
1191 * adds the sleep offset to the timekeeping variables.
1193 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1194 struct timespec64 *delta)
1196 if (!timespec64_valid_strict(delta)) {
1197 printk_deferred(KERN_WARNING
1198 "__timekeeping_inject_sleeptime: Invalid "
1199 "sleep delta value!\n");
1202 tk_xtime_add(tk, delta);
1203 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1204 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1205 tk_debug_account_sleep_time(delta);
1209 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1210 * @delta: pointer to a timespec64 delta value
1212 * This hook is for architectures that cannot support read_persistent_clock
1213 * because their RTC/persistent clock is only accessible when irqs are enabled.
1215 * This function should only be called by rtc_resume(), and allows
1216 * a suspend offset to be injected into the timekeeping values.
1218 void timekeeping_inject_sleeptime64(struct timespec64 *delta)
1220 struct timekeeper *tk = &tk_core.timekeeper;
1221 unsigned long flags;
1224 * Make sure we don't set the clock twice, as timekeeping_resume()
1227 if (has_persistent_clock())
1230 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1231 write_seqcount_begin(&tk_core.seq);
1233 timekeeping_forward_now(tk);
1235 __timekeeping_inject_sleeptime(tk, delta);
1237 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1239 write_seqcount_end(&tk_core.seq);
1240 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1242 /* signal hrtimers about time change */
1247 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1249 * This is for the generic clocksource timekeeping.
1250 * xtime/wall_to_monotonic/jiffies/etc are
1251 * still managed by arch specific suspend/resume code.
1253 void timekeeping_resume(void)
1255 struct timekeeper *tk = &tk_core.timekeeper;
1256 struct clocksource *clock = tk->tkr.clock;
1257 unsigned long flags;
1258 struct timespec64 ts_new, ts_delta;
1259 struct timespec tmp;
1260 cycle_t cycle_now, cycle_delta;
1261 bool suspendtime_found = false;
1263 read_persistent_clock(&tmp);
1264 ts_new = timespec_to_timespec64(tmp);
1266 clockevents_resume();
1267 clocksource_resume();
1269 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1270 write_seqcount_begin(&tk_core.seq);
1273 * After system resumes, we need to calculate the suspended time and
1274 * compensate it for the OS time. There are 3 sources that could be
1275 * used: Nonstop clocksource during suspend, persistent clock and rtc
1278 * One specific platform may have 1 or 2 or all of them, and the
1279 * preference will be:
1280 * suspend-nonstop clocksource -> persistent clock -> rtc
1281 * The less preferred source will only be tried if there is no better
1282 * usable source. The rtc part is handled separately in rtc core code.
1284 cycle_now = tk->tkr.read(clock);
1285 if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
1286 cycle_now > tk->tkr.cycle_last) {
1287 u64 num, max = ULLONG_MAX;
1288 u32 mult = clock->mult;
1289 u32 shift = clock->shift;
1292 cycle_delta = clocksource_delta(cycle_now, tk->tkr.cycle_last,
1296 * "cycle_delta * mutl" may cause 64 bits overflow, if the
1297 * suspended time is too long. In that case we need do the
1298 * 64 bits math carefully
1301 if (cycle_delta > max) {
1302 num = div64_u64(cycle_delta, max);
1303 nsec = (((u64) max * mult) >> shift) * num;
1304 cycle_delta -= num * max;
1306 nsec += ((u64) cycle_delta * mult) >> shift;
1308 ts_delta = ns_to_timespec64(nsec);
1309 suspendtime_found = true;
1310 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1311 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1312 suspendtime_found = true;
1315 if (suspendtime_found)
1316 __timekeeping_inject_sleeptime(tk, &ts_delta);
1318 /* Re-base the last cycle value */
1319 tk->tkr.cycle_last = cycle_now;
1321 timekeeping_suspended = 0;
1322 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1323 write_seqcount_end(&tk_core.seq);
1324 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1326 touch_softlockup_watchdog();
1328 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
1330 /* Resume hrtimers */
1334 int timekeeping_suspend(void)
1336 struct timekeeper *tk = &tk_core.timekeeper;
1337 unsigned long flags;
1338 struct timespec64 delta, delta_delta;
1339 static struct timespec64 old_delta;
1340 struct timespec tmp;
1342 read_persistent_clock(&tmp);
1343 timekeeping_suspend_time = timespec_to_timespec64(tmp);
1346 * On some systems the persistent_clock can not be detected at
1347 * timekeeping_init by its return value, so if we see a valid
1348 * value returned, update the persistent_clock_exists flag.
