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;
58 * struct tk_fast - NMI safe timekeeper
59 * @seq: Sequence counter for protecting updates. The lowest bit
60 * is the index for the tk_read_base array
61 * @base: tk_read_base array. Access is indexed by the lowest bit of
64 * See @update_fast_timekeeper() below.
67 seqcount_raw_spinlock_t seq;
68 struct tk_read_base base[2];
71 /* Suspend-time cycles value for halted fast timekeeper. */
72 static u64 cycles_at_suspend;
74 static u64 dummy_clock_read(struct clocksource *cs)
76 return cycles_at_suspend;
79 static struct clocksource dummy_clock = {
80 .read = dummy_clock_read,
83 static struct tk_fast tk_fast_mono ____cacheline_aligned = {
84 .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_fast_mono.seq, &timekeeper_lock),
85 .base[0] = { .clock = &dummy_clock, },
86 .base[1] = { .clock = &dummy_clock, },
89 static struct tk_fast tk_fast_raw ____cacheline_aligned = {
90 .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_fast_raw.seq, &timekeeper_lock),
91 .base[0] = { .clock = &dummy_clock, },
92 .base[1] = { .clock = &dummy_clock, },
95 /* flag for if timekeeping is suspended */
96 int __read_mostly timekeeping_suspended;
98 static inline void tk_normalize_xtime(struct timekeeper *tk)
100 while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) {
101 tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
104 while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) {
105 tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
110 static inline struct timespec64 tk_xtime(const struct timekeeper *tk)
112 struct timespec64 ts;
114 ts.tv_sec = tk->xtime_sec;
115 ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
119 static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
121 tk->xtime_sec = ts->tv_sec;
122 tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift;
125 static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
127 tk->xtime_sec += ts->tv_sec;
128 tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift;
129 tk_normalize_xtime(tk);
132 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
134 struct timespec64 tmp;
137 * Verify consistency of: offset_real = -wall_to_monotonic
138 * before modifying anything
140 set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
141 -tk->wall_to_monotonic.tv_nsec);
142 WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp));
143 tk->wall_to_monotonic = wtm;
144 set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
145 tk->offs_real = timespec64_to_ktime(tmp);
146 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
149 static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
151 tk->offs_boot = ktime_add(tk->offs_boot, delta);
153 * Timespec representation for VDSO update to avoid 64bit division
156 tk->monotonic_to_boot = ktime_to_timespec64(tk->offs_boot);
160 * tk_clock_read - atomic clocksource read() helper
162 * This helper is necessary to use in the read paths because, while the
163 * seqcount ensures we don't return a bad value while structures are updated,
164 * it doesn't protect from potential crashes. There is the possibility that
165 * the tkr's clocksource may change between the read reference, and the
166 * clock reference passed to the read function. This can cause crashes if
167 * the wrong clocksource is passed to the wrong read function.
168 * This isn't necessary to use when holding the timekeeper_lock or doing
169 * a read of the fast-timekeeper tkrs (which is protected by its own locking
172 static inline u64 tk_clock_read(const struct tk_read_base *tkr)
174 struct clocksource *clock = READ_ONCE(tkr->clock);
176 return clock->read(clock);
179 #ifdef CONFIG_DEBUG_TIMEKEEPING
180 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
182 static void timekeeping_check_update(struct timekeeper *tk, u64 offset)
185 u64 max_cycles = tk->tkr_mono.clock->max_cycles;
186 const char *name = tk->tkr_mono.clock->name;
188 if (offset > max_cycles) {
189 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
190 offset, name, max_cycles);
191 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
193 if (offset > (max_cycles >> 1)) {
194 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
195 offset, name, max_cycles >> 1);
196 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
200 if (tk->underflow_seen) {
201 if (jiffies - tk->last_warning > WARNING_FREQ) {
202 printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name);
203 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
204 printk_deferred(" Your kernel is probably still fine.\n");
205 tk->last_warning = jiffies;
207 tk->underflow_seen = 0;
210 if (tk->overflow_seen) {
211 if (jiffies - tk->last_warning > WARNING_FREQ) {
212 printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name);
213 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
214 printk_deferred(" Your kernel is probably still fine.\n");
215 tk->last_warning = jiffies;
217 tk->overflow_seen = 0;
221 static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
223 struct timekeeper *tk = &tk_core.timekeeper;
224 u64 now, last, mask, max, delta;
228 * Since we're called holding a seqcount, the data may shift
229 * under us while we're doing the calculation. This can cause
230 * false positives, since we'd note a problem but throw the
231 * results away. So nest another seqcount here to atomically
232 * grab the points we are checking with.
235 seq = read_seqcount_begin(&tk_core.seq);
236 now = tk_clock_read(tkr);
237 last = tkr->cycle_last;
239 max = tkr->clock->max_cycles;
240 } while (read_seqcount_retry(&tk_core.seq, seq));
242 delta = clocksource_delta(now, last, mask);
245 * Try to catch underflows by checking if we are seeing small
246 * mask-relative negative values.
248 if (unlikely((~delta & mask) < (mask >> 3))) {
249 tk->underflow_seen = 1;
253 /* Cap delta value to the max_cycles values to avoid mult overflows */
254 if (unlikely(delta > max)) {
255 tk->overflow_seen = 1;
256 delta = tkr->clock->max_cycles;
262 static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset)
265 static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
267 u64 cycle_now, delta;
269 /* read clocksource */
270 cycle_now = tk_clock_read(tkr);
272 /* calculate the delta since the last update_wall_time */
273 delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
280 * tk_setup_internals - Set up internals to use clocksource clock.
282 * @tk: The target timekeeper to setup.
283 * @clock: Pointer to clocksource.
285 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
286 * pair and interval request.
288 * Unless you're the timekeeping code, you should not be using this!
290 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
293 u64 tmp, ntpinterval;
294 struct clocksource *old_clock;
296 ++tk->cs_was_changed_seq;
297 old_clock = tk->tkr_mono.clock;
298 tk->tkr_mono.clock = clock;
299 tk->tkr_mono.mask = clock->mask;
300 tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono);
302 tk->tkr_raw.clock = clock;
303 tk->tkr_raw.mask = clock->mask;
304 tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last;
306 /* Do the ns -> cycle conversion first, using original mult */
307 tmp = NTP_INTERVAL_LENGTH;
308 tmp <<= clock->shift;
310 tmp += clock->mult/2;
311 do_div(tmp, clock->mult);
315 interval = (u64) tmp;
316 tk->cycle_interval = interval;
318 /* Go back from cycles -> shifted ns */
319 tk->xtime_interval = interval * clock->mult;
320 tk->xtime_remainder = ntpinterval - tk->xtime_interval;
321 tk->raw_interval = interval * clock->mult;
323 /* if changing clocks, convert xtime_nsec shift units */
325 int shift_change = clock->shift - old_clock->shift;
326 if (shift_change < 0) {
327 tk->tkr_mono.xtime_nsec >>= -shift_change;
328 tk->tkr_raw.xtime_nsec >>= -shift_change;
330 tk->tkr_mono.xtime_nsec <<= shift_change;
331 tk->tkr_raw.xtime_nsec <<= shift_change;
335 tk->tkr_mono.shift = clock->shift;
336 tk->tkr_raw.shift = clock->shift;
339 tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
340 tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
343 * The timekeeper keeps its own mult values for the currently
344 * active clocksource. These value will be adjusted via NTP
345 * to counteract clock drifting.
