1 // SPDX-License-Identifier: GPL-2.0
3 * Kernel timekeeping code and accessor functions. Based on code from
4 * timer.c, moved in commit 8524070b7982.
6 #include <linux/timekeeper_internal.h>
7 #include <linux/module.h>
8 #include <linux/interrupt.h>
9 #include <linux/percpu.h>
10 #include <linux/init.h>
12 #include <linux/nmi.h>
13 #include <linux/sched.h>
14 #include <linux/sched/loadavg.h>
15 #include <linux/sched/clock.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/clocksource.h>
18 #include <linux/jiffies.h>
19 #include <linux/time.h>
20 #include <linux/tick.h>
21 #include <linux/stop_machine.h>
22 #include <linux/pvclock_gtod.h>
23 #include <linux/compiler.h>
24 #include <linux/audit.h>
26 #include "tick-internal.h"
27 #include "ntp_internal.h"
28 #include "timekeeping_internal.h"
30 #define TK_CLEAR_NTP (1 << 0)
31 #define TK_MIRROR (1 << 1)
32 #define TK_CLOCK_WAS_SET (1 << 2)
34 enum timekeeping_adv_mode {
35 /* Update timekeeper when a tick has passed */
38 /* Update timekeeper on a direct frequency change */
42 DEFINE_RAW_SPINLOCK(timekeeper_lock);
45 * The most important data for readout fits into a single 64 byte
49 seqcount_raw_spinlock_t seq;
50 struct timekeeper timekeeper;
51 } tk_core ____cacheline_aligned = {
52 .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_core.seq, &timekeeper_lock),
55 static struct timekeeper shadow_timekeeper;
57 /* flag for if timekeeping is suspended */
58 int __read_mostly timekeeping_suspended;
61 * struct tk_fast - NMI safe timekeeper
62 * @seq: Sequence counter for protecting updates. The lowest bit
63 * is the index for the tk_read_base array
64 * @base: tk_read_base array. Access is indexed by the lowest bit of
67 * See @update_fast_timekeeper() below.
71 struct tk_read_base base[2];
74 /* Suspend-time cycles value for halted fast timekeeper. */
75 static u64 cycles_at_suspend;
77 static u64 dummy_clock_read(struct clocksource *cs)
79 if (timekeeping_suspended)
80 return cycles_at_suspend;
84 static struct clocksource dummy_clock = {
85 .read = dummy_clock_read,
89 * Boot time initialization which allows local_clock() to be utilized
90 * during early boot when clocksources are not available. local_clock()
91 * returns nanoseconds already so no conversion is required, hence mult=1
92 * and shift=0. When the first proper clocksource is installed then
93 * the fast time keepers are updated with the correct values.
95 #define FAST_TK_INIT \
97 .clock = &dummy_clock, \
98 .mask = CLOCKSOURCE_MASK(64), \
103 static struct tk_fast tk_fast_mono ____cacheline_aligned = {
104 .seq = SEQCNT_LATCH_ZERO(tk_fast_mono.seq),
105 .base[0] = FAST_TK_INIT,
106 .base[1] = FAST_TK_INIT,
109 static struct tk_fast tk_fast_raw ____cacheline_aligned = {
110 .seq = SEQCNT_LATCH_ZERO(tk_fast_raw.seq),
111 .base[0] = FAST_TK_INIT,
112 .base[1] = FAST_TK_INIT,
115 static inline void tk_normalize_xtime(struct timekeeper *tk)
117 while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) {
118 tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
121 while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) {
122 tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
127 static inline struct timespec64 tk_xtime(const struct timekeeper *tk)
129 struct timespec64 ts;
131 ts.tv_sec = tk->xtime_sec;
132 ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
136 static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
138 tk->xtime_sec = ts->tv_sec;
139 tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift;
142 static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
144 tk->xtime_sec += ts->tv_sec;
145 tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift;
146 tk_normalize_xtime(tk);
149 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
151 struct timespec64 tmp;
154 * Verify consistency of: offset_real = -wall_to_monotonic
155 * before modifying anything
157 set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
158 -tk->wall_to_monotonic.tv_nsec);
159 WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp));
160 tk->wall_to_monotonic = wtm;
161 set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
162 tk->offs_real = timespec64_to_ktime(tmp);
163 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
166 static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
168 tk->offs_boot = ktime_add(tk->offs_boot, delta);
170 * Timespec representation for VDSO update to avoid 64bit division
173 tk->monotonic_to_boot = ktime_to_timespec64(tk->offs_boot);
177 * tk_clock_read - atomic clocksource read() helper
179 * This helper is necessary to use in the read paths because, while the
180 * seqcount ensures we don't return a bad value while structures are updated,
181 * it doesn't protect from potential crashes. There is the possibility that
182 * the tkr's clocksource may change between the read reference, and the
183 * clock reference passed to the read function. This can cause crashes if
184 * the wrong clocksource is passed to the wrong read function.
185 * This isn't necessary to use when holding the timekeeper_lock or doing
186 * a read of the fast-timekeeper tkrs (which is protected by its own locking
189 static inline u64 tk_clock_read(const struct tk_read_base *tkr)
191 struct clocksource *clock = READ_ONCE(tkr->clock);
193 return clock->read(clock);
196 #ifdef CONFIG_DEBUG_TIMEKEEPING
197 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
199 static void timekeeping_check_update(struct timekeeper *tk, u64 offset)
202 u64 max_cycles = tk->tkr_mono.clock->max_cycles;
203 const char *name = tk->tkr_mono.clock->name;
205 if (offset > max_cycles) {
206 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
207 offset, name, max_cycles);
208 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
210 if (offset > (max_cycles >> 1)) {
211 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
212 offset, name, max_cycles >> 1);
213 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
217 if (tk->underflow_seen) {
218 if (jiffies - tk->last_warning > WARNING_FREQ) {
219 printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name);
220 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
221 printk_deferred(" Your kernel is probably still fine.\n");
222 tk->last_warning = jiffies;
224 tk->underflow_seen = 0;
227 if (tk->overflow_seen) {
228 if (jiffies - tk->last_warning > WARNING_FREQ) {
229 printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name);
230 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
231 printk_deferred(" Your kernel is probably still fine.\n");
232 tk->last_warning = jiffies;
234 tk->overflow_seen = 0;
238 static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
240 struct timekeeper *tk = &tk_core.timekeeper;
241 u64 now, last, mask, max, delta;
245 * Since we're called holding a seqcount, the data may shift
246 * under us while we're doing the calculation. This can cause
247 * false positives, since we'd note a problem but throw the
248 * results away. So nest another seqcount here to atomically
249 * grab the points we are checking with.
252 seq = read_seqcount_begin(&tk_core.seq);
253 now = tk_clock_read(tkr);
254 last = tkr->cycle_last;
256 max = tkr->clock->max_cycles;
257 } while (read_seqcount_retry(&tk_core.seq, seq));
259 delta = clocksource_delta(now, last, mask);
262 * Try to catch underflows by checking if we are seeing small
263 * mask-relative negative values.
265 if (unlikely((~delta & mask) < (mask >> 3))) {
266 tk->underflow_seen = 1;
270 /* Cap delta value to the max_cycles values to avoid mult overflows */
271 if (unlikely(delta > max)) {
272 tk->overflow_seen = 1;
273 delta = tkr->clock->max_cycles;
279 static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset)
282 static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
284 u64 cycle_now, delta;
286 /* read clocksource */
287 cycle_now = tk_clock_read(tkr);
289 /* calculate the delta since the last update_wall_time */
290 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
297 * tk_setup_internals - Set up internals to use clocksource clock.
299 * @tk: The target timekeeper to setup.
300 * @clock: Pointer to clocksource.
302 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
303 * pair and interval request.
305 * Unless you're the timekeeping code, you should not be using this!
