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 */
43 * The most important data for readout fits into a single 64 byte
48 struct timekeeper timekeeper
;
49 } tk_core ____cacheline_aligned
= {
50 .seq
= SEQCNT_ZERO(tk_core
.seq
),
53 static DEFINE_RAW_SPINLOCK(timekeeper_lock
);
54 static struct timekeeper shadow_timekeeper
;
57 * struct tk_fast - NMI safe timekeeper
58 * @seq: Sequence counter for protecting updates. The lowest bit
59 * is the index for the tk_read_base array
60 * @base: tk_read_base array. Access is indexed by the lowest bit of
63 * See @update_fast_timekeeper() below.
67 struct tk_read_base base
[2];
70 /* Suspend-time cycles value for halted fast timekeeper. */
71 static u64 cycles_at_suspend
;
73 static u64
dummy_clock_read(struct clocksource
*cs
)
75 return cycles_at_suspend
;
78 static struct clocksource dummy_clock
= {
79 .read
= dummy_clock_read
,
82 static struct tk_fast tk_fast_mono ____cacheline_aligned
= {
83 .base
[0] = { .clock
= &dummy_clock
, },
84 .base
[1] = { .clock
= &dummy_clock
, },
87 static struct tk_fast tk_fast_raw ____cacheline_aligned
= {
88 .base
[0] = { .clock
= &dummy_clock
, },
89 .base
[1] = { .clock
= &dummy_clock
, },
92 /* flag for if timekeeping is suspended */
93 int __read_mostly timekeeping_suspended
;
95 static inline void tk_normalize_xtime(struct timekeeper
*tk
)
97 while (tk
->tkr_mono
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
)) {
98 tk
->tkr_mono
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
101 while (tk
->tkr_raw
.xtime_nsec
>= ((u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
)) {
102 tk
->tkr_raw
.xtime_nsec
-= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
107 static inline struct timespec64
tk_xtime(const struct timekeeper
*tk
)
109 struct timespec64 ts
;
111 ts
.tv_sec
= tk
->xtime_sec
;
112 ts
.tv_nsec
= (long)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
116 static void tk_set_xtime(struct timekeeper
*tk
, const struct timespec64
*ts
)
118 tk
->xtime_sec
= ts
->tv_sec
;
119 tk
->tkr_mono
.xtime_nsec
= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
122 static void tk_xtime_add(struct timekeeper
*tk
, const struct timespec64
*ts
)
124 tk
->xtime_sec
+= ts
->tv_sec
;
125 tk
->tkr_mono
.xtime_nsec
+= (u64
)ts
->tv_nsec
<< tk
->tkr_mono
.shift
;
126 tk_normalize_xtime(tk
);
129 static void tk_set_wall_to_mono(struct timekeeper
*tk
, struct timespec64 wtm
)
131 struct timespec64 tmp
;
134 * Verify consistency of: offset_real = -wall_to_monotonic
135 * before modifying anything
137 set_normalized_timespec64(&tmp
, -tk
->wall_to_monotonic
.tv_sec
,
138 -tk
->wall_to_monotonic
.tv_nsec
);
139 WARN_ON_ONCE(tk
->offs_real
!= timespec64_to_ktime(tmp
));
140 tk
->wall_to_monotonic
= wtm
;
141 set_normalized_timespec64(&tmp
, -wtm
.tv_sec
, -wtm
.tv_nsec
);
142 tk
->offs_real
= timespec64_to_ktime(tmp
);
143 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tk
->tai_offset
, 0));
146 static inline void tk_update_sleep_time(struct timekeeper
*tk
, ktime_t delta
)
148 tk
->offs_boot
= ktime_add(tk
->offs_boot
, delta
);
150 * Timespec representation for VDSO update to avoid 64bit division
153 tk
->monotonic_to_boot
= ktime_to_timespec64(tk
->offs_boot
);
157 * tk_clock_read - atomic clocksource read() helper
159 * This helper is necessary to use in the read paths because, while the
160 * seqlock ensures we don't return a bad value while structures are updated,
161 * it doesn't protect from potential crashes. There is the possibility that
162 * the tkr's clocksource may change between the read reference, and the
163 * clock reference passed to the read function. This can cause crashes if
164 * the wrong clocksource is passed to the wrong read function.
165 * This isn't necessary to use when holding the timekeeper_lock or doing
166 * a read of the fast-timekeeper tkrs (which is protected by its own locking
169 static inline u64
tk_clock_read(const struct tk_read_base
*tkr
)
171 struct clocksource
*clock
= READ_ONCE(tkr
->clock
);
173 return clock
->read(clock
);
176 #ifdef CONFIG_DEBUG_TIMEKEEPING
177 #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
179 static void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
182 u64 max_cycles
= tk
->tkr_mono
.clock
->max_cycles
;
183 const char *name
= tk
->tkr_mono
.clock
->name
;
185 if (offset
> max_cycles
) {
186 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
187 offset
, name
, max_cycles
);
188 printk_deferred(" timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
190 if (offset
> (max_cycles
>> 1)) {
191 printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
192 offset
, name
, max_cycles
>> 1);
193 printk_deferred(" timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
197 if (tk
->underflow_seen
) {
198 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
199 printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name
);
200 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
201 printk_deferred(" Your kernel is probably still fine.\n");
202 tk
->last_warning
= jiffies
;
204 tk
->underflow_seen
= 0;
207 if (tk
->overflow_seen
) {
208 if (jiffies
- tk
->last_warning
> WARNING_FREQ
) {
209 printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name
);
210 printk_deferred(" Please report this, consider using a different clocksource, if possible.\n");
211 printk_deferred(" Your kernel is probably still fine.\n");
212 tk
->last_warning
= jiffies
;
214 tk
->overflow_seen
= 0;
218 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
220 struct timekeeper
*tk
= &tk_core
.timekeeper
;
221 u64 now
, last
, mask
, max
, delta
;
225 * Since we're called holding a seqlock, the data may shift
226 * under us while we're doing the calculation. This can cause
227 * false positives, since we'd note a problem but throw the
228 * results away. So nest another seqlock here to atomically
229 * grab the points we are checking with.
232 seq
= read_seqcount_begin(&tk_core
.seq
);
233 now
= tk_clock_read(tkr
);
234 last
= tkr
->cycle_last
;
236 max
= tkr
->clock
->max_cycles
;
237 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
239 delta
= clocksource_delta(now
, last
, mask
);
242 * Try to catch underflows by checking if we are seeing small
243 * mask-relative negative values.
245 if (unlikely((~delta
& mask
) < (mask
>> 3))) {
246 tk
->underflow_seen
= 1;
250 /* Cap delta value to the max_cycles values to avoid mult overflows */
251 if (unlikely(delta
> max
)) {
252 tk
->overflow_seen
= 1;
253 delta
= tkr
->clock
->max_cycles
;
259 static inline void timekeeping_check_update(struct timekeeper
*tk
, u64 offset
)
262 static inline u64
timekeeping_get_delta(const struct tk_read_base
*tkr
)
264 u64 cycle_now
, delta
;
266 /* read clocksource */
267 cycle_now
= tk_clock_read(tkr
);
269 /* calculate the delta since the last update_wall_time */
270 delta
= clocksource_delta(cycle_now
, tkr
->cycle_last
, tkr
->mask
);
277 * tk_setup_internals - Set up internals to use clocksource clock.