347 tk->tkr_mono.mult = clock->mult;
348 tk->tkr_raw.mult = clock->mult;
349 tk->ntp_err_mult = 0;
350 tk->skip_second_overflow = 0;
353 /* Timekeeper helper functions. */
355 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
356 static u32 default_arch_gettimeoffset(void) { return 0; }
357 u32 (*arch_gettimeoffset)(void) = default_arch_gettimeoffset;
359 static inline u32 arch_gettimeoffset(void) { return 0; }
362 static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta)
366 nsec = delta * tkr->mult + tkr->xtime_nsec;
369 /* If arch requires, add in get_arch_timeoffset() */
370 return nsec + arch_gettimeoffset();
373 static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr)
377 delta = timekeeping_get_delta(tkr);
378 return timekeeping_delta_to_ns(tkr, delta);
381 static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles)
385 /* calculate the delta since the last update_wall_time */
386 delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask);
387 return timekeeping_delta_to_ns(tkr, delta);
391 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
392 * @tkr: Timekeeping readout base from which we take the update
394 * We want to use this from any context including NMI and tracing /
395 * instrumenting the timekeeping code itself.
397 * Employ the latch technique; see @raw_write_seqcount_latch.
399 * So if a NMI hits the update of base[0] then it will use base[1]
400 * which is still consistent. In the worst case this can result is a
401 * slightly wrong timestamp (a few nanoseconds). See
402 * @ktime_get_mono_fast_ns.
404 static void update_fast_timekeeper(const struct tk_read_base *tkr,
407 struct tk_read_base *base = tkf->base;
409 /* Force readers off to base[1] */
410 raw_write_seqcount_latch(&tkf->seq);
413 memcpy(base, tkr, sizeof(*base));
415 /* Force readers back to base[0] */
416 raw_write_seqcount_latch(&tkf->seq);
419 memcpy(base + 1, base, sizeof(*base));
423 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
425 * This timestamp is not guaranteed to be monotonic across an update.
426 * The timestamp is calculated by:
428 * now = base_mono + clock_delta * slope
430 * So if the update lowers the slope, readers who are forced to the
431 * not yet updated second array are still using the old steeper slope.
440 * |12345678---> reader order
446 * So reader 6 will observe time going backwards versus reader 5.
448 * While other CPUs are likely to be able observe that, the only way
449 * for a CPU local observation is when an NMI hits in the middle of
450 * the update. Timestamps taken from that NMI context might be ahead
451 * of the following timestamps. Callers need to be aware of that and
454 static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf)
456 struct tk_read_base *tkr;
461 seq = raw_read_seqcount_latch(&tkf->seq);
462 tkr = tkf->base + (seq & 0x01);
463 now = ktime_to_ns(tkr->base);
465 now += timekeeping_delta_to_ns(tkr,
470 } while (read_seqcount_retry(&tkf->seq, seq));
475 u64 ktime_get_mono_fast_ns(void)
477 return __ktime_get_fast_ns(&tk_fast_mono);
479 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
481 u64 ktime_get_raw_fast_ns(void)
483 return __ktime_get_fast_ns(&tk_fast_raw);
485 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
488 * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
490 * To keep it NMI safe since we're accessing from tracing, we're not using a
491 * separate timekeeper with updates to monotonic clock and boot offset
492 * protected with seqcounts. This has the following minor side effects:
494 * (1) Its possible that a timestamp be taken after the boot offset is updated
495 * but before the timekeeper is updated. If this happens, the new boot offset
496 * is added to the old timekeeping making the clock appear to update slightly
499 * timekeeping_inject_sleeptime64()
500 * __timekeeping_inject_sleeptime(tk, delta);
502 * timekeeping_update(tk, TK_CLEAR_NTP...);
504 * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
505 * partially updated. Since the tk->offs_boot update is a rare event, this
506 * should be a rare occurrence which postprocessing should be able to handle.
508 u64 notrace ktime_get_boot_fast_ns(void)
510 struct timekeeper *tk = &tk_core.timekeeper;
512 return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
514 EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
518 * See comment for __ktime_get_fast_ns() vs. timestamp ordering
520 static __always_inline u64 __ktime_get_real_fast_ns(struct tk_fast *tkf)
522 struct tk_read_base *tkr;
527 seq = raw_read_seqcount_latch(&tkf->seq);
528 tkr = tkf->base + (seq & 0x01);
529 now = ktime_to_ns(tkr->base_real);
531 now += timekeeping_delta_to_ns(tkr,
536 } while (read_seqcount_retry(&tkf->seq, seq));
542 * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
544 u64 ktime_get_real_fast_ns(void)
546 return __ktime_get_real_fast_ns(&tk_fast_mono);
548 EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns);
551 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
552 * @tk: Timekeeper to snapshot.
554 * It generally is unsafe to access the clocksource after timekeeping has been
555 * suspended, so take a snapshot of the readout base of @tk and use it as the
556 * fast timekeeper's readout base while suspended. It will return the same
557 * number of cycles every time until timekeeping is resumed at which time the
558 * proper readout base for the fast timekeeper will be restored automatically.
560 static void halt_fast_timekeeper(const struct timekeeper *tk)
562 static struct tk_read_base tkr_dummy;
563 const struct tk_read_base *tkr = &tk->tkr_mono;
565 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
566 cycles_at_suspend = tk_clock_read(tkr);
567 tkr_dummy.clock = &dummy_clock;
568 tkr_dummy.base_real = tkr->base + tk->offs_real;
569 update_fast_timekeeper(&tkr_dummy, &tk_fast_mono);
572 memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
573 tkr_dummy.clock = &dummy_clock;
574 update_fast_timekeeper(&tkr_dummy, &tk_fast_raw);
577 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
579 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
581 raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
585 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
587 int pvclock_gtod_register_notifier(struct notifier_block *nb)
589 struct timekeeper *tk = &tk_core.timekeeper;
593 raw_spin_lock_irqsave(&timekeeper_lock, flags);
594 ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
595 update_pvclock_gtod(tk, true);
596 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
600 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
603 * pvclock_gtod_unregister_notifier - unregister a pvclock
604 * timedata update listener
606 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
611 raw_spin_lock_irqsave(&timekeeper_lock, flags);
612 ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
613 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
617 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
620 * tk_update_leap_state - helper to update the next_leap_ktime
622 static inline void tk_update_leap_state(struct timekeeper *tk)
624 tk->next_leap_ktime = ntp_get_next_leap();
625 if (tk->next_leap_ktime != KTIME_MAX)
626 /* Convert to monotonic time */
627 tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real);
631 * Update the ktime_t based scalar nsec members of the timekeeper
633 static inline void tk_update_ktime_data(struct timekeeper *tk)
639 * The xtime based monotonic readout is:
640 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
641 * The ktime based monotonic readout is:
642 * nsec = base_mono + now();
643 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
645 seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
646 nsec = (u32) tk->wall_to_monotonic.tv_nsec;
647 tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
650 * The sum of the nanoseconds portions of xtime and
651 * wall_to_monotonic can be greater/equal one second. Take
652 * this into account before updating tk->ktime_sec.
654 nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
655 if (nsec >= NSEC_PER_SEC)
657 tk->ktime_sec = seconds;
659 /* Update the monotonic raw base */
660 tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC);
663 /* must hold timekeeper_lock */
664 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
666 if (action & TK_CLEAR_NTP) {
671 tk_update_leap_state(tk);
672 tk_update_ktime_data(tk);
675 update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
677 tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real;
678 update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
679 update_fast_timekeeper(&tk->tkr_raw, &tk_fast_raw);
681 if (action & TK_CLOCK_WAS_SET)
682 tk->clock_was_set_seq++;
684 * The mirroring of the data to the shadow-timekeeper needs
685 * to happen last here to ensure we don't over-write the
686 * timekeeper structure on the next update with stale data
688 if (action & TK_MIRROR)
689 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
690 sizeof(tk_core.timekeeper));
694 * timekeeping_forward_now - update clock to the current time
696 * Forward the current clock to update its state since the last call to
697 * update_wall_time(). This is useful before significant clock changes,
698 * as it avoids having to deal with this time offset explicitly.