307 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
310 u64 tmp, ntpinterval;
311 struct clocksource *old_clock;
313 ++tk->cs_was_changed_seq;
314 old_clock = tk->tkr_mono.clock;
315 tk->tkr_mono.clock = clock;
316 tk->tkr_mono.mask = clock->mask;
317 tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono);
319 tk->tkr_raw.clock = clock;
320 tk->tkr_raw.mask = clock->mask;
321 tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last;
323 /* Do the ns -> cycle conversion first, using original mult */
324 tmp = NTP_INTERVAL_LENGTH;
325 tmp <<= clock->shift;
327 tmp += clock->mult/2;
328 do_div(tmp, clock->mult);
332 interval = (u64) tmp;
333 tk->cycle_interval = interval;
335 /* Go back from cycles -> shifted ns */
336 tk->xtime_interval = interval * clock->mult;
337 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
338 tk->raw_interval = interval * clock->mult;
340 /* if changing clocks, convert xtime_nsec shift units */
342 int shift_change = clock->shift - old_clock->shift;
343 if (shift_change < 0) {
344 tk->tkr_mono.xtime_nsec >>= -shift_change;
345 tk->tkr_raw.xtime_nsec >>= -shift_change;
347 tk->tkr_mono.xtime_nsec <<= shift_change;
348 tk->tkr_raw.xtime_nsec <<= shift_change;
352 tk->tkr_mono.shift = clock->shift;
353 tk->tkr_raw.shift = clock->shift;
356 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
357 tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
360 * The timekeeper keeps its own mult values for the currently
361 * active clocksource. These value will be adjusted via NTP
362 * to counteract clock drifting.
364 tk->tkr_mono.mult = clock->mult;
365 tk->tkr_raw.mult = clock->mult;
366 tk->ntp_err_mult = 0;
367 tk->skip_second_overflow = 0;
370 /* Timekeeper helper functions. */
372 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
373 static u32 default_arch_gettimeoffset(void) { return 0; }
374 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
376 static inline u32 arch_gettimeoffset(void) { return 0; }
379 static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta)
383 nsec = delta * tkr->mult + tkr->xtime_nsec;
386 /* If arch requires, add in get_arch_timeoffset() */
387 return nsec + arch_gettimeoffset();
390 static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr)
394 delta = timekeeping_get_delta(tkr);
395 return timekeeping_delta_to_ns(tkr, delta);
398 static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles)
402 /* calculate the delta since the last update_wall_time */
403 delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask);
404 return timekeeping_delta_to_ns(tkr, delta);
408 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
409 * @tkr: Timekeeping readout base from which we take the update
411 * We want to use this from any context including NMI and tracing /
412 * instrumenting the timekeeping code itself.
414 * Employ the latch technique; see @raw_write_seqcount_latch.
416 * So if a NMI hits the update of base[0] then it will use base[1]
417 * which is still consistent. In the worst case this can result is a
418 * slightly wrong timestamp (a few nanoseconds). See
419 * @ktime_get_mono_fast_ns.
421 static void update_fast_timekeeper(const struct tk_read_base *tkr,
424 struct tk_read_base *base = tkf->base;
426 /* Force readers off to base[1] */
427 raw_write_seqcount_latch(&tkf->seq);
430 memcpy(base, tkr, sizeof(*base));
432 /* Force readers back to base[0] */
433 raw_write_seqcount_latch(&tkf->seq);
436 memcpy(base + 1, base, sizeof(*base));
440 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
442 * This timestamp is not guaranteed to be monotonic across an update.
443 * The timestamp is calculated by:
445 * now = base_mono + clock_delta * slope
447 * So if the update lowers the slope, readers who are forced to the
448 * not yet updated second array are still using the old steeper slope.
457 * |12345678---> reader order
463 * So reader 6 will observe time going backwards versus reader 5.
465 * While other CPUs are likely to be able observe that, the only way
466 * for a CPU local observation is when an NMI hits in the middle of
467 * the update. Timestamps taken from that NMI context might be ahead
468 * of the following timestamps. Callers need to be aware of that and
471 static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf)
473 struct tk_read_base *tkr;
478 seq = raw_read_seqcount_latch(&tkf->seq);
479 tkr = tkf->base + (seq & 0x01);
480 now = ktime_to_ns(tkr->base);
482 now += timekeeping_delta_to_ns(tkr,
487 } while (read_seqcount_latch_retry(&tkf->seq, seq));
492 u64 ktime_get_mono_fast_ns(void)
494 return __ktime_get_fast_ns(&tk_fast_mono);
496 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
498 u64 ktime_get_raw_fast_ns(void)
500 return __ktime_get_fast_ns(&tk_fast_raw);
502 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
505 * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
507 * To keep it NMI safe since we're accessing from tracing, we're not using a
508 * separate timekeeper with updates to monotonic clock and boot offset
509 * protected with seqcounts. This has the following minor side effects:
511 * (1) Its possible that a timestamp be taken after the boot offset is updated
512 * but before the timekeeper is updated. If this happens, the new boot offset
513 * is added to the old timekeeping making the clock appear to update slightly
516 * timekeeping_inject_sleeptime64()
517 * __timekeeping_inject_sleeptime(tk, delta);
519 * timekeeping_update(tk, TK_CLEAR_NTP...);
521 * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
522 * partially updated. Since the tk->offs_boot update is a rare event, this
523 * should be a rare occurrence which postprocessing should be able to handle.
525 u64 notrace ktime_get_boot_fast_ns(void)
527 struct timekeeper *tk = &tk_core.timekeeper;
529 return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
531 EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
534 * See comment for __ktime_get_fast_ns() vs. timestamp ordering
536 static __always_inline u64 __ktime_get_real_fast(struct tk_fast *tkf, u64 *mono)
538 struct tk_read_base *tkr;
539 u64 basem, baser, delta;
543 seq = raw_read_seqcount_latch(&tkf->seq);
544 tkr = tkf->base + (seq & 0x01);
545 basem = ktime_to_ns(tkr->base);
546 baser = ktime_to_ns(tkr->base_real);
548 delta = timekeeping_delta_to_ns(tkr,
549 clocksource_delta(tk_clock_read(tkr),
550 tkr->cycle_last, tkr->mask));
551 } while (read_seqcount_latch_retry(&tkf->seq, seq));
554 *mono = basem + delta;
555 return baser + delta;
559 * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
561 u64 ktime_get_real_fast_ns(void)
563 return __ktime_get_real_fast(&tk_fast_mono, NULL);
565 EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns);
568 * ktime_get_fast_timestamps: - NMI safe timestamps
569 * @snapshot: Pointer to timestamp storage
571 * Stores clock monotonic, boottime and realtime timestamps.
573 * Boot time is a racy access on 32bit systems if the sleep time injection
574 * happens late during resume and not in timekeeping_resume(). That could
575 * be avoided by expanding struct tk_read_base with boot offset for 32bit
576 * and adding more overhead to the update. As this is a hard to observe
577 * once per resume event which can be filtered with reasonable effort using
578 * the accurate mono/real timestamps, it's probably not worth the trouble.
580 * Aside of that it might be possible on 32 and 64 bit to observe the
581 * following when the sleep time injection happens late:
584 * timekeeping_resume()
585 * ktime_get_fast_timestamps()
586 * mono, real = __ktime_get_real_fast()
587 * inject_sleep_time()
589 * boot = mono + bootoffset;
591 * That means that boot time already has the sleep time adjustment, but
592 * real time does not. On the next readout both are in sync again.
594 * Preventing this for 64bit is not really feasible without destroying the
595 * careful cache layout of the timekeeper because the sequence count and
596 * struct tk_read_base would then need two cache lines instead of one.
598 * Access to the time keeper clock source is disabled accross the innermost
599 * steps of suspend/resume. The accessors still work, but the timestamps
600 * are frozen until time keeping is resumed which happens very early.
602 * For regular suspend/resume there is no observable difference vs. sched
603 * clock, but it might affect some of the nasty low level debug printks.
605 * OTOH, access to sched clock is not guaranteed accross suspend/resume on
606 * all systems either so it depends on the hardware in use.
608 * If that turns out to be a real problem then this could be mitigated by
609 * using sched clock in a similar way as during early boot. But it's not as
610 * trivial as on early boot because it needs some careful protection
611 * against the clock monotonic timestamp jumping backwards on resume.
613 void ktime_get_fast_timestamps(struct ktime_timestamps *snapshot)
615 struct timekeeper *tk = &tk_core.timekeeper;
617 snapshot->real = __ktime_get_real_fast(&tk_fast_mono, &snapshot->mono);
618 snapshot->boot = snapshot->mono + ktime_to_ns(data_race(tk->offs_boot));
622 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
623 * @tk: Timekeeper to snapshot.