279 * @tk: The target timekeeper to setup.
280 * @clock: Pointer to clocksource.
282 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
283 * pair and interval request.
285 * Unless you're the timekeeping code, you should not be using this!
287 static void tk_setup_internals(struct timekeeper
*tk
, struct clocksource
*clock
)
290 u64 tmp
, ntpinterval
;
291 struct clocksource
*old_clock
;
293 ++tk
->cs_was_changed_seq
;
294 old_clock
= tk
->tkr_mono
.clock
;
295 tk
->tkr_mono
.clock
= clock
;
296 tk
->tkr_mono
.mask
= clock
->mask
;
297 tk
->tkr_mono
.cycle_last
= tk_clock_read(&tk
->tkr_mono
);
299 tk
->tkr_raw
.clock
= clock
;
300 tk
->tkr_raw
.mask
= clock
->mask
;
301 tk
->tkr_raw
.cycle_last
= tk
->tkr_mono
.cycle_last
;
303 /* Do the ns -> cycle conversion first, using original mult */
304 tmp
= NTP_INTERVAL_LENGTH
;
305 tmp
<<= clock
->shift
;
307 tmp
+= clock
->mult
/2;
308 do_div(tmp
, clock
->mult
);
312 interval
= (u64
) tmp
;
313 tk
->cycle_interval
= interval
;
315 /* Go back from cycles -> shifted ns */
316 tk
->xtime_interval
= interval
* clock
->mult
;
317 tk
->xtime_remainder
= ntpinterval
- tk
->xtime_interval
;
318 tk
->raw_interval
= interval
* clock
->mult
;
320 /* if changing clocks, convert xtime_nsec shift units */
322 int shift_change
= clock
->shift
- old_clock
->shift
;
323 if (shift_change
< 0) {
324 tk
->tkr_mono
.xtime_nsec
>>= -shift_change
;
325 tk
->tkr_raw
.xtime_nsec
>>= -shift_change
;
327 tk
->tkr_mono
.xtime_nsec
<<= shift_change
;
328 tk
->tkr_raw
.xtime_nsec
<<= shift_change
;
332 tk
->tkr_mono
.shift
= clock
->shift
;
333 tk
->tkr_raw
.shift
= clock
->shift
;
336 tk
->ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
337 tk
->ntp_tick
= ntpinterval
<< tk
->ntp_error_shift
;
340 * The timekeeper keeps its own mult values for the currently
341 * active clocksource. These value will be adjusted via NTP
342 * to counteract clock drifting.
344 tk
->tkr_mono
.mult
= clock
->mult
;
345 tk
->tkr_raw
.mult
= clock
->mult
;
346 tk
->ntp_err_mult
= 0;
347 tk
->skip_second_overflow
= 0;
350 /* Timekeeper helper functions. */
352 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
353 static u32
default_arch_gettimeoffset(void) { return 0; }
354 u32 (*arch_gettimeoffset
)(void) = default_arch_gettimeoffset
;
356 static inline u32
arch_gettimeoffset(void) { return 0; }
359 static inline u64
timekeeping_delta_to_ns(const struct tk_read_base
*tkr
, u64 delta
)
363 nsec
= delta
* tkr
->mult
+ tkr
->xtime_nsec
;
366 /* If arch requires, add in get_arch_timeoffset() */
367 return nsec
+ arch_gettimeoffset();
370 static inline u64
timekeeping_get_ns(const struct tk_read_base
*tkr
)
374 delta
= timekeeping_get_delta(tkr
);
375 return timekeeping_delta_to_ns(tkr
, delta
);
378 static inline u64
timekeeping_cycles_to_ns(const struct tk_read_base
*tkr
, u64 cycles
)
382 /* calculate the delta since the last update_wall_time */
383 delta
= clocksource_delta(cycles
, tkr
->cycle_last
, tkr
->mask
);
384 return timekeeping_delta_to_ns(tkr
, delta
);
388 * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
389 * @tkr: Timekeeping readout base from which we take the update
391 * We want to use this from any context including NMI and tracing /
392 * instrumenting the timekeeping code itself.
394 * Employ the latch technique; see @raw_write_seqcount_latch.
396 * So if a NMI hits the update of base[0] then it will use base[1]
397 * which is still consistent. In the worst case this can result is a
398 * slightly wrong timestamp (a few nanoseconds). See
399 * @ktime_get_mono_fast_ns.
401 static void update_fast_timekeeper(const struct tk_read_base
*tkr
,
404 struct tk_read_base
*base
= tkf
->base
;
406 /* Force readers off to base[1] */
407 raw_write_seqcount_latch(&tkf
->seq
);
410 memcpy(base
, tkr
, sizeof(*base
));
412 /* Force readers back to base[0] */
413 raw_write_seqcount_latch(&tkf
->seq
);
416 memcpy(base
+ 1, base
, sizeof(*base
));
420 * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
422 * This timestamp is not guaranteed to be monotonic across an update.
423 * The timestamp is calculated by:
425 * now = base_mono + clock_delta * slope
427 * So if the update lowers the slope, readers who are forced to the
428 * not yet updated second array are still using the old steeper slope.
437 * |12345678---> reader order
443 * So reader 6 will observe time going backwards versus reader 5.
445 * While other CPUs are likely to be able observe that, the only way
446 * for a CPU local observation is when an NMI hits in the middle of
447 * the update. Timestamps taken from that NMI context might be ahead
448 * of the following timestamps. Callers need to be aware of that and
451 static __always_inline u64
__ktime_get_fast_ns(struct tk_fast
*tkf
)
453 struct tk_read_base
*tkr
;
458 seq
= raw_read_seqcount_latch(&tkf
->seq
);
459 tkr
= tkf
->base
+ (seq
& 0x01);
460 now
= ktime_to_ns(tkr
->base
);
462 now
+= timekeeping_delta_to_ns(tkr
,
467 } while (read_seqcount_retry(&tkf
->seq
, seq
));
472 u64
ktime_get_mono_fast_ns(void)
474 return __ktime_get_fast_ns(&tk_fast_mono
);
476 EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns
);
478 u64
ktime_get_raw_fast_ns(void)
480 return __ktime_get_fast_ns(&tk_fast_raw
);
482 EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns
);
485 * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
487 * To keep it NMI safe since we're accessing from tracing, we're not using a
488 * separate timekeeper with updates to monotonic clock and boot offset
489 * protected with seqlocks. This has the following minor side effects:
491 * (1) Its possible that a timestamp be taken after the boot offset is updated
492 * but before the timekeeper is updated. If this happens, the new boot offset
493 * is added to the old timekeeping making the clock appear to update slightly
496 * timekeeping_inject_sleeptime64()
497 * __timekeeping_inject_sleeptime(tk, delta);
499 * timekeeping_update(tk, TK_CLEAR_NTP...);
501 * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
502 * partially updated. Since the tk->offs_boot update is a rare event, this
503 * should be a rare occurrence which postprocessing should be able to handle.