1350 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1351 persistent_clock_exist = true;
1353 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1354 write_seqcount_begin(&tk_core.seq);
1355 timekeeping_forward_now(tk);
1356 timekeeping_suspended = 1;
1359 * To avoid drift caused by repeated suspend/resumes,
1360 * which each can add ~1 second drift error,
1361 * try to compensate so the difference in system time
1362 * and persistent_clock time stays close to constant.
1364 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1365 delta_delta = timespec64_sub(delta, old_delta);
1366 if (abs(delta_delta.tv_sec) >= 2) {
1368 * if delta_delta is too large, assume time correction
1369 * has occured and set old_delta to the current delta.
1373 /* Otherwise try to adjust old_system to compensate */
1374 timekeeping_suspend_time =
1375 timespec64_add(timekeeping_suspend_time, delta_delta);
1378 timekeeping_update(tk, TK_MIRROR);
1379 halt_fast_timekeeper(tk);
1380 write_seqcount_end(&tk_core.seq);
1381 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1383 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1384 clocksource_suspend();
1385 clockevents_suspend();
1390 /* sysfs resume/suspend bits for timekeeping */
1391 static struct syscore_ops timekeeping_syscore_ops = {
1392 .resume = timekeeping_resume,
1393 .suspend = timekeeping_suspend,
1396 static int __init timekeeping_init_ops(void)
1398 register_syscore_ops(&timekeeping_syscore_ops);
1401 device_initcall(timekeeping_init_ops);
1404 * Apply a multiplier adjustment to the timekeeper
1406 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1411 s64 interval = tk->cycle_interval;
1415 mult_adj = -mult_adj;
1416 interval = -interval;
1419 mult_adj <<= adj_scale;
1420 interval <<= adj_scale;
1421 offset <<= adj_scale;
1424 * So the following can be confusing.
1426 * To keep things simple, lets assume mult_adj == 1 for now.
1428 * When mult_adj != 1, remember that the interval and offset values
1429 * have been appropriately scaled so the math is the same.
1431 * The basic idea here is that we're increasing the multiplier
1432 * by one, this causes the xtime_interval to be incremented by
1433 * one cycle_interval. This is because:
1434 * xtime_interval = cycle_interval * mult
1435 * So if mult is being incremented by one:
1436 * xtime_interval = cycle_interval * (mult + 1)
1438 * xtime_interval = (cycle_interval * mult) + cycle_interval
1439 * Which can be shortened to:
1440 * xtime_interval += cycle_interval
1442 * So offset stores the non-accumulated cycles. Thus the current
1443 * time (in shifted nanoseconds) is:
1444 * now = (offset * adj) + xtime_nsec
1445 * Now, even though we're adjusting the clock frequency, we have
1446 * to keep time consistent. In other words, we can't jump back
1447 * in time, and we also want to avoid jumping forward in time.
1449 * So given the same offset value, we need the time to be the same
1450 * both before and after the freq adjustment.
1451 * now = (offset * adj_1) + xtime_nsec_1
1452 * now = (offset * adj_2) + xtime_nsec_2
1454 * (offset * adj_1) + xtime_nsec_1 =
1455 * (offset * adj_2) + xtime_nsec_2
1459 * (offset * adj_1) + xtime_nsec_1 =
1460 * (offset * (adj_1+1)) + xtime_nsec_2
1461 * (offset * adj_1) + xtime_nsec_1 =
1462 * (offset * adj_1) + offset + xtime_nsec_2
1463 * Canceling the sides:
1464 * xtime_nsec_1 = offset + xtime_nsec_2
1466 * xtime_nsec_2 = xtime_nsec_1 - offset
1467 * Which simplfies to:
1468 * xtime_nsec -= offset
1470 * XXX - TODO: Doc ntp_error calculation.