625 * It generally is unsafe to access the clocksource after timekeeping has been
626 * suspended, so take a snapshot of the readout base of @tk and use it as the
627 * fast timekeeper's readout base while suspended. It will return the same
628 * number of cycles every time until timekeeping is resumed at which time the
629 * proper readout base for the fast timekeeper will be restored automatically.
631 static void halt_fast_timekeeper(const struct timekeeper *tk)
633 static struct tk_read_base tkr_dummy;
634 const struct tk_read_base *tkr = &tk->tkr_mono;
636 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
637 cycles_at_suspend = tk_clock_read(tkr);
638 tkr_dummy.clock = &dummy_clock;
639 tkr_dummy.base_real = tkr->base + tk->offs_real;
640 update_fast_timekeeper(&tkr_dummy, &tk_fast_mono);
643 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
644 tkr_dummy.clock = &dummy_clock;
645 update_fast_timekeeper(&tkr_dummy, &tk_fast_raw);
648 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
650 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
652 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
656 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
658 int pvclock_gtod_register_notifier(struct notifier_block *nb)
660 struct timekeeper *tk = &tk_core.timekeeper;
664 raw_spin_lock_irqsave(&timekeeper_lock, flags);
665 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
666 update_pvclock_gtod(tk, true);
667 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
671 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
674 * pvclock_gtod_unregister_notifier - unregister a pvclock
675 * timedata update listener
677 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
682 raw_spin_lock_irqsave(&timekeeper_lock, flags);
683 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
684 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
688 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
691 * tk_update_leap_state - helper to update the next_leap_ktime
693 static inline void tk_update_leap_state(struct timekeeper *tk)
695 tk->next_leap_ktime = ntp_get_next_leap();
696 if (tk->next_leap_ktime != KTIME_MAX)
697 /* Convert to monotonic time */
698 tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real);
702 * Update the ktime_t based scalar nsec members of the timekeeper
704 static inline void tk_update_ktime_data(struct timekeeper *tk)
710 * The xtime based monotonic readout is:
711 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
712 * The ktime based monotonic readout is:
713 * nsec = base_mono + now();
714 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
716 seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
717 nsec = (u32) tk->wall_to_monotonic.tv_nsec;
718 tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
721 * The sum of the nanoseconds portions of xtime and
722 * wall_to_monotonic can be greater/equal one second. Take
723 * this into account before updating tk->ktime_sec.
725 nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
726 if (nsec >= NSEC_PER_SEC)
728 tk->ktime_sec = seconds;
730 /* Update the monotonic raw base */
731 tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC);
734 /* must hold timekeeper_lock */
735 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
737 if (action & TK_CLEAR_NTP) {
742 tk_update_leap_state(tk);
743 tk_update_ktime_data(tk);
746 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
748 tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real;
749 update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
750 update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw);
752 if (action & TK_CLOCK_WAS_SET)
753 tk->clock_was_set_seq++;
755 * The mirroring of the data to the shadow-timekeeper needs
756 * to happen last here to ensure we don't over-write the
757 * timekeeper structure on the next update with stale data
759 if (action & TK_MIRROR)
760 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
761 sizeof(tk_core.timekeeper));
765 * timekeeping_forward_now - update clock to the current time
767 * Forward the current clock to update its state since the last call to
768 * update_wall_time(). This is useful before significant clock changes,
769 * as it avoids having to deal with this time offset explicitly.
771 static void timekeeping_forward_now(struct timekeeper *tk)
773 u64 cycle_now, delta;
775 cycle_now = tk_clock_read(&tk->tkr_mono);
776 delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
777 tk->tkr_mono.cycle_last = cycle_now;
778 tk->tkr_raw.cycle_last = cycle_now;
780 tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult;
782 /* If arch requires, add in get_arch_timeoffset() */
783 tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift;
786 tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult;
788 /* If arch requires, add in get_arch_timeoffset() */
789 tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift;
791 tk_normalize_xtime(tk);
795 * ktime_get_real_ts64 - Returns the time of day in a timespec64.
796 * @ts: pointer to the timespec to be set
798 * Returns the time of day in a timespec64 (WARN if suspended).
800 void ktime_get_real_ts64(struct timespec64 *ts)
802 struct timekeeper *tk = &tk_core.timekeeper;
806 WARN_ON(timekeeping_suspended);
809 seq = read_seqcount_begin(&tk_core.seq);
811 ts->tv_sec = tk->xtime_sec;
812 nsecs = timekeeping_get_ns(&tk->tkr_mono);
814 } while (read_seqcount_retry(&tk_core.seq, seq));
817 timespec64_add_ns(ts, nsecs);
819 EXPORT_SYMBOL(ktime_get_real_ts64);
821 ktime_t ktime_get(void)
823 struct timekeeper *tk = &tk_core.timekeeper;
828 WARN_ON(timekeeping_suspended);
831 seq = read_seqcount_begin(&tk_core.seq);
832 base = tk->tkr_mono.base;
833 nsecs = timekeeping_get_ns(&tk->tkr_mono);
835 } while (read_seqcount_retry(&tk_core.seq, seq));
837 return ktime_add_ns(base, nsecs);
839 EXPORT_SYMBOL_GPL(ktime_get);
841 u32 ktime_get_resolution_ns(void)
843 struct timekeeper *tk = &tk_core.timekeeper;
847 WARN_ON(timekeeping_suspended);
850 seq = read_seqcount_begin(&tk_core.seq);
851 nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift;
852 } while (read_seqcount_retry(&tk_core.seq, seq));
856 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns);
858 static ktime_t *offsets[TK_OFFS_MAX] = {
859 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
860 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
861 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
864 ktime_t ktime_get_with_offset(enum tk_offsets offs)
866 struct timekeeper *tk = &tk_core.timekeeper;
868 ktime_t base, *offset = offsets[offs];
871 WARN_ON(timekeeping_suspended);
874 seq = read_seqcount_begin(&tk_core.seq);
875 base = ktime_add(tk->tkr_mono.base, *offset);
876 nsecs = timekeeping_get_ns(&tk->tkr_mono);
878 } while (read_seqcount_retry(&tk_core.seq, seq));
880 return ktime_add_ns(base, nsecs);
883 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
885 ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs)
887 struct timekeeper *tk = &tk_core.timekeeper;
889 ktime_t base, *offset = offsets[offs];
892 WARN_ON(timekeeping_suspended);
895 seq = read_seqcount_begin(&tk_core.seq);
896 base = ktime_add(tk->tkr_mono.base, *offset);
897 nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift;
899 } while (read_seqcount_retry(&tk_core.seq, seq));
901 return ktime_add_ns(base, nsecs);
903 EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset);
906 * ktime_mono_to_any() - convert mononotic time to any other time
907 * @tmono: time to convert.
908 * @offs: which offset to use
910 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
912 ktime_t *offset = offsets[offs];
917 seq = read_seqcount_begin(&tk_core.seq);
918 tconv = ktime_add(tmono, *offset);
919 } while (read_seqcount_retry(&tk_core.seq, seq));
923 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
926 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
928 ktime_t ktime_get_raw(void)
930 struct timekeeper *tk = &tk_core.timekeeper;
936 seq = read_seqcount_begin(&tk_core.seq);
937 base = tk->tkr_raw.base;
938 nsecs = timekeeping_get_ns(&tk->tkr_raw);
940 } while (read_seqcount_retry(&tk_core.seq, seq));
942 return ktime_add_ns(base, nsecs);
944 EXPORT_SYMBOL_GPL(ktime_get_raw);
947 * ktime_get_ts64 - get the monotonic clock in timespec64 format
948 * @ts: pointer to timespec variable
950 * The function calculates the monotonic clock from the realtime
951 * clock and the wall_to_monotonic offset and stores the result
952 * in normalized timespec64 format in the variable pointed to by @ts.