505 u64 notrace
ktime_get_boot_fast_ns(void)
507 struct timekeeper
*tk
= &tk_core
.timekeeper
;
509 return (ktime_get_mono_fast_ns() + ktime_to_ns(tk
->offs_boot
));
511 EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns
);
515 * See comment for __ktime_get_fast_ns() vs. timestamp ordering
517 static __always_inline u64
__ktime_get_real_fast_ns(struct tk_fast
*tkf
)
519 struct tk_read_base
*tkr
;
524 seq
= raw_read_seqcount_latch(&tkf
->seq
);
525 tkr
= tkf
->base
+ (seq
& 0x01);
526 now
= ktime_to_ns(tkr
->base_real
);
528 now
+= timekeeping_delta_to_ns(tkr
,
533 } while (read_seqcount_retry(&tkf
->seq
, seq
));
539 * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
541 u64
ktime_get_real_fast_ns(void)
543 return __ktime_get_real_fast_ns(&tk_fast_mono
);
545 EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns
);
548 * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
549 * @tk: Timekeeper to snapshot.
551 * It generally is unsafe to access the clocksource after timekeeping has been
552 * suspended, so take a snapshot of the readout base of @tk and use it as the
553 * fast timekeeper's readout base while suspended. It will return the same
554 * number of cycles every time until timekeeping is resumed at which time the
555 * proper readout base for the fast timekeeper will be restored automatically.
557 static void halt_fast_timekeeper(const struct timekeeper
*tk
)
559 static struct tk_read_base tkr_dummy
;
560 const struct tk_read_base
*tkr
= &tk
->tkr_mono
;
562 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
563 cycles_at_suspend
= tk_clock_read(tkr
);
564 tkr_dummy
.clock
= &dummy_clock
;
565 tkr_dummy
.base_real
= tkr
->base
+ tk
->offs_real
;
566 update_fast_timekeeper(&tkr_dummy
, &tk_fast_mono
);
569 memcpy(&tkr_dummy
, tkr
, sizeof(tkr_dummy
));
570 tkr_dummy
.clock
= &dummy_clock
;
571 update_fast_timekeeper(&tkr_dummy
, &tk_fast_raw
);
574 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain
);
576 static void update_pvclock_gtod(struct timekeeper
*tk
, bool was_set
)
578 raw_notifier_call_chain(&pvclock_gtod_chain
, was_set
, tk
);
582 * pvclock_gtod_register_notifier - register a pvclock timedata update listener
584 int pvclock_gtod_register_notifier(struct notifier_block
*nb
)
586 struct timekeeper
*tk
= &tk_core
.timekeeper
;
590 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
591 ret
= raw_notifier_chain_register(&pvclock_gtod_chain
, nb
);
592 update_pvclock_gtod(tk
, true);
593 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
597 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier
);
600 * pvclock_gtod_unregister_notifier - unregister a pvclock
601 * timedata update listener
603 int pvclock_gtod_unregister_notifier(struct notifier_block
*nb
)
608 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
609 ret
= raw_notifier_chain_unregister(&pvclock_gtod_chain
, nb
);
610 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
614 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier
);
617 * tk_update_leap_state - helper to update the next_leap_ktime
619 static inline void tk_update_leap_state(struct timekeeper
*tk
)
621 tk
->next_leap_ktime
= ntp_get_next_leap();
622 if (tk
->next_leap_ktime
!= KTIME_MAX
)
623 /* Convert to monotonic time */
624 tk
->next_leap_ktime
= ktime_sub(tk
->next_leap_ktime
, tk
->offs_real
);
628 * Update the ktime_t based scalar nsec members of the timekeeper
630 static inline void tk_update_ktime_data(struct timekeeper
*tk
)
636 * The xtime based monotonic readout is:
637 * nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
638 * The ktime based monotonic readout is:
639 * nsec = base_mono + now();
640 * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
642 seconds
= (u64
)(tk
->xtime_sec
+ tk
->wall_to_monotonic
.tv_sec
);
643 nsec
= (u32
) tk
->wall_to_monotonic
.tv_nsec
;
644 tk
->tkr_mono
.base
= ns_to_ktime(seconds
* NSEC_PER_SEC
+ nsec
);
647 * The sum of the nanoseconds portions of xtime and
648 * wall_to_monotonic can be greater/equal one second. Take
649 * this into account before updating tk->ktime_sec.
651 nsec
+= (u32
)(tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
);
652 if (nsec
>= NSEC_PER_SEC
)
654 tk
->ktime_sec
= seconds
;
656 /* Update the monotonic raw base */
657 tk
->tkr_raw
.base
= ns_to_ktime(tk
->raw_sec
* NSEC_PER_SEC
);
660 /* must hold timekeeper_lock */
661 static void timekeeping_update(struct timekeeper
*tk
, unsigned int action
)
663 if (action
& TK_CLEAR_NTP
) {
668 tk_update_leap_state(tk
);
669 tk_update_ktime_data(tk
);
672 update_pvclock_gtod(tk
, action
& TK_CLOCK_WAS_SET
);
674 tk
->tkr_mono
.base_real
= tk
->tkr_mono
.base
+ tk
->offs_real
;
675 update_fast_timekeeper(&tk
->tkr_mono
, &tk_fast_mono
);
676 update_fast_timekeeper(&tk
->tkr_raw
, &tk_fast_raw
);
678 if (action
& TK_CLOCK_WAS_SET
)
679 tk
->clock_was_set_seq
++;
681 * The mirroring of the data to the shadow-timekeeper needs
682 * to happen last here to ensure we don't over-write the
683 * timekeeper structure on the next update with stale data
685 if (action
& TK_MIRROR
)
686 memcpy(&shadow_timekeeper
, &tk_core
.timekeeper
,
687 sizeof(tk_core
.timekeeper
));
691 * timekeeping_forward_now - update clock to the current time
693 * Forward the current clock to update its state since the last call to
694 * update_wall_time(). This is useful before significant clock changes,
695 * as it avoids having to deal with this time offset explicitly.
697 static void timekeeping_forward_now(struct timekeeper
*tk
)
699 u64 cycle_now
, delta
;
701 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
702 delta
= clocksource_delta(cycle_now
, tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
703 tk
->tkr_mono
.cycle_last
= cycle_now
;
704 tk
->tkr_raw
.cycle_last
= cycle_now
;
706 tk
->tkr_mono
.xtime_nsec
+= delta
* tk
->tkr_mono
.mult
;
708 /* If arch requires, add in get_arch_timeoffset() */
709 tk
->tkr_mono
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_mono
.shift
;
712 tk
->tkr_raw
.xtime_nsec
+= delta
* tk
->tkr_raw
.mult
;
714 /* If arch requires, add in get_arch_timeoffset() */
715 tk
->tkr_raw
.xtime_nsec
+= (u64
)arch_gettimeoffset() << tk
->tkr_raw
.shift
;
717 tk_normalize_xtime(tk
);
721 * ktime_get_real_ts64 - Returns the time of day in a timespec64.