700 static void timekeeping_forward_now(struct timekeeper *tk)
702 u64 cycle_now, delta;
704 cycle_now = tk_clock_read(&tk->tkr_mono);
705 delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
706 tk->tkr_mono.cycle_last = cycle_now;
707 tk->tkr_raw.cycle_last = cycle_now;
709 tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult;
711 /* If arch requires, add in get_arch_timeoffset() */
712 tk->tkr_mono.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_mono.shift;
715 tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult;
717 /* If arch requires, add in get_arch_timeoffset() */
718 tk->tkr_raw.xtime_nsec += (u64)arch_gettimeoffset() << tk->tkr_raw.shift;
720 tk_normalize_xtime(tk);
724 * ktime_get_real_ts64 - Returns the time of day in a timespec64.
725 * @ts: pointer to the timespec to be set
727 * Returns the time of day in a timespec64 (WARN if suspended).
729 void ktime_get_real_ts64(struct timespec64 *ts)
731 struct timekeeper *tk = &tk_core.timekeeper;
735 WARN_ON(timekeeping_suspended);
738 seq = read_seqcount_begin(&tk_core.seq);
740 ts->tv_sec = tk->xtime_sec;
741 nsecs = timekeeping_get_ns(&tk->tkr_mono);
743 } while (read_seqcount_retry(&tk_core.seq, seq));
746 timespec64_add_ns(ts, nsecs);
748 EXPORT_SYMBOL(ktime_get_real_ts64);
750 ktime_t ktime_get(void)
752 struct timekeeper *tk = &tk_core.timekeeper;
757 WARN_ON(timekeeping_suspended);
760 seq = read_seqcount_begin(&tk_core.seq);
761 base = tk->tkr_mono.base;
762 nsecs = timekeeping_get_ns(&tk->tkr_mono);
764 } while (read_seqcount_retry(&tk_core.seq, seq));
766 return ktime_add_ns(base, nsecs);
768 EXPORT_SYMBOL_GPL(ktime_get);
770 u32 ktime_get_resolution_ns(void)
772 struct timekeeper *tk = &tk_core.timekeeper;
776 WARN_ON(timekeeping_suspended);
779 seq = read_seqcount_begin(&tk_core.seq);
780 nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift;
781 } while (read_seqcount_retry(&tk_core.seq, seq));
785 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns);
787 static ktime_t *offsets[TK_OFFS_MAX] = {
788 [TK_OFFS_REAL] = &tk_core.timekeeper.offs_real,
789 [TK_OFFS_BOOT] = &tk_core.timekeeper.offs_boot,
790 [TK_OFFS_TAI] = &tk_core.timekeeper.offs_tai,
793 ktime_t ktime_get_with_offset(enum tk_offsets offs)
795 struct timekeeper *tk = &tk_core.timekeeper;
797 ktime_t base, *offset = offsets[offs];
800 WARN_ON(timekeeping_suspended);
803 seq = read_seqcount_begin(&tk_core.seq);
804 base = ktime_add(tk->tkr_mono.base, *offset);
805 nsecs = timekeeping_get_ns(&tk->tkr_mono);
807 } while (read_seqcount_retry(&tk_core.seq, seq));
809 return ktime_add_ns(base, nsecs);
812 EXPORT_SYMBOL_GPL(ktime_get_with_offset);
814 ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs)
816 struct timekeeper *tk = &tk_core.timekeeper;
818 ktime_t base, *offset = offsets[offs];
821 WARN_ON(timekeeping_suspended);
824 seq = read_seqcount_begin(&tk_core.seq);
825 base = ktime_add(tk->tkr_mono.base, *offset);
826 nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift;
828 } while (read_seqcount_retry(&tk_core.seq, seq));
830 return ktime_add_ns(base, nsecs);
832 EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset);
835 * ktime_mono_to_any() - convert mononotic time to any other time
836 * @tmono: time to convert.
837 * @offs: which offset to use
839 ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
841 ktime_t *offset = offsets[offs];
846 seq = read_seqcount_begin(&tk_core.seq);
847 tconv = ktime_add(tmono, *offset);
848 } while (read_seqcount_retry(&tk_core.seq, seq));
852 EXPORT_SYMBOL_GPL(ktime_mono_to_any);
855 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
857 ktime_t ktime_get_raw(void)
859 struct timekeeper *tk = &tk_core.timekeeper;
865 seq = read_seqcount_begin(&tk_core.seq);
866 base = tk->tkr_raw.base;
867 nsecs = timekeeping_get_ns(&tk->tkr_raw);
869 } while (read_seqcount_retry(&tk_core.seq, seq));
871 return ktime_add_ns(base, nsecs);
873 EXPORT_SYMBOL_GPL(ktime_get_raw);
876 * ktime_get_ts64 - get the monotonic clock in timespec64 format
877 * @ts: pointer to timespec variable
879 * The function calculates the monotonic clock from the realtime
880 * clock and the wall_to_monotonic offset and stores the result
881 * in normalized timespec64 format in the variable pointed to by @ts.
883 void ktime_get_ts64(struct timespec64 *ts)
885 struct timekeeper *tk = &tk_core.timekeeper;
886 struct timespec64 tomono;
890 WARN_ON(timekeeping_suspended);
893 seq = read_seqcount_begin(&tk_core.seq);
894 ts->tv_sec = tk->xtime_sec;
895 nsec = timekeeping_get_ns(&tk->tkr_mono);
896 tomono = tk->wall_to_monotonic;
898 } while (read_seqcount_retry(&tk_core.seq, seq));
900 ts->tv_sec += tomono.tv_sec;
902 timespec64_add_ns(ts, nsec + tomono.tv_nsec);
904 EXPORT_SYMBOL_GPL(ktime_get_ts64);
907 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
909 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
910 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
911 * works on both 32 and 64 bit systems. On 32 bit systems the readout
912 * covers ~136 years of uptime which should be enough to prevent
913 * premature wrap arounds.
915 time64_t ktime_get_seconds(void)
917 struct timekeeper *tk = &tk_core.timekeeper;
919 WARN_ON(timekeeping_suspended);
920 return tk->ktime_sec;
922 EXPORT_SYMBOL_GPL(ktime_get_seconds);
925 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
927 * Returns the wall clock seconds since 1970. This replaces the
928 * get_seconds() interface which is not y2038 safe on 32bit systems.
930 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
931 * 32bit systems the access must be protected with the sequence
932 * counter to provide "atomic" access to the 64bit tk->xtime_sec
935 time64_t ktime_get_real_seconds(void)
937 struct timekeeper *tk = &tk_core.timekeeper;
941 if (IS_ENABLED(CONFIG_64BIT))
942 return tk->xtime_sec;
945 seq = read_seqcount_begin(&tk_core.seq);
946 seconds = tk->xtime_sec;
948 } while (read_seqcount_retry(&tk_core.seq, seq));
952 EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
955 * __ktime_get_real_seconds - The same as ktime_get_real_seconds
956 * but without the sequence counter protect. This internal function
957 * is called just when timekeeping lock is already held.