954 void ktime_get_ts64(struct timespec64 *ts)
956 struct timekeeper *tk = &tk_core.timekeeper;
957 struct timespec64 tomono;
961 WARN_ON(timekeeping_suspended);
964 seq = read_seqcount_begin(&tk_core.seq);
965 ts->tv_sec = tk->xtime_sec;
966 nsec = timekeeping_get_ns(&tk->tkr_mono);
967 tomono = tk->wall_to_monotonic;
969 } while (read_seqcount_retry(&tk_core.seq, seq));
971 ts->tv_sec += tomono.tv_sec;
973 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
975 EXPORT_SYMBOL_GPL(ktime_get_ts64);
978 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
980 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
981 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
982 * works on both 32 and 64 bit systems. On 32 bit systems the readout
983 * covers ~136 years of uptime which should be enough to prevent
984 * premature wrap arounds.
986 time64_t ktime_get_seconds(void)
988 struct timekeeper *tk = &tk_core.timekeeper;
990 WARN_ON(timekeeping_suspended);
991 return tk->ktime_sec;
993 EXPORT_SYMBOL_GPL(ktime_get_seconds);
996 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
998 * Returns the wall clock seconds since 1970. This replaces the
999 * get_seconds() interface which is not y2038 safe on 32bit systems.
1001 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
1002 * 32bit systems the access must be protected with the sequence
1003 * counter to provide "atomic" access to the 64bit tk->xtime_sec
1006 time64_t ktime_get_real_seconds(void)
1008 struct timekeeper *tk = &tk_core.timekeeper;
1012 if (IS_ENABLED(CONFIG_64BIT))
1013 return tk->xtime_sec;
1016 seq = read_seqcount_begin(&tk_core.seq);
1017 seconds = tk->xtime_sec;
1019 } while (read_seqcount_retry(&tk_core.seq, seq));
1023 EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
1026 * __ktime_get_real_seconds - The same as ktime_get_real_seconds
1027 * but without the sequence counter protect. This internal function
1028 * is called just when timekeeping lock is already held.
1030 noinstr time64_t __ktime_get_real_seconds(void)
1032 struct timekeeper *tk = &tk_core.timekeeper;
1034 return tk->xtime_sec;
1038 * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
1039 * @systime_snapshot: pointer to struct receiving the system time snapshot
1041 void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot)
1043 struct timekeeper *tk = &tk_core.timekeeper;
1051 WARN_ON_ONCE(timekeeping_suspended);
1054 seq = read_seqcount_begin(&tk_core.seq);
1055 now = tk_clock_read(&tk->tkr_mono);
1056 systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq;
1057 systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq;
1058 base_real = ktime_add(tk->tkr_mono.base,
1059 tk_core.timekeeper.offs_real);
1060 base_raw = tk->tkr_raw.base;
1061 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now);
1062 nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now);
1063 } while (read_seqcount_retry(&tk_core.seq, seq));
1065 systime_snapshot->cycles = now;
1066 systime_snapshot->real = ktime_add_ns(base_real, nsec_real);
1067 systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw);
1069 EXPORT_SYMBOL_GPL(ktime_get_snapshot);
1071 /* Scale base by mult/div checking for overflow */
1072 static int scale64_check_overflow(u64 mult, u64 div, u64 *base)
1076 tmp = div64_u64_rem(*base, div, &rem);
1078 if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) ||
1079 ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem)))
1083 rem = div64_u64(rem * mult, div);
1089 * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1090 * @history: Snapshot representing start of history
1091 * @partial_history_cycles: Cycle offset into history (fractional part)
1092 * @total_history_cycles: Total history length in cycles
1093 * @discontinuity: True indicates clock was set on history period
1094 * @ts: Cross timestamp that should be adjusted using
1095 * partial/total ratio
1097 * Helper function used by get_device_system_crosststamp() to correct the
1098 * crosstimestamp corresponding to the start of the current interval to the
1099 * system counter value (timestamp point) provided by the driver. The
1100 * total_history_* quantities are the total history starting at the provided
1101 * reference point and ending at the start of the current interval. The cycle
1102 * count between the driver timestamp point and the start of the current
1103 * interval is partial_history_cycles.
1105 static int adjust_historical_crosststamp(struct system_time_snapshot *history,
1106 u64 partial_history_cycles,
1107 u64 total_history_cycles,
1109 struct system_device_crosststamp *ts)
1111 struct timekeeper *tk = &tk_core.timekeeper;
1112 u64 corr_raw, corr_real;
1113 bool interp_forward;
1116 if (total_history_cycles == 0 || partial_history_cycles == 0)
1119 /* Interpolate shortest distance from beginning or end of history */
1120 interp_forward = partial_history_cycles > total_history_cycles / 2;
1121 partial_history_cycles = interp_forward ?
1122 total_history_cycles - partial_history_cycles :
1123 partial_history_cycles;
1126 * Scale the monotonic raw time delta by:
1127 * partial_history_cycles / total_history_cycles
1129 corr_raw = (u64)ktime_to_ns(
1130 ktime_sub(ts->sys_monoraw, history->raw));
1131 ret = scale64_check_overflow(partial_history_cycles,
1132 total_history_cycles, &corr_raw);
1137 * If there is a discontinuity in the history, scale monotonic raw
1139 * mult(real)/mult(raw) yielding the realtime correction
1140 * Otherwise, calculate the realtime correction similar to monotonic
1143 if (discontinuity) {
1144 corr_real = mul_u64_u32_div
1145 (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult);
1147 corr_real = (u64)ktime_to_ns(
1148 ktime_sub(ts->sys_realtime, history->real));
1149 ret = scale64_check_overflow(partial_history_cycles,
1150 total_history_cycles, &corr_real);
1155 /* Fixup monotonic raw and real time time values */
1156 if (interp_forward) {
1157 ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw);
1158 ts->sys_realtime = ktime_add_ns(history->real, corr_real);
1160 ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw);
1161 ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real);
1168 * cycle_between - true if test occurs chronologically between before and after
1170 static bool cycle_between(u64 before, u64 test, u64 after)
1172 if (test > before && test < after)
1174 if (test < before && before > after)
1180 * get_device_system_crosststamp - Synchronously capture system/device timestamp
1181 * @get_time_fn: Callback to get simultaneous device time and
1182 * system counter from the device driver
1183 * @ctx: Context passed to get_time_fn()
1184 * @history_begin: Historical reference point used to interpolate system
1185 * time when counter provided by the driver is before the current interval
1186 * @xtstamp: Receives simultaneously captured system and device time
1188 * Reads a timestamp from a device and correlates it to system time
1190 int get_device_system_crosststamp(int (*get_time_fn)
1191 (ktime_t *device_time,
1192 struct system_counterval_t *sys_counterval,
1195 struct system_time_snapshot *history_begin,
1196 struct system_device_crosststamp *xtstamp)
1198 struct system_counterval_t system_counterval;
1199 struct timekeeper *tk = &tk_core.timekeeper;
1200 u64 cycles, now, interval_start;
1201 unsigned int clock_was_set_seq = 0;
1202 ktime_t base_real, base_raw;
1203 u64 nsec_real, nsec_raw;
1204 u8 cs_was_changed_seq;
1210 seq = read_seqcount_begin(&tk_core.seq);
1212 * Try to synchronously capture device time and a system
1213 * counter value calling back into the device driver
1215 ret = get_time_fn(&xtstamp->device, &system_counterval, ctx);
1220 * Verify that the clocksource associated with the captured
1221 * system counter value is the same as the currently installed
1222 * timekeeper clocksource
1224 if (tk->tkr_mono.clock != system_counterval.cs)
1226 cycles = system_counterval.cycles;
1229 * Check whether the system counter value provided by the
1230 * device driver is on the current timekeeping interval.