722 * @ts: pointer to the timespec to be set
724 * Returns the time of day in a timespec64 (WARN if suspended).
726 void ktime_get_real_ts64(struct timespec64
*ts
)
728 struct timekeeper
*tk
= &tk_core
.timekeeper
;
732 WARN_ON(timekeeping_suspended
);
735 seq
= read_seqcount_begin(&tk_core
.seq
);
737 ts
->tv_sec
= tk
->xtime_sec
;
738 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
740 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
743 timespec64_add_ns(ts
, nsecs
);
745 EXPORT_SYMBOL(ktime_get_real_ts64
);
747 ktime_t
ktime_get(void)
749 struct timekeeper
*tk
= &tk_core
.timekeeper
;
754 WARN_ON(timekeeping_suspended
);
757 seq
= read_seqcount_begin(&tk_core
.seq
);
758 base
= tk
->tkr_mono
.base
;
759 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
761 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
763 return ktime_add_ns(base
, nsecs
);
765 EXPORT_SYMBOL_GPL(ktime_get
);
767 u32
ktime_get_resolution_ns(void)
769 struct timekeeper
*tk
= &tk_core
.timekeeper
;
773 WARN_ON(timekeeping_suspended
);
776 seq
= read_seqcount_begin(&tk_core
.seq
);
777 nsecs
= tk
->tkr_mono
.mult
>> tk
->tkr_mono
.shift
;
778 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
782 EXPORT_SYMBOL_GPL(ktime_get_resolution_ns
);
784 static ktime_t
*offsets
[TK_OFFS_MAX
] = {
785 [TK_OFFS_REAL
] = &tk_core
.timekeeper
.offs_real
,
786 [TK_OFFS_BOOT
] = &tk_core
.timekeeper
.offs_boot
,
787 [TK_OFFS_TAI
] = &tk_core
.timekeeper
.offs_tai
,
790 ktime_t
ktime_get_with_offset(enum tk_offsets offs
)
792 struct timekeeper
*tk
= &tk_core
.timekeeper
;
794 ktime_t base
, *offset
= offsets
[offs
];
797 WARN_ON(timekeeping_suspended
);
800 seq
= read_seqcount_begin(&tk_core
.seq
);
801 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
802 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
804 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
806 return ktime_add_ns(base
, nsecs
);
809 EXPORT_SYMBOL_GPL(ktime_get_with_offset
);
811 ktime_t
ktime_get_coarse_with_offset(enum tk_offsets offs
)
813 struct timekeeper
*tk
= &tk_core
.timekeeper
;
815 ktime_t base
, *offset
= offsets
[offs
];
818 WARN_ON(timekeeping_suspended
);
821 seq
= read_seqcount_begin(&tk_core
.seq
);
822 base
= ktime_add(tk
->tkr_mono
.base
, *offset
);
823 nsecs
= tk
->tkr_mono
.xtime_nsec
>> tk
->tkr_mono
.shift
;
825 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
827 return ktime_add_ns(base
, nsecs
);
829 EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset
);
832 * ktime_mono_to_any() - convert mononotic time to any other time
833 * @tmono: time to convert.
834 * @offs: which offset to use
836 ktime_t
ktime_mono_to_any(ktime_t tmono
, enum tk_offsets offs
)
838 ktime_t
*offset
= offsets
[offs
];
843 seq
= read_seqcount_begin(&tk_core
.seq
);
844 tconv
= ktime_add(tmono
, *offset
);
845 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
849 EXPORT_SYMBOL_GPL(ktime_mono_to_any
);
852 * ktime_get_raw - Returns the raw monotonic time in ktime_t format
854 ktime_t
ktime_get_raw(void)
856 struct timekeeper
*tk
= &tk_core
.timekeeper
;
862 seq
= read_seqcount_begin(&tk_core
.seq
);
863 base
= tk
->tkr_raw
.base
;
864 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
866 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
868 return ktime_add_ns(base
, nsecs
);
870 EXPORT_SYMBOL_GPL(ktime_get_raw
);
873 * ktime_get_ts64 - get the monotonic clock in timespec64 format
874 * @ts: pointer to timespec variable
876 * The function calculates the monotonic clock from the realtime
877 * clock and the wall_to_monotonic offset and stores the result
878 * in normalized timespec64 format in the variable pointed to by @ts.
880 void ktime_get_ts64(struct timespec64
*ts
)
882 struct timekeeper
*tk
= &tk_core
.timekeeper
;
883 struct timespec64 tomono
;
887 WARN_ON(timekeeping_suspended
);
890 seq
= read_seqcount_begin(&tk_core
.seq
);
891 ts
->tv_sec
= tk
->xtime_sec
;
892 nsec
= timekeeping_get_ns(&tk
->tkr_mono
);
893 tomono
= tk
->wall_to_monotonic
;
895 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
897 ts
->tv_sec
+= tomono
.tv_sec
;
899 timespec64_add_ns(ts
, nsec
+ tomono
.tv_nsec
);
901 EXPORT_SYMBOL_GPL(ktime_get_ts64
);
904 * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
906 * Returns the seconds portion of CLOCK_MONOTONIC with a single non
907 * serialized read. tk->ktime_sec is of type 'unsigned long' so this
908 * works on both 32 and 64 bit systems. On 32 bit systems the readout
909 * covers ~136 years of uptime which should be enough to prevent
910 * premature wrap arounds.
912 time64_t
ktime_get_seconds(void)
914 struct timekeeper
*tk
= &tk_core
.timekeeper
;
916 WARN_ON(timekeeping_suspended
);
917 return tk
->ktime_sec
;
919 EXPORT_SYMBOL_GPL(ktime_get_seconds
);
922 * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
924 * Returns the wall clock seconds since 1970. This replaces the
925 * get_seconds() interface which is not y2038 safe on 32bit systems.
927 * For 64bit systems the fast access to tk->xtime_sec is preserved. On
928 * 32bit systems the access must be protected with the sequence
929 * counter to provide "atomic" access to the 64bit tk->xtime_sec
932 time64_t
ktime_get_real_seconds(void)
934 struct timekeeper
*tk
= &tk_core
.timekeeper
;
938 if (IS_ENABLED(CONFIG_64BIT
))
939 return tk
->xtime_sec
;
942 seq
= read_seqcount_begin(&tk_core
.seq
);
943 seconds
= tk
->xtime_sec
;
945 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
949 EXPORT_SYMBOL_GPL(ktime_get_real_seconds
);
952 * __ktime_get_real_seconds - The same as ktime_get_real_seconds
953 * but without the sequence counter protect. This internal function
954 * is called just when timekeeping lock is already held.