959 noinstr time64_t __ktime_get_real_seconds(void)
961 struct timekeeper *tk = &tk_core.timekeeper;
963 return tk->xtime_sec;
967 * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
968 * @systime_snapshot: pointer to struct receiving the system time snapshot
970 void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot)
972 struct timekeeper *tk = &tk_core.timekeeper;
980 WARN_ON_ONCE(timekeeping_suspended);
983 seq = read_seqcount_begin(&tk_core.seq);
984 now = tk_clock_read(&tk->tkr_mono);
985 systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq;
986 systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq;
987 base_real = ktime_add(tk->tkr_mono.base,
988 tk_core.timekeeper.offs_real);
989 base_raw = tk->tkr_raw.base;
990 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now);
991 nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw, now);
992 } while (read_seqcount_retry(&tk_core.seq, seq));
994 systime_snapshot->cycles = now;
995 systime_snapshot->real = ktime_add_ns(base_real, nsec_real);
996 systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw);
998 EXPORT_SYMBOL_GPL(ktime_get_snapshot);
1000 /* Scale base by mult/div checking for overflow */
1001 static int scale64_check_overflow(u64 mult, u64 div, u64 *base)
1005 tmp = div64_u64_rem(*base, div, &rem);
1007 if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) ||
1008 ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem)))
1012 rem = div64_u64(rem * mult, div);
1018 * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1019 * @history: Snapshot representing start of history
1020 * @partial_history_cycles: Cycle offset into history (fractional part)
1021 * @total_history_cycles: Total history length in cycles
1022 * @discontinuity: True indicates clock was set on history period
1023 * @ts: Cross timestamp that should be adjusted using
1024 * partial/total ratio
1026 * Helper function used by get_device_system_crosststamp() to correct the
1027 * crosstimestamp corresponding to the start of the current interval to the
1028 * system counter value (timestamp point) provided by the driver. The
1029 * total_history_* quantities are the total history starting at the provided
1030 * reference point and ending at the start of the current interval. The cycle
1031 * count between the driver timestamp point and the start of the current
1032 * interval is partial_history_cycles.
1034 static int adjust_historical_crosststamp(struct system_time_snapshot *history,
1035 u64 partial_history_cycles,
1036 u64 total_history_cycles,
1038 struct system_device_crosststamp *ts)
1040 struct timekeeper *tk = &tk_core.timekeeper;
1041 u64 corr_raw, corr_real;
1042 bool interp_forward;
1045 if (total_history_cycles == 0 || partial_history_cycles == 0)
1048 /* Interpolate shortest distance from beginning or end of history */
1049 interp_forward = partial_history_cycles > total_history_cycles / 2;
1050 partial_history_cycles = interp_forward ?
1051 total_history_cycles - partial_history_cycles :
1052 partial_history_cycles;
1055 * Scale the monotonic raw time delta by:
1056 * partial_history_cycles / total_history_cycles
1058 corr_raw = (u64)ktime_to_ns(
1059 ktime_sub(ts->sys_monoraw, history->raw));
1060 ret = scale64_check_overflow(partial_history_cycles,
1061 total_history_cycles, &corr_raw);
1066 * If there is a discontinuity in the history, scale monotonic raw
1068 * mult(real)/mult(raw) yielding the realtime correction
1069 * Otherwise, calculate the realtime correction similar to monotonic
1072 if (discontinuity) {
1073 corr_real = mul_u64_u32_div
1074 (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult);
1076 corr_real = (u64)ktime_to_ns(
1077 ktime_sub(ts->sys_realtime, history->real));
1078 ret = scale64_check_overflow(partial_history_cycles,
1079 total_history_cycles, &corr_real);
1084 /* Fixup monotonic raw and real time time values */
1085 if (interp_forward) {
1086 ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw);
1087 ts->sys_realtime = ktime_add_ns(history->real, corr_real);
1089 ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw);
1090 ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real);
1097 * cycle_between - true if test occurs chronologically between before and after
1099 static bool cycle_between(u64 before, u64 test, u64 after)
1101 if (test > before && test < after)
1103 if (test < before && before > after)
1109 * get_device_system_crosststamp - Synchronously capture system/device timestamp
1110 * @get_time_fn: Callback to get simultaneous device time and
1111 * system counter from the device driver
1112 * @ctx: Context passed to get_time_fn()
1113 * @history_begin: Historical reference point used to interpolate system
1114 * time when counter provided by the driver is before the current interval
1115 * @xtstamp: Receives simultaneously captured system and device time
1117 * Reads a timestamp from a device and correlates it to system time
1119 int get_device_system_crosststamp(int (*get_time_fn)
1120 (ktime_t *device_time,
1121 struct system_counterval_t *sys_counterval,
1124 struct system_time_snapshot *history_begin,
1125 struct system_device_crosststamp *xtstamp)
1127 struct system_counterval_t system_counterval;
1128 struct timekeeper *tk = &tk_core.timekeeper;
1129 u64 cycles, now, interval_start;
1130 unsigned int clock_was_set_seq = 0;
1131 ktime_t base_real, base_raw;
1132 u64 nsec_real, nsec_raw;
1133 u8 cs_was_changed_seq;
1139 seq = read_seqcount_begin(&tk_core.seq);
1141 * Try to synchronously capture device time and a system
1142 * counter value calling back into the device driver
1144 ret = get_time_fn(&xtstamp->device, &system_counterval, ctx);
1149 * Verify that the clocksource associated with the captured
1150 * system counter value is the same as the currently installed
1151 * timekeeper clocksource
1153 if (tk->tkr_mono.clock != system_counterval.cs)
1155 cycles = system_counterval.cycles;
1158 * Check whether the system counter value provided by the
1159 * device driver is on the current timekeeping interval.
1161 now = tk_clock_read(&tk->tkr_mono);
1162 interval_start = tk->tkr_mono.cycle_last;
1163 if (!cycle_between(interval_start, cycles, now)) {
1164 clock_was_set_seq = tk->clock_was_set_seq;
1165 cs_was_changed_seq = tk->cs_was_changed_seq;
1166 cycles = interval_start;
1172 base_real = ktime_add(tk->tkr_mono.base,
1173 tk_core.timekeeper.offs_real);
1174 base_raw = tk->tkr_raw.base;
1176 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono,
1177 system_counterval.cycles);
1178 nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw,
1179 system_counterval.cycles);
1180 } while (read_seqcount_retry(&tk_core.seq, seq));
1182 xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
1183 xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw);
1186 * Interpolate if necessary, adjusting back from the start of the
1190 u64 partial_history_cycles, total_history_cycles;
1194 * Check that the counter value occurs after the provided
1195 * history reference and that the history doesn't cross a
1196 * clocksource change
1198 if (!history_begin ||
1199 !cycle_between(history_begin->cycles,
1200 system_counterval.cycles, cycles) ||
1201 history_begin->cs_was_changed_seq != cs_was_changed_seq)
1203 partial_history_cycles = cycles - system_counterval.cycles;
1204 total_history_cycles = cycles - history_begin->cycles;
1206 history_begin->clock_was_set_seq != clock_was_set_seq;
1208 ret = adjust_historical_crosststamp(history_begin,
1209 partial_history_cycles,
1210 total_history_cycles,
1211 discontinuity, xtstamp);
1218 EXPORT_SYMBOL_GPL(get_device_system_crosststamp);
1221 * do_settimeofday64 - Sets the time of day.
1222 * @ts: pointer to the timespec64 variable containing the new time
1224 * Sets the time of day to the new time and update NTP and notify hrtimers
1226 int do_settimeofday64(const struct timespec64 *ts)
1228 struct timekeeper *tk = &tk_core.timekeeper;
1229 struct timespec64 ts_delta, xt;
1230 unsigned long flags;
1233 if (!timespec64_valid_settod(ts))
1236 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1237 write_seqcount_begin(&tk_core.seq);
1239 timekeeping_forward_now(tk);
1242 ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
1243 ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
1245 if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) {
1250 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
1252 tk_set_xtime(tk, ts);
1254 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1256 write_seqcount_end(&tk_core.seq);
1257 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1259 /* signal hrtimers about time change */
1263 audit_tk_injoffset(ts_delta);
1267 EXPORT_SYMBOL(do_settimeofday64);
1270 * timekeeping_inject_offset - Adds or subtracts from the current time.
1271 * @tv: pointer to the timespec variable containing the offset
1273 * Adds or subtracts an offset value from the current time.