1232 now = tk_clock_read(&tk->tkr_mono);
1233 interval_start = tk->tkr_mono.cycle_last;
1234 if (!cycle_between(interval_start, cycles, now)) {
1235 clock_was_set_seq = tk->clock_was_set_seq;
1236 cs_was_changed_seq = tk->cs_was_changed_seq;
1237 cycles = interval_start;
1243 base_real = ktime_add(tk->tkr_mono.base,
1244 tk_core.timekeeper.offs_real);
1245 base_raw = tk->tkr_raw.base;
1247 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono,
1248 system_counterval.cycles);
1249 nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw,
1250 system_counterval.cycles);
1251 } while (read_seqcount_retry(&tk_core.seq, seq));
1253 xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
1254 xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw);
1257 * Interpolate if necessary, adjusting back from the start of the
1261 u64 partial_history_cycles, total_history_cycles;
1265 * Check that the counter value occurs after the provided
1266 * history reference and that the history doesn't cross a
1267 * clocksource change
1269 if (!history_begin ||
1270 !cycle_between(history_begin->cycles,
1271 system_counterval.cycles, cycles) ||
1272 history_begin->cs_was_changed_seq != cs_was_changed_seq)
1274 partial_history_cycles = cycles - system_counterval.cycles;
1275 total_history_cycles = cycles - history_begin->cycles;
1277 history_begin->clock_was_set_seq != clock_was_set_seq;
1279 ret = adjust_historical_crosststamp(history_begin,
1280 partial_history_cycles,
1281 total_history_cycles,
1282 discontinuity, xtstamp);
1289 EXPORT_SYMBOL_GPL(get_device_system_crosststamp);
1292 * do_settimeofday64 - Sets the time of day.
1293 * @ts: pointer to the timespec64 variable containing the new time
1295 * Sets the time of day to the new time and update NTP and notify hrtimers
1297 int do_settimeofday64(const struct timespec64 *ts)
1299 struct timekeeper *tk = &tk_core.timekeeper;
1300 struct timespec64 ts_delta, xt;
1301 unsigned long flags;
1304 if (!timespec64_valid_settod(ts))
1307 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1308 write_seqcount_begin(&tk_core.seq);
1310 timekeeping_forward_now(tk);
1313 ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
1314 ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
1316 if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) {
1321 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
1323 tk_set_xtime(tk, ts);
1325 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1327 write_seqcount_end(&tk_core.seq);
1328 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1330 /* signal hrtimers about time change */
1334 audit_tk_injoffset(ts_delta);
1338 EXPORT_SYMBOL(do_settimeofday64);
1341 * timekeeping_inject_offset - Adds or subtracts from the current time.
1342 * @tv: pointer to the timespec variable containing the offset
1344 * Adds or subtracts an offset value from the current time.
1346 static int timekeeping_inject_offset(const struct timespec64 *ts)
1348 struct timekeeper *tk = &tk_core.timekeeper;
1349 unsigned long flags;
1350 struct timespec64 tmp;
1353 if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
1356 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1357 write_seqcount_begin(&tk_core.seq);
1359 timekeeping_forward_now(tk);
1361 /* Make sure the proposed value is valid */
1362 tmp = timespec64_add(tk_xtime(tk), *ts);
1363 if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 ||
1364 !timespec64_valid_settod(&tmp)) {
1369 tk_xtime_add(tk, ts);
1370 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts));
1372 error: /* even if we error out, we forwarded the time, so call update */
1373 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1375 write_seqcount_end(&tk_core.seq);
1376 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1378 /* signal hrtimers about time change */
1385 * Indicates if there is an offset between the system clock and the hardware
1386 * clock/persistent clock/rtc.
1388 int persistent_clock_is_local;
1391 * Adjust the time obtained from the CMOS to be UTC time instead of
1394 * This is ugly, but preferable to the alternatives. Otherwise we
1395 * would either need to write a program to do it in /etc/rc (and risk
1396 * confusion if the program gets run more than once; it would also be
1397 * hard to make the program warp the clock precisely n hours) or
1398 * compile in the timezone information into the kernel. Bad, bad....
1402 * The best thing to do is to keep the CMOS clock in universal time (UTC)
1403 * as real UNIX machines always do it. This avoids all headaches about
1404 * daylight saving times and warping kernel clocks.
1406 void timekeeping_warp_clock(void)
1408 if (sys_tz.tz_minuteswest != 0) {
1409 struct timespec64 adjust;
1411 persistent_clock_is_local = 1;
1412 adjust.tv_sec = sys_tz.tz_minuteswest * 60;
1414 timekeeping_inject_offset(&adjust);
1419 * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1422 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
1424 tk->tai_offset = tai_offset;
1425 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
1429 * change_clocksource - Swaps clocksources if a new one is available
1431 * Accumulates current time interval and initializes new clocksource
1433 static int change_clocksource(void *data)
1435 struct timekeeper *tk = &tk_core.timekeeper;
1436 struct clocksource *new, *old;
1437 unsigned long flags;
1439 new = (struct clocksource *) data;
1441 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1442 write_seqcount_begin(&tk_core.seq);
1444 timekeeping_forward_now(tk);
1446 * If the cs is in module, get a module reference. Succeeds
1447 * for built-in code (owner == NULL) as well.
1449 if (try_module_get(new->owner)) {
1450 if (!new->enable || new->enable(new) == 0) {
1451 old = tk->tkr_mono.clock;
1452 tk_setup_internals(tk, new);
1455 module_put(old->owner);
1457 module_put(new->owner);
1460 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1462 write_seqcount_end(&tk_core.seq);
1463 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1469 * timekeeping_notify - Install a new clock source
1470 * @clock: pointer to the clock source
1472 * This function is called from clocksource.c after a new, better clock
1473 * source has been registered. The caller holds the clocksource_mutex.
1475 int timekeeping_notify(struct clocksource *clock)
1477 struct timekeeper *tk = &tk_core.timekeeper;
1479 if (tk->tkr_mono.clock == clock)
1481 stop_machine(change_clocksource, clock, NULL);
1482 tick_clock_notify();
1483 return tk->tkr_mono.clock == clock ? 0 : -1;
1487 * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1488 * @ts: pointer to the timespec64 to be set
1490 * Returns the raw monotonic time (completely un-modified by ntp)
1492 void ktime_get_raw_ts64(struct timespec64 *ts)
1494 struct timekeeper *tk = &tk_core.timekeeper;
1499 seq = read_seqcount_begin(&tk_core.seq);
1500 ts->tv_sec = tk->raw_sec;
1501 nsecs = timekeeping_get_ns(&tk->tkr_raw);
1503 } while (read_seqcount_retry(&tk_core.seq, seq));
1506 timespec64_add_ns(ts, nsecs);
1508 EXPORT_SYMBOL(ktime_get_raw_ts64);
1512 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1514 int timekeeping_valid_for_hres(void)
1516 struct timekeeper *tk = &tk_core.timekeeper;
1521 seq = read_seqcount_begin(&tk_core.seq);
1523 ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1525 } while (read_seqcount_retry(&tk_core.seq, seq));
1531 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1533 u64 timekeeping_max_deferment(void)
1535 struct timekeeper *tk = &tk_core.timekeeper;
1540 seq = read_seqcount_begin(&tk_core.seq);
1542 ret = tk->tkr_mono.clock->max_idle_ns;
1544 } while (read_seqcount_retry(&tk_core.seq, seq));
1550 * read_persistent_clock64 - Return time from the persistent clock.
1552 * Weak dummy function for arches that do not yet support it.
1553 * Reads the time from the battery backed persistent clock.
1554 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1556 * XXX - Do be sure to remove it once all arches implement it.
1558 void __weak read_persistent_clock64(struct timespec64 *ts)
1565 * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1568 * Weak dummy function for arches that do not yet support it.
1569 * wall_time - current time as returned by persistent clock
1570 * boot_offset - offset that is defined as wall_time - boot_time
1571 * The default function calculates offset based on the current value of
1572 * local_clock(). This way architectures that support sched_clock() but don't
1573 * support dedicated boot time clock will provide the best estimate of the
1577 read_persistent_wall_and_boot_offset(struct timespec64 *wall_time,
1578 struct timespec64 *boot_offset)
1580 read_persistent_clock64(wall_time);
1581 *boot_offset = ns_to_timespec64(local_clock());
1585 * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1587 * The flag starts of false and is only set when a suspend reaches
1588 * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1589 * timekeeper clocksource is not stopping across suspend and has been
1590 * used to update sleep time. If the timekeeper clocksource has stopped
1591 * then the flag stays true and is used by the RTC resume code to decide
1592 * whether sleeptime must be injected and if so the flag gets false then.
1594 * If a suspend fails before reaching timekeeping_resume() then the flag
1595 * stays false and prevents erroneous sleeptime injection.