956 time64_t
__ktime_get_real_seconds(void)
958 struct timekeeper
*tk
= &tk_core
.timekeeper
;
960 return tk
->xtime_sec
;
964 * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
965 * @systime_snapshot: pointer to struct receiving the system time snapshot
967 void ktime_get_snapshot(struct system_time_snapshot
*systime_snapshot
)
969 struct timekeeper
*tk
= &tk_core
.timekeeper
;
977 WARN_ON_ONCE(timekeeping_suspended
);
980 seq
= read_seqcount_begin(&tk_core
.seq
);
981 now
= tk_clock_read(&tk
->tkr_mono
);
982 systime_snapshot
->cs_was_changed_seq
= tk
->cs_was_changed_seq
;
983 systime_snapshot
->clock_was_set_seq
= tk
->clock_was_set_seq
;
984 base_real
= ktime_add(tk
->tkr_mono
.base
,
985 tk_core
.timekeeper
.offs_real
);
986 base_raw
= tk
->tkr_raw
.base
;
987 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
, now
);
988 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
, now
);
989 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
991 systime_snapshot
->cycles
= now
;
992 systime_snapshot
->real
= ktime_add_ns(base_real
, nsec_real
);
993 systime_snapshot
->raw
= ktime_add_ns(base_raw
, nsec_raw
);
995 EXPORT_SYMBOL_GPL(ktime_get_snapshot
);
997 /* Scale base by mult/div checking for overflow */
998 static int scale64_check_overflow(u64 mult
, u64 div
, u64
*base
)
1002 tmp
= div64_u64_rem(*base
, div
, &rem
);
1004 if (((int)sizeof(u64
)*8 - fls64(mult
) < fls64(tmp
)) ||
1005 ((int)sizeof(u64
)*8 - fls64(mult
) < fls64(rem
)))
1016 * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
1017 * @history: Snapshot representing start of history
1018 * @partial_history_cycles: Cycle offset into history (fractional part)
1019 * @total_history_cycles: Total history length in cycles
1020 * @discontinuity: True indicates clock was set on history period
1021 * @ts: Cross timestamp that should be adjusted using
1022 * partial/total ratio
1024 * Helper function used by get_device_system_crosststamp() to correct the
1025 * crosstimestamp corresponding to the start of the current interval to the
1026 * system counter value (timestamp point) provided by the driver. The
1027 * total_history_* quantities are the total history starting at the provided
1028 * reference point and ending at the start of the current interval. The cycle
1029 * count between the driver timestamp point and the start of the current
1030 * interval is partial_history_cycles.
1032 static int adjust_historical_crosststamp(struct system_time_snapshot
*history
,
1033 u64 partial_history_cycles
,
1034 u64 total_history_cycles
,
1036 struct system_device_crosststamp
*ts
)
1038 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1039 u64 corr_raw
, corr_real
;
1040 bool interp_forward
;
1043 if (total_history_cycles
== 0 || partial_history_cycles
== 0)
1046 /* Interpolate shortest distance from beginning or end of history */
1047 interp_forward
= partial_history_cycles
> total_history_cycles
/ 2;
1048 partial_history_cycles
= interp_forward
?
1049 total_history_cycles
- partial_history_cycles
:
1050 partial_history_cycles
;
1053 * Scale the monotonic raw time delta by:
1054 * partial_history_cycles / total_history_cycles
1056 corr_raw
= (u64
)ktime_to_ns(
1057 ktime_sub(ts
->sys_monoraw
, history
->raw
));
1058 ret
= scale64_check_overflow(partial_history_cycles
,
1059 total_history_cycles
, &corr_raw
);
1064 * If there is a discontinuity in the history, scale monotonic raw
1066 * mult(real)/mult(raw) yielding the realtime correction
1067 * Otherwise, calculate the realtime correction similar to monotonic
1070 if (discontinuity
) {
1071 corr_real
= mul_u64_u32_div
1072 (corr_raw
, tk
->tkr_mono
.mult
, tk
->tkr_raw
.mult
);
1074 corr_real
= (u64
)ktime_to_ns(
1075 ktime_sub(ts
->sys_realtime
, history
->real
));
1076 ret
= scale64_check_overflow(partial_history_cycles
,
1077 total_history_cycles
, &corr_real
);
1082 /* Fixup monotonic raw and real time time values */
1083 if (interp_forward
) {
1084 ts
->sys_monoraw
= ktime_add_ns(history
->raw
, corr_raw
);
1085 ts
->sys_realtime
= ktime_add_ns(history
->real
, corr_real
);
1087 ts
->sys_monoraw
= ktime_sub_ns(ts
->sys_monoraw
, corr_raw
);
1088 ts
->sys_realtime
= ktime_sub_ns(ts
->sys_realtime
, corr_real
);
1095 * cycle_between - true if test occurs chronologically between before and after
1097 static bool cycle_between(u64 before
, u64 test
, u64 after
)
1099 if (test
> before
&& test
< after
)
1101 if (test
< before
&& before
> after
)
1107 * get_device_system_crosststamp - Synchronously capture system/device timestamp
1108 * @get_time_fn: Callback to get simultaneous device time and
1109 * system counter from the device driver
1110 * @ctx: Context passed to get_time_fn()
1111 * @history_begin: Historical reference point used to interpolate system
1112 * time when counter provided by the driver is before the current interval
1113 * @xtstamp: Receives simultaneously captured system and device time
1115 * Reads a timestamp from a device and correlates it to system time
1117 int get_device_system_crosststamp(int (*get_time_fn
)
1118 (ktime_t
*device_time
,
1119 struct system_counterval_t
*sys_counterval
,
1122 struct system_time_snapshot
*history_begin
,
1123 struct system_device_crosststamp
*xtstamp
)
1125 struct system_counterval_t system_counterval
;
1126 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1127 u64 cycles
, now
, interval_start
;
1128 unsigned int clock_was_set_seq
= 0;
1129 ktime_t base_real
, base_raw
;
1130 u64 nsec_real
, nsec_raw
;
1131 u8 cs_was_changed_seq
;
1137 seq
= read_seqcount_begin(&tk_core
.seq
);
1139 * Try to synchronously capture device time and a system
1140 * counter value calling back into the device driver
1142 ret
= get_time_fn(&xtstamp
->device
, &system_counterval
, ctx
);
1147 * Verify that the clocksource associated with the captured
1148 * system counter value is the same as the currently installed
1149 * timekeeper clocksource
1151 if (tk
->tkr_mono
.clock
!= system_counterval
.cs
)
1153 cycles
= system_counterval
.cycles
;
1156 * Check whether the system counter value provided by the
1157 * device driver is on the current timekeeping interval.