1275 static int timekeeping_inject_offset(const struct timespec64 *ts)
1277 struct timekeeper *tk = &tk_core.timekeeper;
1278 unsigned long flags;
1279 struct timespec64 tmp;
1282 if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
1285 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1286 write_seqcount_begin(&tk_core.seq);
1288 timekeeping_forward_now(tk);
1290 /* Make sure the proposed value is valid */
1291 tmp = timespec64_add(tk_xtime(tk), *ts);
1292 if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 ||
1293 !timespec64_valid_settod(&tmp)) {
1298 tk_xtime_add(tk, ts);
1299 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts));
1301 error: /* even if we error out, we forwarded the time, so call update */
1302 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1304 write_seqcount_end(&tk_core.seq);
1305 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1307 /* signal hrtimers about time change */
1314 * Indicates if there is an offset between the system clock and the hardware
1315 * clock/persistent clock/rtc.
1317 int persistent_clock_is_local;
1320 * Adjust the time obtained from the CMOS to be UTC time instead of
1323 * This is ugly, but preferable to the alternatives. Otherwise we
1324 * would either need to write a program to do it in /etc/rc (and risk
1325 * confusion if the program gets run more than once; it would also be
1326 * hard to make the program warp the clock precisely n hours) or
1327 * compile in the timezone information into the kernel. Bad, bad....
1331 * The best thing to do is to keep the CMOS clock in universal time (UTC)
1332 * as real UNIX machines always do it. This avoids all headaches about
1333 * daylight saving times and warping kernel clocks.
1335 void timekeeping_warp_clock(void)
1337 if (sys_tz.tz_minuteswest != 0) {
1338 struct timespec64 adjust;
1340 persistent_clock_is_local = 1;
1341 adjust.tv_sec = sys_tz.tz_minuteswest * 60;
1343 timekeeping_inject_offset(&adjust);
1348 * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1351 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
1353 tk->tai_offset = tai_offset;
1354 tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
1358 * change_clocksource - Swaps clocksources if a new one is available
1360 * Accumulates current time interval and initializes new clocksource
1362 static int change_clocksource(void *data)
1364 struct timekeeper *tk = &tk_core.timekeeper;
1365 struct clocksource *new, *old;
1366 unsigned long flags;
1368 new = (struct clocksource *) data;
1370 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1371 write_seqcount_begin(&tk_core.seq);
1373 timekeeping_forward_now(tk);
1375 * If the cs is in module, get a module reference. Succeeds
1376 * for built-in code (owner == NULL) as well.
1378 if (try_module_get(new->owner)) {
1379 if (!new->enable || new->enable(new) == 0) {
1380 old = tk->tkr_mono.clock;
1381 tk_setup_internals(tk, new);
1384 module_put(old->owner);
1386 module_put(new->owner);
1389 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1391 write_seqcount_end(&tk_core.seq);
1392 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1398 * timekeeping_notify - Install a new clock source
1399 * @clock: pointer to the clock source
1401 * This function is called from clocksource.c after a new, better clock
1402 * source has been registered. The caller holds the clocksource_mutex.
1404 int timekeeping_notify(struct clocksource *clock)
1406 struct timekeeper *tk = &tk_core.timekeeper;
1408 if (tk->tkr_mono.clock == clock)
1410 stop_machine(change_clocksource, clock, NULL);
1411 tick_clock_notify();
1412 return tk->tkr_mono.clock == clock ? 0 : -1;
1416 * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1417 * @ts: pointer to the timespec64 to be set
1419 * Returns the raw monotonic time (completely un-modified by ntp)
1421 void ktime_get_raw_ts64(struct timespec64 *ts)
1423 struct timekeeper *tk = &tk_core.timekeeper;
1428 seq = read_seqcount_begin(&tk_core.seq);
1429 ts->tv_sec = tk->raw_sec;
1430 nsecs = timekeeping_get_ns(&tk->tkr_raw);
1432 } while (read_seqcount_retry(&tk_core.seq, seq));
1435 timespec64_add_ns(ts, nsecs);
1437 EXPORT_SYMBOL(ktime_get_raw_ts64);
1441 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1443 int timekeeping_valid_for_hres(void)
1445 struct timekeeper *tk = &tk_core.timekeeper;
1450 seq = read_seqcount_begin(&tk_core.seq);
1452 ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
1454 } while (read_seqcount_retry(&tk_core.seq, seq));
1460 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1462 u64 timekeeping_max_deferment(void)
1464 struct timekeeper *tk = &tk_core.timekeeper;
1469 seq = read_seqcount_begin(&tk_core.seq);
1471 ret = tk->tkr_mono.clock->max_idle_ns;
1473 } while (read_seqcount_retry(&tk_core.seq, seq));
1479 * read_persistent_clock64 - Return time from the persistent clock.
1481 * Weak dummy function for arches that do not yet support it.
1482 * Reads the time from the battery backed persistent clock.
1483 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1485 * XXX - Do be sure to remove it once all arches implement it.
1487 void __weak read_persistent_clock64(struct timespec64 *ts)
1494 * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1497 * Weak dummy function for arches that do not yet support it.
1498 * wall_time - current time as returned by persistent clock
1499 * boot_offset - offset that is defined as wall_time - boot_time
1500 * The default function calculates offset based on the current value of
1501 * local_clock(). This way architectures that support sched_clock() but don't
1502 * support dedicated boot time clock will provide the best estimate of the
1506 read_persistent_wall_and_boot_offset(struct timespec64 *wall_time,
1507 struct timespec64 *boot_offset)
1509 read_persistent_clock64(wall_time);
1510 *boot_offset = ns_to_timespec64(local_clock());
1514 * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1516 * The flag starts of false and is only set when a suspend reaches
1517 * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1518 * timekeeper clocksource is not stopping across suspend and has been
1519 * used to update sleep time. If the timekeeper clocksource has stopped
1520 * then the flag stays true and is used by the RTC resume code to decide
1521 * whether sleeptime must be injected and if so the flag gets false then.
1523 * If a suspend fails before reaching timekeeping_resume() then the flag
1524 * stays false and prevents erroneous sleeptime injection.
1526 static bool suspend_timing_needed;
1528 /* Flag for if there is a persistent clock on this platform */
1529 static bool persistent_clock_exists;
1532 * timekeeping_init - Initializes the clocksource and common timekeeping values
1534 void __init timekeeping_init(void)
1536 struct timespec64 wall_time, boot_offset, wall_to_mono;
1537 struct timekeeper *tk = &tk_core.timekeeper;
1538 struct clocksource *clock;
1539 unsigned long flags;
1541 read_persistent_wall_and_boot_offset(&wall_time, &boot_offset);
1542 if (timespec64_valid_settod(&wall_time) &&
1543 timespec64_to_ns(&wall_time) > 0) {
1544 persistent_clock_exists = true;
1545 } else if (timespec64_to_ns(&wall_time) != 0) {
1546 pr_warn("Persistent clock returned invalid value");
1547 wall_time = (struct timespec64){0};
1550 if (timespec64_compare(&wall_time, &boot_offset) < 0)
1551 boot_offset = (struct timespec64){0};
1554 * We want set wall_to_mono, so the following is true:
1555 * wall time + wall_to_mono = boot time
1557 wall_to_mono = timespec64_sub(boot_offset, wall_time);
1559 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1560 write_seqcount_begin(&tk_core.seq);
1563 clock = clocksource_default_clock();
1565 clock->enable(clock);
1566 tk_setup_internals(tk, clock);
1568 tk_set_xtime(tk, &wall_time);
1571 tk_set_wall_to_mono(tk, wall_to_mono);
1573 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1575 write_seqcount_end(&tk_core.seq);
1576 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1579 /* time in seconds when suspend began for persistent clock */
1580 static struct timespec64 timekeeping_suspend_time;
1583 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1584 * @delta: pointer to a timespec delta value
1586 * Takes a timespec offset measuring a suspend interval and properly
1587 * adds the sleep offset to the timekeeping variables.