1597 static bool suspend_timing_needed;
1599 /* Flag for if there is a persistent clock on this platform */
1600 static bool persistent_clock_exists;
1603 * timekeeping_init - Initializes the clocksource and common timekeeping values
1605 void __init timekeeping_init(void)
1607 struct timespec64 wall_time, boot_offset, wall_to_mono;
1608 struct timekeeper *tk = &tk_core.timekeeper;
1609 struct clocksource *clock;
1610 unsigned long flags;
1612 read_persistent_wall_and_boot_offset(&wall_time, &boot_offset);
1613 if (timespec64_valid_settod(&wall_time) &&
1614 timespec64_to_ns(&wall_time) > 0) {
1615 persistent_clock_exists = true;
1616 } else if (timespec64_to_ns(&wall_time) != 0) {
1617 pr_warn("Persistent clock returned invalid value");
1618 wall_time = (struct timespec64){0};
1621 if (timespec64_compare(&wall_time, &boot_offset) < 0)
1622 boot_offset = (struct timespec64){0};
1625 * We want set wall_to_mono, so the following is true:
1626 * wall time + wall_to_mono = boot time
1628 wall_to_mono = timespec64_sub(boot_offset, wall_time);
1630 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1631 write_seqcount_begin(&tk_core.seq);
1634 clock = clocksource_default_clock();
1636 clock->enable(clock);
1637 tk_setup_internals(tk, clock);
1639 tk_set_xtime(tk, &wall_time);
1642 tk_set_wall_to_mono(tk, wall_to_mono);
1644 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1646 write_seqcount_end(&tk_core.seq);
1647 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1650 /* time in seconds when suspend began for persistent clock */
1651 static struct timespec64 timekeeping_suspend_time;
1654 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1655 * @delta: pointer to a timespec delta value
1657 * Takes a timespec offset measuring a suspend interval and properly
1658 * adds the sleep offset to the timekeeping variables.
1660 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1661 const struct timespec64 *delta)
1663 if (!timespec64_valid_strict(delta)) {
1664 printk_deferred(KERN_WARNING
1665 "__timekeeping_inject_sleeptime: Invalid "
1666 "sleep delta value!\n");
1669 tk_xtime_add(tk, delta);
1670 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1671 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1672 tk_debug_account_sleep_time(delta);
1675 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1677 * We have three kinds of time sources to use for sleep time
1678 * injection, the preference order is:
1679 * 1) non-stop clocksource
1680 * 2) persistent clock (ie: RTC accessible when irqs are off)
1683 * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1684 * If system has neither 1) nor 2), 3) will be used finally.
1687 * If timekeeping has injected sleeptime via either 1) or 2),
1688 * 3) becomes needless, so in this case we don't need to call
1689 * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1692 bool timekeeping_rtc_skipresume(void)
1694 return !suspend_timing_needed;
1698 * 1) can be determined whether to use or not only when doing
1699 * timekeeping_resume() which is invoked after rtc_suspend(),
1700 * so we can't skip rtc_suspend() surely if system has 1).
1702 * But if system has 2), 2) will definitely be used, so in this
1703 * case we don't need to call rtc_suspend(), and this is what
1704 * timekeeping_rtc_skipsuspend() means.
1706 bool timekeeping_rtc_skipsuspend(void)
1708 return persistent_clock_exists;
1712 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1713 * @delta: pointer to a timespec64 delta value
1715 * This hook is for architectures that cannot support read_persistent_clock64
1716 * because their RTC/persistent clock is only accessible when irqs are enabled.
1717 * and also don't have an effective nonstop clocksource.
1719 * This function should only be called by rtc_resume(), and allows
1720 * a suspend offset to be injected into the timekeeping values.
1722 void timekeeping_inject_sleeptime64(const struct timespec64 *delta)
1724 struct timekeeper *tk = &tk_core.timekeeper;
1725 unsigned long flags;
1727 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1728 write_seqcount_begin(&tk_core.seq);
1730 suspend_timing_needed = false;
1732 timekeeping_forward_now(tk);
1734 __timekeeping_inject_sleeptime(tk, delta);
1736 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1738 write_seqcount_end(&tk_core.seq);
1739 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1741 /* signal hrtimers about time change */
1747 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1749 void timekeeping_resume(void)
1751 struct timekeeper *tk = &tk_core.timekeeper;
1752 struct clocksource *clock = tk->tkr_mono.clock;
1753 unsigned long flags;
1754 struct timespec64 ts_new, ts_delta;
1755 u64 cycle_now, nsec;
1756 bool inject_sleeptime = false;
1758 read_persistent_clock64(&ts_new);
1760 clockevents_resume();
1761 clocksource_resume();
1763 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1764 write_seqcount_begin(&tk_core.seq);
1767 * After system resumes, we need to calculate the suspended time and
1768 * compensate it for the OS time. There are 3 sources that could be
1769 * used: Nonstop clocksource during suspend, persistent clock and rtc
1772 * One specific platform may have 1 or 2 or all of them, and the
1773 * preference will be:
1774 * suspend-nonstop clocksource -> persistent clock -> rtc
1775 * The less preferred source will only be tried if there is no better
1776 * usable source. The rtc part is handled separately in rtc core code.
1778 cycle_now = tk_clock_read(&tk->tkr_mono);
1779 nsec = clocksource_stop_suspend_timing(clock, cycle_now);
1781 ts_delta = ns_to_timespec64(nsec);
1782 inject_sleeptime = true;
1783 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1784 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1785 inject_sleeptime = true;
1788 if (inject_sleeptime) {
1789 suspend_timing_needed = false;
1790 __timekeeping_inject_sleeptime(tk, &ts_delta);
1793 /* Re-base the last cycle value */
1794 tk->tkr_mono.cycle_last = cycle_now;
1795 tk->tkr_raw.cycle_last = cycle_now;
1798 timekeeping_suspended = 0;
1799 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1800 write_seqcount_end(&tk_core.seq);
1801 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1803 touch_softlockup_watchdog();
1809 int timekeeping_suspend(void)
1811 struct timekeeper *tk = &tk_core.timekeeper;
1812 unsigned long flags;
1813 struct timespec64 delta, delta_delta;
1814 static struct timespec64 old_delta;
1815 struct clocksource *curr_clock;
1818 read_persistent_clock64(&timekeeping_suspend_time);
1821 * On some systems the persistent_clock can not be detected at
1822 * timekeeping_init by its return value, so if we see a valid
1823 * value returned, update the persistent_clock_exists flag.
1825 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1826 persistent_clock_exists = true;
1828 suspend_timing_needed = true;
1830 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1831 write_seqcount_begin(&tk_core.seq);
1832 timekeeping_forward_now(tk);
1833 timekeeping_suspended = 1;
1836 * Since we've called forward_now, cycle_last stores the value
1837 * just read from the current clocksource. Save this to potentially
1838 * use in suspend timing.
1840 curr_clock = tk->tkr_mono.clock;
1841 cycle_now = tk->tkr_mono.cycle_last;
1842 clocksource_start_suspend_timing(curr_clock, cycle_now);
1844 if (persistent_clock_exists) {
1846 * To avoid drift caused by repeated suspend/resumes,
1847 * which each can add ~1 second drift error,
1848 * try to compensate so the difference in system time
1849 * and persistent_clock time stays close to constant.
1851 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1852 delta_delta = timespec64_sub(delta, old_delta);
1853 if (abs(delta_delta.tv_sec) >= 2) {
1855 * if delta_delta is too large, assume time correction
1856 * has occurred and set old_delta to the current delta.
1860 /* Otherwise try to adjust old_system to compensate */
1861 timekeeping_suspend_time =
1862 timespec64_add(timekeeping_suspend_time, delta_delta);
1866 timekeeping_update(tk, TK_MIRROR);
1867 halt_fast_timekeeper(tk);
1868 write_seqcount_end(&tk_core.seq);
1869 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1872 clocksource_suspend();
1873 clockevents_suspend();
1878 /* sysfs resume/suspend bits for timekeeping */
1879 static struct syscore_ops timekeeping_syscore_ops = {
1880 .resume = timekeeping_resume,
1881 .suspend = timekeeping_suspend,
1884 static int __init timekeeping_init_ops(void)
1886 register_syscore_ops(&timekeeping_syscore_ops);
1889 device_initcall(timekeeping_init_ops);
1892 * Apply a multiplier adjustment to the timekeeper
1894 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1898 s64 interval = tk->cycle_interval;
1900 if (mult_adj == 0) {
1902 } else if (mult_adj == -1) {
1903 interval = -interval;
1905 } else if (mult_adj != 1) {
1906 interval *= mult_adj;
1911 * So the following can be confusing.