1159 now
= tk_clock_read(&tk
->tkr_mono
);
1160 interval_start
= tk
->tkr_mono
.cycle_last
;
1161 if (!cycle_between(interval_start
, cycles
, now
)) {
1162 clock_was_set_seq
= tk
->clock_was_set_seq
;
1163 cs_was_changed_seq
= tk
->cs_was_changed_seq
;
1164 cycles
= interval_start
;
1170 base_real
= ktime_add(tk
->tkr_mono
.base
,
1171 tk_core
.timekeeper
.offs_real
);
1172 base_raw
= tk
->tkr_raw
.base
;
1174 nsec_real
= timekeeping_cycles_to_ns(&tk
->tkr_mono
,
1175 system_counterval
.cycles
);
1176 nsec_raw
= timekeeping_cycles_to_ns(&tk
->tkr_raw
,
1177 system_counterval
.cycles
);
1178 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1180 xtstamp
->sys_realtime
= ktime_add_ns(base_real
, nsec_real
);
1181 xtstamp
->sys_monoraw
= ktime_add_ns(base_raw
, nsec_raw
);
1184 * Interpolate if necessary, adjusting back from the start of the
1188 u64 partial_history_cycles
, total_history_cycles
;
1192 * Check that the counter value occurs after the provided
1193 * history reference and that the history doesn't cross a
1194 * clocksource change
1196 if (!history_begin
||
1197 !cycle_between(history_begin
->cycles
,
1198 system_counterval
.cycles
, cycles
) ||
1199 history_begin
->cs_was_changed_seq
!= cs_was_changed_seq
)
1201 partial_history_cycles
= cycles
- system_counterval
.cycles
;
1202 total_history_cycles
= cycles
- history_begin
->cycles
;
1204 history_begin
->clock_was_set_seq
!= clock_was_set_seq
;
1206 ret
= adjust_historical_crosststamp(history_begin
,
1207 partial_history_cycles
,
1208 total_history_cycles
,
1209 discontinuity
, xtstamp
);
1216 EXPORT_SYMBOL_GPL(get_device_system_crosststamp
);
1219 * do_settimeofday64 - Sets the time of day.
1220 * @ts: pointer to the timespec64 variable containing the new time
1222 * Sets the time of day to the new time and update NTP and notify hrtimers
1224 int do_settimeofday64(const struct timespec64
*ts
)
1226 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1227 struct timespec64 ts_delta
, xt
;
1228 unsigned long flags
;
1231 if (!timespec64_valid_settod(ts
))
1234 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1235 write_seqcount_begin(&tk_core
.seq
);
1237 timekeeping_forward_now(tk
);
1240 ts_delta
.tv_sec
= ts
->tv_sec
- xt
.tv_sec
;
1241 ts_delta
.tv_nsec
= ts
->tv_nsec
- xt
.tv_nsec
;
1243 if (timespec64_compare(&tk
->wall_to_monotonic
, &ts_delta
) > 0) {
1248 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, ts_delta
));
1250 tk_set_xtime(tk
, ts
);
1252 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1254 write_seqcount_end(&tk_core
.seq
);
1255 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1257 /* signal hrtimers about time change */
1261 audit_tk_injoffset(ts_delta
);
1265 EXPORT_SYMBOL(do_settimeofday64
);
1268 * timekeeping_inject_offset - Adds or subtracts from the current time.
1269 * @tv: pointer to the timespec variable containing the offset
1271 * Adds or subtracts an offset value from the current time.
1273 static int timekeeping_inject_offset(const struct timespec64
*ts
)
1275 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1276 unsigned long flags
;
1277 struct timespec64 tmp
;
1280 if (ts
->tv_nsec
< 0 || ts
->tv_nsec
>= NSEC_PER_SEC
)
1283 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1284 write_seqcount_begin(&tk_core
.seq
);
1286 timekeeping_forward_now(tk
);
1288 /* Make sure the proposed value is valid */
1289 tmp
= timespec64_add(tk_xtime(tk
), *ts
);
1290 if (timespec64_compare(&tk
->wall_to_monotonic
, ts
) > 0 ||
1291 !timespec64_valid_settod(&tmp
)) {
1296 tk_xtime_add(tk
, ts
);
1297 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *ts
));
1299 error
: /* even if we error out, we forwarded the time, so call update */
1300 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1302 write_seqcount_end(&tk_core
.seq
);
1303 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1305 /* signal hrtimers about time change */
1312 * Indicates if there is an offset between the system clock and the hardware
1313 * clock/persistent clock/rtc.
1315 int persistent_clock_is_local
;
1318 * Adjust the time obtained from the CMOS to be UTC time instead of
1321 * This is ugly, but preferable to the alternatives. Otherwise we
1322 * would either need to write a program to do it in /etc/rc (and risk
1323 * confusion if the program gets run more than once; it would also be
1324 * hard to make the program warp the clock precisely n hours) or
1325 * compile in the timezone information into the kernel. Bad, bad....
1329 * The best thing to do is to keep the CMOS clock in universal time (UTC)
1330 * as real UNIX machines always do it. This avoids all headaches about
1331 * daylight saving times and warping kernel clocks.
1333 void timekeeping_warp_clock(void)
1335 if (sys_tz
.tz_minuteswest
!= 0) {
1336 struct timespec64 adjust
;
1338 persistent_clock_is_local
= 1;
1339 adjust
.tv_sec
= sys_tz
.tz_minuteswest
* 60;
1341 timekeeping_inject_offset(&adjust
);
1346 * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
1349 static void __timekeeping_set_tai_offset(struct timekeeper
*tk
, s32 tai_offset
)
1351 tk
->tai_offset
= tai_offset
;
1352 tk
->offs_tai
= ktime_add(tk
->offs_real
, ktime_set(tai_offset
, 0));
1356 * change_clocksource - Swaps clocksources if a new one is available
1358 * Accumulates current time interval and initializes new clocksource
1360 static int change_clocksource(void *data
)
1362 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1363 struct clocksource
*new, *old
;
1364 unsigned long flags
;
1366 new = (struct clocksource
*) data
;
1368 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1369 write_seqcount_begin(&tk_core
.seq
);
1371 timekeeping_forward_now(tk
);
1373 * If the cs is in module, get a module reference. Succeeds
1374 * for built-in code (owner == NULL) as well.
1376 if (try_module_get(new->owner
)) {
1377 if (!new->enable
|| new->enable(new) == 0) {
1378 old
= tk
->tkr_mono
.clock
;
1379 tk_setup_internals(tk
, new);
1382 module_put(old
->owner
);
1384 module_put(new->owner
);
1387 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1389 write_seqcount_end(&tk_core
.seq
);
1390 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1396 * timekeeping_notify - Install a new clock source
1397 * @clock: pointer to the clock source
1399 * This function is called from clocksource.c after a new, better clock
1400 * source has been registered. The caller holds the clocksource_mutex.
1402 int timekeeping_notify(struct clocksource
*clock
)
1404 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1406 if (tk
->tkr_mono
.clock
== clock
)
1408 stop_machine(change_clocksource
, clock
, NULL
);
1409 tick_clock_notify();
1410 return tk
->tkr_mono
.clock
== clock
? 0 : -1;
1414 * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
1415 * @ts: pointer to the timespec64 to be set
1417 * Returns the raw monotonic time (completely un-modified by ntp)
1419 void ktime_get_raw_ts64(struct timespec64
*ts
)
1421 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1426 seq
= read_seqcount_begin(&tk_core
.seq
);
1427 ts
->tv_sec
= tk
->raw_sec
;
1428 nsecs
= timekeeping_get_ns(&tk
->tkr_raw
);
1430 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1433 timespec64_add_ns(ts
, nsecs
);
1435 EXPORT_SYMBOL(ktime_get_raw_ts64
);
1439 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
1441 int timekeeping_valid_for_hres(void)
1443 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1448 seq
= read_seqcount_begin(&tk_core
.seq
);
1450 ret
= tk
->tkr_mono
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
1452 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1458 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
1460 u64
timekeeping_max_deferment(void)
1462 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1467 seq
= read_seqcount_begin(&tk_core
.seq
);
1469 ret
= tk
->tkr_mono
.clock
->max_idle_ns
;
1471 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
1477 * read_persistent_clock64 - Return time from the persistent clock.