1589 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
1590 const struct timespec64 *delta)
1592 if (!timespec64_valid_strict(delta)) {
1593 printk_deferred(KERN_WARNING
1594 "__timekeeping_inject_sleeptime: Invalid "
1595 "sleep delta value!\n");
1598 tk_xtime_add(tk, delta);
1599 tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
1600 tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
1601 tk_debug_account_sleep_time(delta);
1604 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1606 * We have three kinds of time sources to use for sleep time
1607 * injection, the preference order is:
1608 * 1) non-stop clocksource
1609 * 2) persistent clock (ie: RTC accessible when irqs are off)
1612 * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1613 * If system has neither 1) nor 2), 3) will be used finally.
1616 * If timekeeping has injected sleeptime via either 1) or 2),
1617 * 3) becomes needless, so in this case we don't need to call
1618 * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1621 bool timekeeping_rtc_skipresume(void)
1623 return !suspend_timing_needed;
1627 * 1) can be determined whether to use or not only when doing
1628 * timekeeping_resume() which is invoked after rtc_suspend(),
1629 * so we can't skip rtc_suspend() surely if system has 1).
1631 * But if system has 2), 2) will definitely be used, so in this
1632 * case we don't need to call rtc_suspend(), and this is what
1633 * timekeeping_rtc_skipsuspend() means.
1635 bool timekeeping_rtc_skipsuspend(void)
1637 return persistent_clock_exists;
1641 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1642 * @delta: pointer to a timespec64 delta value
1644 * This hook is for architectures that cannot support read_persistent_clock64
1645 * because their RTC/persistent clock is only accessible when irqs are enabled.
1646 * and also don't have an effective nonstop clocksource.
1648 * This function should only be called by rtc_resume(), and allows
1649 * a suspend offset to be injected into the timekeeping values.
1651 void timekeeping_inject_sleeptime64(const struct timespec64 *delta)
1653 struct timekeeper *tk = &tk_core.timekeeper;
1654 unsigned long flags;
1656 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1657 write_seqcount_begin(&tk_core.seq);
1659 suspend_timing_needed = false;
1661 timekeeping_forward_now(tk);
1663 __timekeeping_inject_sleeptime(tk, delta);
1665 timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
1667 write_seqcount_end(&tk_core.seq);
1668 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1670 /* signal hrtimers about time change */
1676 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1678 void timekeeping_resume(void)
1680 struct timekeeper *tk = &tk_core.timekeeper;
1681 struct clocksource *clock = tk->tkr_mono.clock;
1682 unsigned long flags;
1683 struct timespec64 ts_new, ts_delta;
1684 u64 cycle_now, nsec;
1685 bool inject_sleeptime = false;
1687 read_persistent_clock64(&ts_new);
1689 clockevents_resume();
1690 clocksource_resume();
1692 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1693 write_seqcount_begin(&tk_core.seq);
1696 * After system resumes, we need to calculate the suspended time and
1697 * compensate it for the OS time. There are 3 sources that could be
1698 * used: Nonstop clocksource during suspend, persistent clock and rtc
1701 * One specific platform may have 1 or 2 or all of them, and the
1702 * preference will be:
1703 * suspend-nonstop clocksource -> persistent clock -> rtc
1704 * The less preferred source will only be tried if there is no better
1705 * usable source. The rtc part is handled separately in rtc core code.
1707 cycle_now = tk_clock_read(&tk->tkr_mono);
1708 nsec = clocksource_stop_suspend_timing(clock, cycle_now);
1710 ts_delta = ns_to_timespec64(nsec);
1711 inject_sleeptime = true;
1712 } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
1713 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
1714 inject_sleeptime = true;
1717 if (inject_sleeptime) {
1718 suspend_timing_needed = false;
1719 __timekeeping_inject_sleeptime(tk, &ts_delta);
1722 /* Re-base the last cycle value */
1723 tk->tkr_mono.cycle_last = cycle_now;
1724 tk->tkr_raw.cycle_last = cycle_now;
1727 timekeeping_suspended = 0;
1728 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1729 write_seqcount_end(&tk_core.seq);
1730 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1732 touch_softlockup_watchdog();
1738 int timekeeping_suspend(void)
1740 struct timekeeper *tk = &tk_core.timekeeper;
1741 unsigned long flags;
1742 struct timespec64 delta, delta_delta;
1743 static struct timespec64 old_delta;
1744 struct clocksource *curr_clock;
1747 read_persistent_clock64(&timekeeping_suspend_time);
1750 * On some systems the persistent_clock can not be detected at
1751 * timekeeping_init by its return value, so if we see a valid
1752 * value returned, update the persistent_clock_exists flag.
1754 if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1755 persistent_clock_exists = true;
1757 suspend_timing_needed = true;
1759 raw_spin_lock_irqsave(&timekeeper_lock, flags);
1760 write_seqcount_begin(&tk_core.seq);
1761 timekeeping_forward_now(tk);
1762 timekeeping_suspended = 1;
1765 * Since we've called forward_now, cycle_last stores the value
1766 * just read from the current clocksource. Save this to potentially
1767 * use in suspend timing.
1769 curr_clock = tk->tkr_mono.clock;
1770 cycle_now = tk->tkr_mono.cycle_last;
1771 clocksource_start_suspend_timing(curr_clock, cycle_now);
1773 if (persistent_clock_exists) {
1775 * To avoid drift caused by repeated suspend/resumes,
1776 * which each can add ~1 second drift error,
1777 * try to compensate so the difference in system time
1778 * and persistent_clock time stays close to constant.
1780 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
1781 delta_delta = timespec64_sub(delta, old_delta);
1782 if (abs(delta_delta.tv_sec) >= 2) {
1784 * if delta_delta is too large, assume time correction
1785 * has occurred and set old_delta to the current delta.
1789 /* Otherwise try to adjust old_system to compensate */
1790 timekeeping_suspend_time =
1791 timespec64_add(timekeeping_suspend_time, delta_delta);
1795 timekeeping_update(tk, TK_MIRROR);
1796 halt_fast_timekeeper(tk);
1797 write_seqcount_end(&tk_core.seq);
1798 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1801 clocksource_suspend();
1802 clockevents_suspend();
1807 /* sysfs resume/suspend bits for timekeeping */
1808 static struct syscore_ops timekeeping_syscore_ops = {
1809 .resume = timekeeping_resume,
1810 .suspend = timekeeping_suspend,
1813 static int __init timekeeping_init_ops(void)
1815 register_syscore_ops(&timekeeping_syscore_ops);
1818 device_initcall(timekeeping_init_ops);
1821 * Apply a multiplier adjustment to the timekeeper
1823 static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
1827 s64 interval = tk->cycle_interval;
1829 if (mult_adj == 0) {
1831 } else if (mult_adj == -1) {
1832 interval = -interval;
1834 } else if (mult_adj != 1) {
1835 interval *= mult_adj;
1840 * So the following can be confusing.
1842 * To keep things simple, lets assume mult_adj == 1 for now.
1844 * When mult_adj != 1, remember that the interval and offset values
1845 * have been appropriately scaled so the math is the same.
1847 * The basic idea here is that we're increasing the multiplier
1848 * by one, this causes the xtime_interval to be incremented by
1849 * one cycle_interval. This is because:
1850 * xtime_interval = cycle_interval * mult
1851 * So if mult is being incremented by one:
1852 * xtime_interval = cycle_interval * (mult + 1)
1854 * xtime_interval = (cycle_interval * mult) + cycle_interval
1855 * Which can be shortened to:
1856 * xtime_interval += cycle_interval
1858 * So offset stores the non-accumulated cycles. Thus the current
1859 * time (in shifted nanoseconds) is:
1860 * now = (offset * adj) + xtime_nsec
1861 * Now, even though we're adjusting the clock frequency, we have
1862 * to keep time consistent. In other words, we can't jump back
1863 * in time, and we also want to avoid jumping forward in time.