1913 * To keep things simple, lets assume mult_adj == 1 for now.
1915 * When mult_adj != 1, remember that the interval and offset values
1916 * have been appropriately scaled so the math is the same.
1918 * The basic idea here is that we're increasing the multiplier
1919 * by one, this causes the xtime_interval to be incremented by
1920 * one cycle_interval. This is because:
1921 * xtime_interval = cycle_interval * mult
1922 * So if mult is being incremented by one:
1923 * xtime_interval = cycle_interval * (mult + 1)
1925 * xtime_interval = (cycle_interval * mult) + cycle_interval
1926 * Which can be shortened to:
1927 * xtime_interval += cycle_interval
1929 * So offset stores the non-accumulated cycles. Thus the current
1930 * time (in shifted nanoseconds) is:
1931 * now = (offset * adj) + xtime_nsec
1932 * Now, even though we're adjusting the clock frequency, we have
1933 * to keep time consistent. In other words, we can't jump back
1934 * in time, and we also want to avoid jumping forward in time.
1936 * So given the same offset value, we need the time to be the same
1937 * both before and after the freq adjustment.
1938 * now = (offset * adj_1) + xtime_nsec_1
1939 * now = (offset * adj_2) + xtime_nsec_2
1941 * (offset * adj_1) + xtime_nsec_1 =
1942 * (offset * adj_2) + xtime_nsec_2
1946 * (offset * adj_1) + xtime_nsec_1 =
1947 * (offset * (adj_1+1)) + xtime_nsec_2
1948 * (offset * adj_1) + xtime_nsec_1 =
1949 * (offset * adj_1) + offset + xtime_nsec_2
1950 * Canceling the sides:
1951 * xtime_nsec_1 = offset + xtime_nsec_2
1953 * xtime_nsec_2 = xtime_nsec_1 - offset
1954 * Which simplfies to:
1955 * xtime_nsec -= offset
1957 if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) {
1958 /* NTP adjustment caused clocksource mult overflow */
1963 tk->tkr_mono.mult += mult_adj;
1964 tk->xtime_interval += interval;
1965 tk->tkr_mono.xtime_nsec -= offset;
1969 * Adjust the timekeeper's multiplier to the correct frequency
1970 * and also to reduce the accumulated error value.
1972 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1977 * Determine the multiplier from the current NTP tick length.
1978 * Avoid expensive division when the tick length doesn't change.
1980 if (likely(tk->ntp_tick == ntp_tick_length())) {
1981 mult = tk->tkr_mono.mult - tk->ntp_err_mult;
1983 tk->ntp_tick = ntp_tick_length();
1984 mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) -
1985 tk->xtime_remainder, tk->cycle_interval);
1989 * If the clock is behind the NTP time, increase the multiplier by 1
1990 * to catch up with it. If it's ahead and there was a remainder in the
1991 * tick division, the clock will slow down. Otherwise it will stay
1992 * ahead until the tick length changes to a non-divisible value.
1994 tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0;
1995 mult += tk->ntp_err_mult;
1997 timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult);
1999 if (unlikely(tk->tkr_mono.clock->maxadj &&
2000 (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult)
2001 > tk->tkr_mono.clock->maxadj))) {
2002 printk_once(KERN_WARNING
2003 "Adjusting %s more than 11%% (%ld vs %ld)\n",
2004 tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult,
2005 (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj);
2009 * It may be possible that when we entered this function, xtime_nsec
2010 * was very small. Further, if we're slightly speeding the clocksource
2011 * in the code above, its possible the required corrective factor to
2012 * xtime_nsec could cause it to underflow.
2014 * Now, since we have already accumulated the second and the NTP
2015 * subsystem has been notified via second_overflow(), we need to skip
2018 if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) {
2019 tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC <<
2022 tk->skip_second_overflow = 1;
2027 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
2029 * Helper function that accumulates the nsecs greater than a second
2030 * from the xtime_nsec field to the xtime_secs field.
2031 * It also calls into the NTP code to handle leapsecond processing.
2034 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
2036 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
2037 unsigned int clock_set = 0;
2039 while (tk->tkr_mono.xtime_nsec >= nsecps) {
2042 tk->tkr_mono.xtime_nsec -= nsecps;
2046 * Skip NTP update if this second was accumulated before,
2047 * i.e. xtime_nsec underflowed in timekeeping_adjust()
2049 if (unlikely(tk->skip_second_overflow)) {
2050 tk->skip_second_overflow = 0;
2054 /* Figure out if its a leap sec and apply if needed */
2055 leap = second_overflow(tk->xtime_sec);
2056 if (unlikely(leap)) {
2057 struct timespec64 ts;
2059 tk->xtime_sec += leap;
2063 tk_set_wall_to_mono(tk,
2064 timespec64_sub(tk->wall_to_monotonic, ts));
2066 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
2068 clock_set = TK_CLOCK_WAS_SET;
2075 * logarithmic_accumulation - shifted accumulation of cycles
2077 * This functions accumulates a shifted interval of cycles into
2078 * a shifted interval nanoseconds. Allows for O(log) accumulation
2081 * Returns the unconsumed cycles.
2083 static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
2084 u32 shift, unsigned int *clock_set)
2086 u64 interval = tk->cycle_interval << shift;
2089 /* If the offset is smaller than a shifted interval, do nothing */
2090 if (offset < interval)
2093 /* Accumulate one shifted interval */
2095 tk->tkr_mono.cycle_last += interval;
2096 tk->tkr_raw.cycle_last += interval;
2098 tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift;
2099 *clock_set |= accumulate_nsecs_to_secs(tk);
2101 /* Accumulate raw time */
2102 tk->tkr_raw.xtime_nsec += tk->raw_interval << shift;
2103 snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
2104 while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) {
2105 tk->tkr_raw.xtime_nsec -= snsec_per_sec;
2109 /* Accumulate error between NTP and clock interval */
2110 tk->ntp_error += tk->ntp_tick << shift;
2111 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
2112 (tk->ntp_error_shift + shift);
2118 * timekeeping_advance - Updates the timekeeper to the current time and
2119 * current NTP tick length
2121 static void timekeeping_advance(enum timekeeping_adv_mode mode)
2123 struct timekeeper *real_tk = &tk_core.timekeeper;
2124 struct timekeeper *tk = &shadow_timekeeper;
2126 int shift = 0, maxshift;
2127 unsigned int clock_set = 0;
2128 unsigned long flags;
2130 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2132 /* Make sure we're fully resumed: */
2133 if (unlikely(timekeeping_suspended))
2136 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
2137 offset = real_tk->cycle_interval;
2139 if (mode != TK_ADV_TICK)
2142 offset = clocksource_delta(tk_clock_read(&tk->tkr_mono),
2143 tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
2145 /* Check if there's really nothing to do */
2146 if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK)
2150 /* Do some additional sanity checking */
2151 timekeeping_check_update(tk, offset);
2154 * With NO_HZ we may have to accumulate many cycle_intervals
2155 * (think "ticks") worth of time at once. To do this efficiently,
2156 * we calculate the largest doubling multiple of cycle_intervals
2157 * that is smaller than the offset. We then accumulate that
2158 * chunk in one go, and then try to consume the next smaller
2161 shift = ilog2(offset) - ilog2(tk->cycle_interval);
2162 shift = max(0, shift);
2163 /* Bound shift to one less than what overflows tick_length */
2164 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
2165 shift = min(shift, maxshift);
2166 while (offset >= tk->cycle_interval) {
2167 offset = logarithmic_accumulation(tk, offset, shift,
2169 if (offset < tk->cycle_interval<<shift)
2173 /* Adjust the multiplier to correct NTP error */
2174 timekeeping_adjust(tk, offset);
2177 * Finally, make sure that after the rounding
2178 * xtime_nsec isn't larger than NSEC_PER_SEC
2180 clock_set |= accumulate_nsecs_to_secs(tk);
2182 write_seqcount_begin(&tk_core.seq);
2184 * Update the real timekeeper.