1479 * Weak dummy function for arches that do not yet support it.
1480 * Reads the time from the battery backed persistent clock.
1481 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
1483 * XXX - Do be sure to remove it once all arches implement it.
1485 void __weak
read_persistent_clock64(struct timespec64
*ts
)
1492 * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
1495 * Weak dummy function for arches that do not yet support it.
1496 * wall_time - current time as returned by persistent clock
1497 * boot_offset - offset that is defined as wall_time - boot_time
1498 * The default function calculates offset based on the current value of
1499 * local_clock(). This way architectures that support sched_clock() but don't
1500 * support dedicated boot time clock will provide the best estimate of the
1504 read_persistent_wall_and_boot_offset(struct timespec64
*wall_time
,
1505 struct timespec64
*boot_offset
)
1507 read_persistent_clock64(wall_time
);
1508 *boot_offset
= ns_to_timespec64(local_clock());
1512 * Flag reflecting whether timekeeping_resume() has injected sleeptime.
1514 * The flag starts of false and is only set when a suspend reaches
1515 * timekeeping_suspend(), timekeeping_resume() sets it to false when the
1516 * timekeeper clocksource is not stopping across suspend and has been
1517 * used to update sleep time. If the timekeeper clocksource has stopped
1518 * then the flag stays true and is used by the RTC resume code to decide
1519 * whether sleeptime must be injected and if so the flag gets false then.
1521 * If a suspend fails before reaching timekeeping_resume() then the flag
1522 * stays false and prevents erroneous sleeptime injection.
1524 static bool suspend_timing_needed
;
1526 /* Flag for if there is a persistent clock on this platform */
1527 static bool persistent_clock_exists
;
1530 * timekeeping_init - Initializes the clocksource and common timekeeping values
1532 void __init
timekeeping_init(void)
1534 struct timespec64 wall_time
, boot_offset
, wall_to_mono
;
1535 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1536 struct clocksource
*clock
;
1537 unsigned long flags
;
1539 read_persistent_wall_and_boot_offset(&wall_time
, &boot_offset
);
1540 if (timespec64_valid_settod(&wall_time
) &&
1541 timespec64_to_ns(&wall_time
) > 0) {
1542 persistent_clock_exists
= true;
1543 } else if (timespec64_to_ns(&wall_time
) != 0) {
1544 pr_warn("Persistent clock returned invalid value");
1545 wall_time
= (struct timespec64
){0};
1548 if (timespec64_compare(&wall_time
, &boot_offset
) < 0)
1549 boot_offset
= (struct timespec64
){0};
1552 * We want set wall_to_mono, so the following is true:
1553 * wall time + wall_to_mono = boot time
1555 wall_to_mono
= timespec64_sub(boot_offset
, wall_time
);
1557 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1558 write_seqcount_begin(&tk_core
.seq
);
1561 clock
= clocksource_default_clock();
1563 clock
->enable(clock
);
1564 tk_setup_internals(tk
, clock
);
1566 tk_set_xtime(tk
, &wall_time
);
1569 tk_set_wall_to_mono(tk
, wall_to_mono
);
1571 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1573 write_seqcount_end(&tk_core
.seq
);
1574 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1577 /* time in seconds when suspend began for persistent clock */
1578 static struct timespec64 timekeeping_suspend_time
;
1581 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
1582 * @delta: pointer to a timespec delta value
1584 * Takes a timespec offset measuring a suspend interval and properly
1585 * adds the sleep offset to the timekeeping variables.
1587 static void __timekeeping_inject_sleeptime(struct timekeeper
*tk
,
1588 const struct timespec64
*delta
)
1590 if (!timespec64_valid_strict(delta
)) {
1591 printk_deferred(KERN_WARNING
1592 "__timekeeping_inject_sleeptime: Invalid "
1593 "sleep delta value!\n");
1596 tk_xtime_add(tk
, delta
);
1597 tk_set_wall_to_mono(tk
, timespec64_sub(tk
->wall_to_monotonic
, *delta
));
1598 tk_update_sleep_time(tk
, timespec64_to_ktime(*delta
));
1599 tk_debug_account_sleep_time(delta
);
1602 #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
1604 * We have three kinds of time sources to use for sleep time
1605 * injection, the preference order is:
1606 * 1) non-stop clocksource
1607 * 2) persistent clock (ie: RTC accessible when irqs are off)
1610 * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
1611 * If system has neither 1) nor 2), 3) will be used finally.
1614 * If timekeeping has injected sleeptime via either 1) or 2),
1615 * 3) becomes needless, so in this case we don't need to call
1616 * rtc_resume(), and this is what timekeeping_rtc_skipresume()
1619 bool timekeeping_rtc_skipresume(void)
1621 return !suspend_timing_needed
;
1625 * 1) can be determined whether to use or not only when doing
1626 * timekeeping_resume() which is invoked after rtc_suspend(),
1627 * so we can't skip rtc_suspend() surely if system has 1).
1629 * But if system has 2), 2) will definitely be used, so in this
1630 * case we don't need to call rtc_suspend(), and this is what
1631 * timekeeping_rtc_skipsuspend() means.
1633 bool timekeeping_rtc_skipsuspend(void)
1635 return persistent_clock_exists
;
1639 * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
1640 * @delta: pointer to a timespec64 delta value
1642 * This hook is for architectures that cannot support read_persistent_clock64
1643 * because their RTC/persistent clock is only accessible when irqs are enabled.
1644 * and also don't have an effective nonstop clocksource.
1646 * This function should only be called by rtc_resume(), and allows
1647 * a suspend offset to be injected into the timekeeping values.
1649 void timekeeping_inject_sleeptime64(const struct timespec64
*delta
)
1651 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1652 unsigned long flags
;
1654 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1655 write_seqcount_begin(&tk_core
.seq
);
1657 suspend_timing_needed
= false;
1659 timekeeping_forward_now(tk
);
1661 __timekeeping_inject_sleeptime(tk
, delta
);
1663 timekeeping_update(tk
, TK_CLEAR_NTP
| TK_MIRROR
| TK_CLOCK_WAS_SET
);
1665 write_seqcount_end(&tk_core
.seq
);
1666 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1668 /* signal hrtimers about time change */
1674 * timekeeping_resume - Resumes the generic timekeeping subsystem.
1676 void timekeeping_resume(void)
1678 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1679 struct clocksource
*clock
= tk
->tkr_mono
.clock
;
1680 unsigned long flags
;
1681 struct timespec64 ts_new
, ts_delta
;
1682 u64 cycle_now
, nsec
;
1683 bool inject_sleeptime
= false;
1685 read_persistent_clock64(&ts_new
);
1687 clockevents_resume();
1688 clocksource_resume();
1690 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1691 write_seqcount_begin(&tk_core
.seq
);
1694 * After system resumes, we need to calculate the suspended time and
1695 * compensate it for the OS time. There are 3 sources that could be
1696 * used: Nonstop clocksource during suspend, persistent clock and rtc
1699 * One specific platform may have 1 or 2 or all of them, and the
1700 * preference will be:
1701 * suspend-nonstop clocksource -> persistent clock -> rtc
1702 * The less preferred source will only be tried if there is no better
1703 * usable source. The rtc part is handled separately in rtc core code.