1472 if ((mult_adj > 0) && (tk->tkr.mult + mult_adj < mult_adj)) {
1473 /* NTP adjustment caused clocksource mult overflow */
1478 tk->tkr.mult += mult_adj;
1479 tk->xtime_interval += interval;
1480 tk->tkr.xtime_nsec -= offset;
1481 tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1485 * Calculate the multiplier adjustment needed to match the frequency
1488 static __always_inline void timekeeping_freqadjust(struct timekeeper *tk,
1491 s64 interval = tk->cycle_interval;
1492 s64 xinterval = tk->xtime_interval;
1497 /* Remove any current error adj from freq calculation */
1498 if (tk->ntp_err_mult)
1499 xinterval -= tk->cycle_interval;
1501 tk->ntp_tick = ntp_tick_length();
1503 /* Calculate current error per tick */
1504 tick_error = ntp_tick_length() >> tk->ntp_error_shift;
1505 tick_error -= (xinterval + tk->xtime_remainder);
1507 /* Don't worry about correcting it if its small */
1508 if (likely((tick_error >= 0) && (tick_error <= interval)))
1511 /* preserve the direction of correction */
1512 negative = (tick_error < 0);
1514 /* Sort out the magnitude of the correction */
1515 tick_error = abs(tick_error);
1516 for (adj = 0; tick_error > interval; adj++)
1519 /* scale the corrections */
1520 timekeeping_apply_adjustment(tk, offset, negative, adj);
1524 * Adjust the timekeeper's multiplier to the correct frequency
1525 * and also to reduce the accumulated error value.
1527 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1529 /* Correct for the current frequency error */
1530 timekeeping_freqadjust(tk, offset);
1532 /* Next make a small adjustment to fix any cumulative error */
1533 if (!tk->ntp_err_mult && (tk->ntp_error > 0)) {
1534 tk->ntp_err_mult = 1;
1535 timekeeping_apply_adjustment(tk, offset, 0, 0);
1536 } else if (tk->ntp_err_mult && (tk->ntp_error <= 0)) {
1537 /* Undo any existing error adjustment */
1538 timekeeping_apply_adjustment(tk, offset, 1, 0);
1539 tk->ntp_err_mult = 0;
1542 if (unlikely(tk->tkr.clock->maxadj &&
1543 (abs(tk->tkr.mult - tk->tkr.clock->mult)
1544 > tk->tkr.clock->maxadj))) {
1545 printk_once(KERN_WARNING
1546 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1547 tk->tkr.clock->name, (long)tk->tkr.mult,
1548 (long)tk->tkr.clock->mult + tk->tkr.clock->maxadj);
1552 * It may be possible that when we entered this function, xtime_nsec
1553 * was very small. Further, if we're slightly speeding the clocksource
1554 * in the code above, its possible the required corrective factor to
1555 * xtime_nsec could cause it to underflow.
1557 * Now, since we already accumulated the second, cannot simply roll
1558 * the accumulated second back, since the NTP subsystem has been
1559 * notified via second_overflow. So instead we push xtime_nsec forward
1560 * by the amount we underflowed, and add that amount into the error.
1562 * We'll correct this error next time through this function, when
1563 * xtime_nsec is not as small.
1565 if (unlikely((s64)tk->tkr.xtime_nsec < 0)) {
1566 s64 neg = -(s64)tk->tkr.xtime_nsec;
1567 tk->tkr.xtime_nsec = 0;
1568 tk->ntp_error += neg << tk->ntp_error_shift;
1573 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1575 * Helper function that accumulates a the nsecs greater then a second
1576 * from the xtime_nsec field to the xtime_secs field.
1577 * It also calls into the NTP code to handle leapsecond processing.
1580 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1582 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr.shift;
1583 unsigned int clock_set = 0;
1585 while (tk->tkr.xtime_nsec >= nsecps) {
1588 tk->tkr.xtime_nsec -= nsecps;
1591 /* Figure out if its a leap sec and apply if needed */
1592 leap = second_overflow(tk->xtime_sec);
1593 if (unlikely(leap)) {
1594 struct timespec64 ts;
1596 tk->xtime_sec += leap;
1600 tk_set_wall_to_mono(tk,
1601 timespec64_sub(tk->wall_to_monotonic, ts));
1603 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1605 clock_set = TK_CLOCK_WAS_SET;
1612 * logarithmic_accumulation - shifted accumulation of cycles
1614 * This functions accumulates a shifted interval of cycles into
1615 * into a shifted interval nanoseconds. Allows for O(log) accumulation
1618 * Returns the unconsumed cycles.