1865 * So given the same offset value, we need the time to be the same
1866 * both before and after the freq adjustment.
1867 * now = (offset * adj_1) + xtime_nsec_1
1868 * now = (offset * adj_2) + xtime_nsec_2
1870 * (offset * adj_1) + xtime_nsec_1 =
1871 * (offset * adj_2) + xtime_nsec_2
1875 * (offset * adj_1) + xtime_nsec_1 =
1876 * (offset * (adj_1+1)) + xtime_nsec_2
1877 * (offset * adj_1) + xtime_nsec_1 =
1878 * (offset * adj_1) + offset + xtime_nsec_2
1879 * Canceling the sides:
1880 * xtime_nsec_1 = offset + xtime_nsec_2
1882 * xtime_nsec_2 = xtime_nsec_1 - offset
1883 * Which simplfies to:
1884 * xtime_nsec -= offset
1886 if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) {
1887 /* NTP adjustment caused clocksource mult overflow */
1892 tk->tkr_mono.mult += mult_adj;
1893 tk->xtime_interval += interval;
1894 tk->tkr_mono.xtime_nsec -= offset;
1898 * Adjust the timekeeper's multiplier to the correct frequency
1899 * and also to reduce the accumulated error value.
1901 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1906 * Determine the multiplier from the current NTP tick length.
1907 * Avoid expensive division when the tick length doesn't change.
1909 if (likely(tk->ntp_tick == ntp_tick_length())) {
1910 mult = tk->tkr_mono.mult - tk->ntp_err_mult;
1912 tk->ntp_tick = ntp_tick_length();
1913 mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) -
1914 tk->xtime_remainder, tk->cycle_interval);
1918 * If the clock is behind the NTP time, increase the multiplier by 1
1919 * to catch up with it. If it's ahead and there was a remainder in the
1920 * tick division, the clock will slow down. Otherwise it will stay
1921 * ahead until the tick length changes to a non-divisible value.
1923 tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0;
1924 mult += tk->ntp_err_mult;
1926 timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult);
1928 if (unlikely(tk->tkr_mono.clock->maxadj &&
1929 (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult)
1930 > tk->tkr_mono.clock->maxadj))) {
1931 printk_once(KERN_WARNING
1932 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1933 tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult,
1934 (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj);
1938 * It may be possible that when we entered this function, xtime_nsec
1939 * was very small. Further, if we're slightly speeding the clocksource
1940 * in the code above, its possible the required corrective factor to
1941 * xtime_nsec could cause it to underflow.
1943 * Now, since we have already accumulated the second and the NTP
1944 * subsystem has been notified via second_overflow(), we need to skip
1947 if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) {
1948 tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC <<
1951 tk->skip_second_overflow = 1;
1956 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1958 * Helper function that accumulates the nsecs greater than a second
1959 * from the xtime_nsec field to the xtime_secs field.
1960 * It also calls into the NTP code to handle leapsecond processing.
1963 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1965 u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
1966 unsigned int clock_set = 0;
1968 while (tk->tkr_mono.xtime_nsec >= nsecps) {
1971 tk->tkr_mono.xtime_nsec -= nsecps;
1975 * Skip NTP update if this second was accumulated before,
1976 * i.e. xtime_nsec underflowed in timekeeping_adjust()
1978 if (unlikely(tk->skip_second_overflow)) {
1979 tk->skip_second_overflow = 0;
1983 /* Figure out if its a leap sec and apply if needed */
1984 leap = second_overflow(tk->xtime_sec);
1985 if (unlikely(leap)) {
1986 struct timespec64 ts;
1988 tk->xtime_sec += leap;
1992 tk_set_wall_to_mono(tk,
1993 timespec64_sub(tk->wall_to_monotonic, ts));
1995 __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1997 clock_set = TK_CLOCK_WAS_SET;
2004 * logarithmic_accumulation - shifted accumulation of cycles
2006 * This functions accumulates a shifted interval of cycles into
2007 * a shifted interval nanoseconds. Allows for O(log) accumulation
2010 * Returns the unconsumed cycles.
2012 static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
2013 u32 shift, unsigned int *clock_set)
2015 u64 interval = tk->cycle_interval << shift;
2018 /* If the offset is smaller than a shifted interval, do nothing */
2019 if (offset < interval)
2022 /* Accumulate one shifted interval */
2024 tk->tkr_mono.cycle_last += interval;
2025 tk->tkr_raw.cycle_last += interval;
2027 tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift;
2028 *clock_set |= accumulate_nsecs_to_secs(tk);
2030 /* Accumulate raw time */
2031 tk->tkr_raw.xtime_nsec += tk->raw_interval << shift;
2032 snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
2033 while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) {
2034 tk->tkr_raw.xtime_nsec -= snsec_per_sec;
2038 /* Accumulate error between NTP and clock interval */
2039 tk->ntp_error += tk->ntp_tick << shift;
2040 tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
2041 (tk->ntp_error_shift + shift);
2047 * timekeeping_advance - Updates the timekeeper to the current time and
2048 * current NTP tick length
2050 static void timekeeping_advance(enum timekeeping_adv_mode mode)
2052 struct timekeeper *real_tk = &tk_core.timekeeper;
2053 struct timekeeper *tk = &shadow_timekeeper;
2055 int shift = 0, maxshift;
2056 unsigned int clock_set = 0;
2057 unsigned long flags;
2059 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2061 /* Make sure we're fully resumed: */
2062 if (unlikely(timekeeping_suspended))
2065 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
2066 offset = real_tk->cycle_interval;
2068 if (mode != TK_ADV_TICK)
2071 offset = clocksource_delta(tk_clock_read(&tk->tkr_mono),
2072 tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
2074 /* Check if there's really nothing to do */
2075 if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK)
2079 /* Do some additional sanity checking */
2080 timekeeping_check_update(tk, offset);
2083 * With NO_HZ we may have to accumulate many cycle_intervals
2084 * (think "ticks") worth of time at once. To do this efficiently,
2085 * we calculate the largest doubling multiple of cycle_intervals
2086 * that is smaller than the offset. We then accumulate that
2087 * chunk in one go, and then try to consume the next smaller
2090 shift = ilog2(offset) - ilog2(tk->cycle_interval);
2091 shift = max(0, shift);
2092 /* Bound shift to one less than what overflows tick_length */
2093 maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
2094 shift = min(shift, maxshift);
2095 while (offset >= tk->cycle_interval) {
2096 offset = logarithmic_accumulation(tk, offset, shift,
2098 if (offset < tk->cycle_interval<<shift)
2102 /* Adjust the multiplier to correct NTP error */
2103 timekeeping_adjust(tk, offset);
2106 * Finally, make sure that after the rounding
2107 * xtime_nsec isn't larger than NSEC_PER_SEC
2109 clock_set |= accumulate_nsecs_to_secs(tk);
2111 write_seqcount_begin(&tk_core.seq);
2113 * Update the real timekeeper.
2115 * We could avoid this memcpy by switching pointers, but that
2116 * requires changes to all other timekeeper usage sites as
2117 * well, i.e. move the timekeeper pointer getter into the
2118 * spinlocked/seqcount protected sections. And we trade this
2119 * memcpy under the tk_core.seq against one before we start
2122 timekeeping_update(tk, clock_set);
2123 memcpy(real_tk, tk, sizeof(*tk));
2124 /* The memcpy must come last. Do not put anything here! */
2125 write_seqcount_end(&tk_core.seq);
2127 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2129 /* Have to call _delayed version, since in irq context*/
2130 clock_was_set_delayed();
2134 * update_wall_time - Uses the current clocksource to increment the wall time
2137 void update_wall_time(void)
2139 timekeeping_advance(TK_ADV_TICK);
2143 * getboottime64 - Return the real time of system boot.