2186 * We could avoid this memcpy by switching pointers, but that
2187 * requires changes to all other timekeeper usage sites as
2188 * well, i.e. move the timekeeper pointer getter into the
2189 * spinlocked/seqcount protected sections. And we trade this
2190 * memcpy under the tk_core.seq against one before we start
2193 timekeeping_update(tk, clock_set);
2194 memcpy(real_tk, tk, sizeof(*tk));
2195 /* The memcpy must come last. Do not put anything here! */
2196 write_seqcount_end(&tk_core.seq);
2198 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2200 /* Have to call _delayed version, since in irq context*/
2201 clock_was_set_delayed();
2205 * update_wall_time - Uses the current clocksource to increment the wall time
2208 void update_wall_time(void)
2210 timekeeping_advance(TK_ADV_TICK);
2214 * getboottime64 - Return the real time of system boot.
2215 * @ts: pointer to the timespec64 to be set
2217 * Returns the wall-time of boot in a timespec64.
2219 * This is based on the wall_to_monotonic offset and the total suspend
2220 * time. Calls to settimeofday will affect the value returned (which
2221 * basically means that however wrong your real time clock is at boot time,
2222 * you get the right time here).
2224 void getboottime64(struct timespec64 *ts)
2226 struct timekeeper *tk = &tk_core.timekeeper;
2227 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
2229 *ts = ktime_to_timespec64(t);
2231 EXPORT_SYMBOL_GPL(getboottime64);
2233 void ktime_get_coarse_real_ts64(struct timespec64 *ts)
2235 struct timekeeper *tk = &tk_core.timekeeper;
2239 seq = read_seqcount_begin(&tk_core.seq);
2242 } while (read_seqcount_retry(&tk_core.seq, seq));
2244 EXPORT_SYMBOL(ktime_get_coarse_real_ts64);
2246 void ktime_get_coarse_ts64(struct timespec64 *ts)
2248 struct timekeeper *tk = &tk_core.timekeeper;
2249 struct timespec64 now, mono;
2253 seq = read_seqcount_begin(&tk_core.seq);
2256 mono = tk->wall_to_monotonic;
2257 } while (read_seqcount_retry(&tk_core.seq, seq));
2259 set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec,
2260 now.tv_nsec + mono.tv_nsec);
2262 EXPORT_SYMBOL(ktime_get_coarse_ts64);
2265 * Must hold jiffies_lock
2267 void do_timer(unsigned long ticks)
2269 jiffies_64 += ticks;
2274 * ktime_get_update_offsets_now - hrtimer helper
2275 * @cwsseq: pointer to check and store the clock was set sequence number
2276 * @offs_real: pointer to storage for monotonic -> realtime offset
2277 * @offs_boot: pointer to storage for monotonic -> boottime offset
2278 * @offs_tai: pointer to storage for monotonic -> clock tai offset
2280 * Returns current monotonic time and updates the offsets if the
2281 * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2284 * Called from hrtimer_interrupt() or retrigger_next_event()
2286 ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real,
2287 ktime_t *offs_boot, ktime_t *offs_tai)
2289 struct timekeeper *tk = &tk_core.timekeeper;
2295 seq = read_seqcount_begin(&tk_core.seq);
2297 base = tk->tkr_mono.base;
2298 nsecs = timekeeping_get_ns(&tk->tkr_mono);
2299 base = ktime_add_ns(base, nsecs);
2301 if (*cwsseq != tk->clock_was_set_seq) {
2302 *cwsseq = tk->clock_was_set_seq;
2303 *offs_real = tk->offs_real;
2304 *offs_boot = tk->offs_boot;
2305 *offs_tai = tk->offs_tai;
2308 /* Handle leapsecond insertion adjustments */
2309 if (unlikely(base >= tk->next_leap_ktime))
2310 *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0));
2312 } while (read_seqcount_retry(&tk_core.seq, seq));
2318 * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2320 static int timekeeping_validate_timex(const struct __kernel_timex *txc)
2322 if (txc->modes & ADJ_ADJTIME) {
2323 /* singleshot must not be used with any other mode bits */
2324 if (!(txc->modes & ADJ_OFFSET_SINGLESHOT))
2326 if (!(txc->modes & ADJ_OFFSET_READONLY) &&
2327 !capable(CAP_SYS_TIME))
2330 /* In order to modify anything, you gotta be super-user! */
2331 if (txc->modes && !capable(CAP_SYS_TIME))
2334 * if the quartz is off by more than 10% then
2335 * something is VERY wrong!
2337 if (txc->modes & ADJ_TICK &&
2338 (txc->tick < 900000/USER_HZ ||
2339 txc->tick > 1100000/USER_HZ))
2343 if (txc->modes & ADJ_SETOFFSET) {
2344 /* In order to inject time, you gotta be super-user! */
2345 if (!capable(CAP_SYS_TIME))
2349 * Validate if a timespec/timeval used to inject a time
2350 * offset is valid. Offsets can be postive or negative, so
2351 * we don't check tv_sec. The value of the timeval/timespec
2352 * is the sum of its fields,but *NOTE*:
2353 * The field tv_usec/tv_nsec must always be non-negative and
2354 * we can't have more nanoseconds/microseconds than a second.
2356 if (txc->time.tv_usec < 0)
2359 if (txc->modes & ADJ_NANO) {
2360 if (txc->time.tv_usec >= NSEC_PER_SEC)
2363 if (txc->time.tv_usec >= USEC_PER_SEC)
2369 * Check for potential multiplication overflows that can
2370 * only happen on 64-bit systems:
2372 if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) {
2373 if (LLONG_MIN / PPM_SCALE > txc->freq)
2375 if (LLONG_MAX / PPM_SCALE < txc->freq)
2384 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2386 int do_adjtimex(struct __kernel_timex *txc)
2388 struct timekeeper *tk = &tk_core.timekeeper;
2389 struct audit_ntp_data ad;
2390 unsigned long flags;
2391 struct timespec64 ts;
2395 /* Validate the data before disabling interrupts */
2396 ret = timekeeping_validate_timex(txc);
2400 if (txc->modes & ADJ_SETOFFSET) {
2401 struct timespec64 delta;
2402 delta.tv_sec = txc->time.tv_sec;
2403 delta.tv_nsec = txc->time.tv_usec;
2404 if (!(txc->modes & ADJ_NANO))
2405 delta.tv_nsec *= 1000;
2406 ret = timekeeping_inject_offset(&delta);
2410 audit_tk_injoffset(delta);
2413 audit_ntp_init(&ad);
2415 ktime_get_real_ts64(&ts);
2417 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2418 write_seqcount_begin(&tk_core.seq);
2420 orig_tai = tai = tk->tai_offset;
2421 ret = __do_adjtimex(txc, &ts, &tai, &ad);
2423 if (tai != orig_tai) {
2424 __timekeeping_set_tai_offset(tk, tai);
2425 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
2427 tk_update_leap_state(tk);
2429 write_seqcount_end(&tk_core.seq);
2430 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2434 /* Update the multiplier immediately if frequency was set directly */
2435 if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK))
2436 timekeeping_advance(TK_ADV_FREQ);
2438 if (tai != orig_tai)
2441 ntp_notify_cmos_timer();
2446 #ifdef CONFIG_NTP_PPS
2448 * hardpps() - Accessor function to NTP __hardpps function
2450 void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts)
2452 unsigned long flags;
2454 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2455 write_seqcount_begin(&tk_core.seq);
2457 __hardpps(phase_ts, raw_ts);
2459 write_seqcount_end(&tk_core.seq);
2460 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2462 EXPORT_SYMBOL(hardpps);
2463 #endif /* CONFIG_NTP_PPS */
2466 * xtime_update() - advances the timekeeping infrastructure
2467 * @ticks: number of ticks, that have elapsed since the last call.
2469 * Must be called with interrupts disabled.
2471 void xtime_update(unsigned long ticks)
2473 raw_spin_lock(&jiffies_lock);
2474 write_seqcount_begin(&jiffies_seq);
2476 write_seqcount_end(&jiffies_seq);
2477 raw_spin_unlock(&jiffies_lock);