1620 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1622 unsigned int *clock_set)
1624 cycle_t interval = tk->cycle_interval << shift;
1627 /* If the offset is smaller then a shifted interval, do nothing */
1628 if (offset < interval)
1631 /* Accumulate one shifted interval */
1633 tk->tkr.cycle_last += interval;
1635 tk->tkr.xtime_nsec += tk->xtime_interval << shift;
1636 *clock_set |= accumulate_nsecs_to_secs(tk);
1638 /* Accumulate raw time */
1639 raw_nsecs = (u64)tk->raw_interval << shift;
1640 raw_nsecs += tk->raw_time.tv_nsec;
1641 if (raw_nsecs >= NSEC_PER_SEC) {
1642 u64 raw_secs = raw_nsecs;
1643 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1644 tk->raw_time.tv_sec += raw_secs;
1646 tk->raw_time.tv_nsec = raw_nsecs;
1648 /* Accumulate error between NTP and clock interval */
1649 tk->ntp_error += tk->ntp_tick << shift;
1650 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1651 (tk->ntp_error_shift + shift);
1657 * update_wall_time - Uses the current clocksource to increment the wall time
1660 void update_wall_time(void)
1662 struct timekeeper *real_tk = &tk_core.timekeeper;
1663 struct timekeeper *tk = &shadow_timekeeper;
1665 int shift = 0, maxshift;
1666 unsigned int clock_set = 0;
1667 unsigned long flags;
1669 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1671 /* Make sure we're fully resumed: */
1672 if (unlikely(timekeeping_suspended))
1675 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1676 offset = real_tk->cycle_interval;
1678 offset = clocksource_delta(tk->tkr.read(tk->tkr.clock),
1679 tk->tkr.cycle_last, tk->tkr.mask);
1682 /* Check if there's really nothing to do */
1683 if (offset < real_tk->cycle_interval)
1686 /* Do some additional sanity checking */
1687 timekeeping_check_update(real_tk, offset);
1690 * With NO_HZ we may have to accumulate many cycle_intervals
1691 * (think "ticks") worth of time at once. To do this efficiently,
1692 * we calculate the largest doubling multiple of cycle_intervals
1693 * that is smaller than the offset. We then accumulate that
1694 * chunk in one go, and then try to consume the next smaller
1697 shift = ilog2(offset) - ilog2(tk->cycle_interval);
1698 shift = max(0, shift);
1699 /* Bound shift to one less than what overflows tick_length */
1700 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1701 shift = min(shift, maxshift);
1702 while (offset >= tk->cycle_interval) {
1703 offset = logarithmic_accumulation(tk, offset, shift,
1705 if (offset < tk->cycle_interval<<shift)
1709 /* correct the clock when NTP error is too big */
1710 timekeeping_adjust(tk, offset);
1713 * XXX This can be killed once everyone converts
1714 * to the new update_vsyscall.
1716 old_vsyscall_fixup(tk);
1719 * Finally, make sure that after the rounding
1720 * xtime_nsec isn't larger than NSEC_PER_SEC
1722 clock_set |= accumulate_nsecs_to_secs(tk);
1724 write_seqcount_begin(&tk_core.seq);
1726 * Update the real timekeeper.
1728 * We could avoid this memcpy by switching pointers, but that
1729 * requires changes to all other timekeeper usage sites as
1730 * well, i.e. move the timekeeper pointer getter into the
1731 * spinlocked/seqcount protected sections. And we trade this
1732 * memcpy under the tk_core.seq against one before we start
1735 memcpy(real_tk, tk, sizeof(*tk));
1736 timekeeping_update(real_tk, clock_set);
1737 write_seqcount_end(&tk_core.seq);
1739 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1741 /* Have to call _delayed version, since in irq context*/
1742 clock_was_set_delayed();
1746 * getboottime64 - Return the real time of system boot.
1747 * @ts: pointer to the timespec64 to be set
1749 * Returns the wall-time of boot in a timespec64.
1751 * This is based on the wall_to_monotonic offset and the total suspend
1752 * time. Calls to settimeofday will affect the value returned (which
1753 * basically means that however wrong your real time clock is at boot time,
1754 * you get the right time here).