2144 * @ts: pointer to the timespec64 to be set
2146 * Returns the wall-time of boot in a timespec64.
2148 * This is based on the wall_to_monotonic offset and the total suspend
2149 * time. Calls to settimeofday will affect the value returned (which
2150 * basically means that however wrong your real time clock is at boot time,
2151 * you get the right time here).
2153 void getboottime64(struct timespec64 *ts)
2155 struct timekeeper *tk = &tk_core.timekeeper;
2156 ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
2158 *ts = ktime_to_timespec64(t);
2160 EXPORT_SYMBOL_GPL(getboottime64);
2162 void ktime_get_coarse_real_ts64(struct timespec64 *ts)
2164 struct timekeeper *tk = &tk_core.timekeeper;
2168 seq = read_seqcount_begin(&tk_core.seq);
2171 } while (read_seqcount_retry(&tk_core.seq, seq));
2173 EXPORT_SYMBOL(ktime_get_coarse_real_ts64);
2175 void ktime_get_coarse_ts64(struct timespec64 *ts)
2177 struct timekeeper *tk = &tk_core.timekeeper;
2178 struct timespec64 now, mono;
2182 seq = read_seqcount_begin(&tk_core.seq);
2185 mono = tk->wall_to_monotonic;
2186 } while (read_seqcount_retry(&tk_core.seq, seq));
2188 set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec,
2189 now.tv_nsec + mono.tv_nsec);
2191 EXPORT_SYMBOL(ktime_get_coarse_ts64);
2194 * Must hold jiffies_lock
2196 void do_timer(unsigned long ticks)
2198 jiffies_64 += ticks;
2203 * ktime_get_update_offsets_now - hrtimer helper
2204 * @cwsseq: pointer to check and store the clock was set sequence number
2205 * @offs_real: pointer to storage for monotonic -> realtime offset
2206 * @offs_boot: pointer to storage for monotonic -> boottime offset
2207 * @offs_tai: pointer to storage for monotonic -> clock tai offset
2209 * Returns current monotonic time and updates the offsets if the
2210 * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2213 * Called from hrtimer_interrupt() or retrigger_next_event()
2215 ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real,
2216 ktime_t *offs_boot, ktime_t *offs_tai)
2218 struct timekeeper *tk = &tk_core.timekeeper;
2224 seq = read_seqcount_begin(&tk_core.seq);
2226 base = tk->tkr_mono.base;
2227 nsecs = timekeeping_get_ns(&tk->tkr_mono);
2228 base = ktime_add_ns(base, nsecs);
2230 if (*cwsseq != tk->clock_was_set_seq) {
2231 *cwsseq = tk->clock_was_set_seq;
2232 *offs_real = tk->offs_real;
2233 *offs_boot = tk->offs_boot;
2234 *offs_tai = tk->offs_tai;
2237 /* Handle leapsecond insertion adjustments */
2238 if (unlikely(base >= tk->next_leap_ktime))
2239 *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0));
2241 } while (read_seqcount_retry(&tk_core.seq, seq));
2247 * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2249 static int timekeeping_validate_timex(const struct __kernel_timex *txc)
2251 if (txc->modes & ADJ_ADJTIME) {
2252 /* singleshot must not be used with any other mode bits */
2253 if (!(txc->modes & ADJ_OFFSET_SINGLESHOT))
2255 if (!(txc->modes & ADJ_OFFSET_READONLY) &&
2256 !capable(CAP_SYS_TIME))
2259 /* In order to modify anything, you gotta be super-user! */
2260 if (txc->modes && !capable(CAP_SYS_TIME))
2263 * if the quartz is off by more than 10% then
2264 * something is VERY wrong!
2266 if (txc->modes & ADJ_TICK &&
2267 (txc->tick < 900000/USER_HZ ||
2268 txc->tick > 1100000/USER_HZ))
2272 if (txc->modes & ADJ_SETOFFSET) {
2273 /* In order to inject time, you gotta be super-user! */
2274 if (!capable(CAP_SYS_TIME))
2278 * Validate if a timespec/timeval used to inject a time
2279 * offset is valid. Offsets can be postive or negative, so
2280 * we don't check tv_sec. The value of the timeval/timespec
2281 * is the sum of its fields,but *NOTE*:
2282 * The field tv_usec/tv_nsec must always be non-negative and
2283 * we can't have more nanoseconds/microseconds than a second.
2285 if (txc->time.tv_usec < 0)
2288 if (txc->modes & ADJ_NANO) {
2289 if (txc->time.tv_usec >= NSEC_PER_SEC)
2292 if (txc->time.tv_usec >= USEC_PER_SEC)
2298 * Check for potential multiplication overflows that can
2299 * only happen on 64-bit systems:
2301 if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) {
2302 if (LLONG_MIN / PPM_SCALE > txc->freq)
2304 if (LLONG_MAX / PPM_SCALE < txc->freq)
2313 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2315 int do_adjtimex(struct __kernel_timex *txc)
2317 struct timekeeper *tk = &tk_core.timekeeper;
2318 struct audit_ntp_data ad;
2319 unsigned long flags;
2320 struct timespec64 ts;
2324 /* Validate the data before disabling interrupts */
2325 ret = timekeeping_validate_timex(txc);
2329 if (txc->modes & ADJ_SETOFFSET) {
2330 struct timespec64 delta;
2331 delta.tv_sec = txc->time.tv_sec;
2332 delta.tv_nsec = txc->time.tv_usec;
2333 if (!(txc->modes & ADJ_NANO))
2334 delta.tv_nsec *= 1000;
2335 ret = timekeeping_inject_offset(&delta);
2339 audit_tk_injoffset(delta);
2342 audit_ntp_init(&ad);
2344 ktime_get_real_ts64(&ts);
2346 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2347 write_seqcount_begin(&tk_core.seq);
2349 orig_tai = tai = tk->tai_offset;
2350 ret = __do_adjtimex(txc, &ts, &tai, &ad);
2352 if (tai != orig_tai) {
2353 __timekeeping_set_tai_offset(tk, tai);
2354 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
2356 tk_update_leap_state(tk);
2358 write_seqcount_end(&tk_core.seq);
2359 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2363 /* Update the multiplier immediately if frequency was set directly */
2364 if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK))
2365 timekeeping_advance(TK_ADV_FREQ);
2367 if (tai != orig_tai)
2370 ntp_notify_cmos_timer();
2375 #ifdef CONFIG_NTP_PPS
2377 * hardpps() - Accessor function to NTP __hardpps function
2379 void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts)
2381 unsigned long flags;
2383 raw_spin_lock_irqsave(&timekeeper_lock, flags);
2384 write_seqcount_begin(&tk_core.seq);
2386 __hardpps(phase_ts, raw_ts);
2388 write_seqcount_end(&tk_core.seq);
2389 raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
2391 EXPORT_SYMBOL(hardpps);
2392 #endif /* CONFIG_NTP_PPS */
2395 * xtime_update() - advances the timekeeping infrastructure
2396 * @ticks: number of ticks, that have elapsed since the last call.
2398 * Must be called with interrupts disabled.
2400 void xtime_update(unsigned long ticks)
2402 raw_spin_lock(&jiffies_lock);
2403 write_seqcount_begin(&jiffies_seq);
2405 write_seqcount_end(&jiffies_seq);
2406 raw_spin_unlock(&jiffies_lock);