1756 void getboottime64(struct timespec64 *ts)
1758 struct timekeeper *tk = &tk_core.timekeeper;
1759 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
1761 *ts = ktime_to_timespec64(t);
1763 EXPORT_SYMBOL_GPL(getboottime64);
1765 unsigned long get_seconds(void)
1767 struct timekeeper *tk = &tk_core.timekeeper;
1769 return tk->xtime_sec;
1771 EXPORT_SYMBOL(get_seconds);
1773 struct timespec __current_kernel_time(void)
1775 struct timekeeper *tk = &tk_core.timekeeper;
1777 return timespec64_to_timespec(tk_xtime(tk));
1780 struct timespec current_kernel_time(void)
1782 struct timekeeper *tk = &tk_core.timekeeper;
1783 struct timespec64 now;
1787 seq = read_seqcount_begin(&tk_core.seq);
1790 } while (read_seqcount_retry(&tk_core.seq, seq));
1792 return timespec64_to_timespec(now);
1794 EXPORT_SYMBOL(current_kernel_time);
1796 struct timespec64 get_monotonic_coarse64(void)
1798 struct timekeeper *tk = &tk_core.timekeeper;
1799 struct timespec64 now, mono;
1803 seq = read_seqcount_begin(&tk_core.seq);
1806 mono = tk->wall_to_monotonic;
1807 } while (read_seqcount_retry(&tk_core.seq, seq));
1809 set_normalized_timespec64(&now, now.tv_sec + mono.tv_sec,
1810 now.tv_nsec + mono.tv_nsec);
1816 * Must hold jiffies_lock
1818 void do_timer(unsigned long ticks)
1820 jiffies_64 += ticks;
1821 calc_global_load(ticks);
1825 * ktime_get_update_offsets_tick - hrtimer helper
1826 * @offs_real: pointer to storage for monotonic -> realtime offset
1827 * @offs_boot: pointer to storage for monotonic -> boottime offset
1828 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1830 * Returns monotonic time at last tick and various offsets
1832 ktime_t ktime_get_update_offsets_tick(ktime_t *offs_real, ktime_t *offs_boot,
1835 struct timekeeper *tk = &tk_core.timekeeper;
1841 seq = read_seqcount_begin(&tk_core.seq);
1843 base = tk->tkr.base_mono;
1844 nsecs = tk->tkr.xtime_nsec >> tk->tkr.shift;
1846 *offs_real = tk->offs_real;
1847 *offs_boot = tk->offs_boot;
1848 *offs_tai = tk->offs_tai;
1849 } while (read_seqcount_retry(&tk_core.seq, seq));
1851 return ktime_add_ns(base, nsecs);
1854 #ifdef CONFIG_HIGH_RES_TIMERS
1856 * ktime_get_update_offsets_now - hrtimer helper
1857 * @offs_real: pointer to storage for monotonic -> realtime offset
1858 * @offs_boot: pointer to storage for monotonic -> boottime offset
1859 * @offs_tai: pointer to storage for monotonic -> clock tai offset
1861 * Returns current monotonic time and updates the offsets
1862 * Called from hrtimer_interrupt() or retrigger_next_event()
1864 ktime_t ktime_get_update_offsets_now(ktime_t *offs_real, ktime_t *offs_boot,
1867 struct timekeeper *tk = &tk_core.timekeeper;
1873 seq = read_seqcount_begin(&tk_core.seq);
1875 base = tk->tkr.base_mono;
1876 nsecs = timekeeping_get_ns(&tk->tkr);
1878 *offs_real = tk->offs_real;
1879 *offs_boot = tk->offs_boot;
1880 *offs_tai = tk->offs_tai;
1881 } while (read_seqcount_retry(&tk_core.seq, seq));
1883 return ktime_add_ns(base, nsecs);
1888 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1890 int do_adjtimex(struct timex *txc)
1892 struct timekeeper *tk = &tk_core.timekeeper;
1893 unsigned long flags;
1894 struct timespec64 ts;
1898 /* Validate the data before disabling interrupts */
1899 ret = ntp_validate_timex(txc);
1903 if (txc->modes & ADJ_SETOFFSET) {
1904 struct timespec delta;
1905 delta.tv_sec = txc->time.tv_sec;
1906 delta.tv_nsec = txc->time.tv_usec;
1907 if (!(txc->modes & ADJ_NANO))
1908 delta.tv_nsec *= 1000;
1909 ret = timekeeping_inject_offset(&delta);
1914 getnstimeofday64(&ts);
1916 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1917 write_seqcount_begin(&tk_core.seq);
1919 orig_tai = tai = tk->tai_offset;
1920 ret = __do_adjtimex(txc, &ts, &tai);
1922 if (tai != orig_tai) {
1923 __timekeeping_set_tai_offset(tk, tai);
1924 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1926 write_seqcount_end(&tk_core.seq);
1927 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1929 if (tai != orig_tai)
1932 ntp_notify_cmos_timer();
1937 #ifdef CONFIG_NTP_PPS
1939 * hardpps() - Accessor function to NTP __hardpps function
1941 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1943 unsigned long flags;
1945 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1946 write_seqcount_begin(&tk_core.seq);
1948 __hardpps(phase_ts, raw_ts);
1950 write_seqcount_end(&tk_core.seq);
1951 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1953 EXPORT_SYMBOL(hardpps);
1957 * xtime_update() - advances the timekeeping infrastructure
1958 * @ticks: number of ticks, that have elapsed since the last call.
1960 * Must be called with interrupts disabled.
1962 void xtime_update(unsigned long ticks)
1964 write_seqlock(&jiffies_lock);
1966 write_sequnlock(&jiffies_lock);