1705 cycle_now
= tk_clock_read(&tk
->tkr_mono
);
1706 nsec
= clocksource_stop_suspend_timing(clock
, cycle_now
);
1708 ts_delta
= ns_to_timespec64(nsec
);
1709 inject_sleeptime
= true;
1710 } else if (timespec64_compare(&ts_new
, &timekeeping_suspend_time
) > 0) {
1711 ts_delta
= timespec64_sub(ts_new
, timekeeping_suspend_time
);
1712 inject_sleeptime
= true;
1715 if (inject_sleeptime
) {
1716 suspend_timing_needed
= false;
1717 __timekeeping_inject_sleeptime(tk
, &ts_delta
);
1720 /* Re-base the last cycle value */
1721 tk
->tkr_mono
.cycle_last
= cycle_now
;
1722 tk
->tkr_raw
.cycle_last
= cycle_now
;
1725 timekeeping_suspended
= 0;
1726 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
1727 write_seqcount_end(&tk_core
.seq
);
1728 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1730 touch_softlockup_watchdog();
1736 int timekeeping_suspend(void)
1738 struct timekeeper
*tk
= &tk_core
.timekeeper
;
1739 unsigned long flags
;
1740 struct timespec64 delta
, delta_delta
;
1741 static struct timespec64 old_delta
;
1742 struct clocksource
*curr_clock
;
1745 read_persistent_clock64(&timekeeping_suspend_time
);
1748 * On some systems the persistent_clock can not be detected at
1749 * timekeeping_init by its return value, so if we see a valid
1750 * value returned, update the persistent_clock_exists flag.
1752 if (timekeeping_suspend_time
.tv_sec
|| timekeeping_suspend_time
.tv_nsec
)
1753 persistent_clock_exists
= true;
1755 suspend_timing_needed
= true;
1757 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
1758 write_seqcount_begin(&tk_core
.seq
);
1759 timekeeping_forward_now(tk
);
1760 timekeeping_suspended
= 1;
1763 * Since we've called forward_now, cycle_last stores the value
1764 * just read from the current clocksource. Save this to potentially
1765 * use in suspend timing.
1767 curr_clock
= tk
->tkr_mono
.clock
;
1768 cycle_now
= tk
->tkr_mono
.cycle_last
;
1769 clocksource_start_suspend_timing(curr_clock
, cycle_now
);
1771 if (persistent_clock_exists
) {
1773 * To avoid drift caused by repeated suspend/resumes,
1774 * which each can add ~1 second drift error,
1775 * try to compensate so the difference in system time
1776 * and persistent_clock time stays close to constant.
1778 delta
= timespec64_sub(tk_xtime(tk
), timekeeping_suspend_time
);
1779 delta_delta
= timespec64_sub(delta
, old_delta
);
1780 if (abs(delta_delta
.tv_sec
) >= 2) {
1782 * if delta_delta is too large, assume time correction
1783 * has occurred and set old_delta to the current delta.
1787 /* Otherwise try to adjust old_system to compensate */
1788 timekeeping_suspend_time
=
1789 timespec64_add(timekeeping_suspend_time
, delta_delta
);
1793 timekeeping_update(tk
, TK_MIRROR
);
1794 halt_fast_timekeeper(tk
);
1795 write_seqcount_end(&tk_core
.seq
);
1796 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
1799 clocksource_suspend();
1800 clockevents_suspend();
1805 /* sysfs resume/suspend bits for timekeeping */
1806 static struct syscore_ops timekeeping_syscore_ops
= {
1807 .resume
= timekeeping_resume
,
1808 .suspend
= timekeeping_suspend
,
1811 static int __init
timekeeping_init_ops(void)
1813 register_syscore_ops(&timekeeping_syscore_ops
);
1816 device_initcall(timekeeping_init_ops
);
1819 * Apply a multiplier adjustment to the timekeeper
1821 static __always_inline
void timekeeping_apply_adjustment(struct timekeeper
*tk
,
1825 s64 interval
= tk
->cycle_interval
;
1827 if (mult_adj
== 0) {
1829 } else if (mult_adj
== -1) {
1830 interval
= -interval
;
1832 } else if (mult_adj
!= 1) {
1833 interval
*= mult_adj
;
1838 * So the following can be confusing.
1840 * To keep things simple, lets assume mult_adj == 1 for now.
1842 * When mult_adj != 1, remember that the interval and offset values
1843 * have been appropriately scaled so the math is the same.
1845 * The basic idea here is that we're increasing the multiplier
1846 * by one, this causes the xtime_interval to be incremented by
1847 * one cycle_interval. This is because:
1848 * xtime_interval = cycle_interval * mult
1849 * So if mult is being incremented by one:
1850 * xtime_interval = cycle_interval * (mult + 1)
1852 * xtime_interval = (cycle_interval * mult) + cycle_interval
1853 * Which can be shortened to:
1854 * xtime_interval += cycle_interval
1856 * So offset stores the non-accumulated cycles. Thus the current
1857 * time (in shifted nanoseconds) is:
1858 * now = (offset * adj) + xtime_nsec
1859 * Now, even though we're adjusting the clock frequency, we have
1860 * to keep time consistent. In other words, we can't jump back
1861 * in time, and we also want to avoid jumping forward in time.
1863 * So given the same offset value, we need the time to be the same
1864 * both before and after the freq adjustment.
1865 * now = (offset * adj_1) + xtime_nsec_1
1866 * now = (offset * adj_2) + xtime_nsec_2
1868 * (offset * adj_1) + xtime_nsec_1 =
1869 * (offset * adj_2) + xtime_nsec_2
1873 * (offset * adj_1) + xtime_nsec_1 =
1874 * (offset * (adj_1+1)) + xtime_nsec_2
1875 * (offset * adj_1) + xtime_nsec_1 =
1876 * (offset * adj_1) + offset + xtime_nsec_2
1877 * Canceling the sides:
1878 * xtime_nsec_1 = offset + xtime_nsec_2
1880 * xtime_nsec_2 = xtime_nsec_1 - offset
1881 * Which simplfies to:
1882 * xtime_nsec -= offset
1884 if ((mult_adj
> 0) && (tk
->tkr_mono
.mult
+ mult_adj
< mult_adj
)) {
1885 /* NTP adjustment caused clocksource mult overflow */
1890 tk
->tkr_mono
.mult
+= mult_adj
;
1891 tk
->xtime_interval
+= interval
;
1892 tk
->tkr_mono
.xtime_nsec
-= offset
;
1896 * Adjust the timekeeper's multiplier to the correct frequency
1897 * and also to reduce the accumulated error value.
1899 static void timekeeping_adjust(struct timekeeper
*tk
, s64 offset
)
1904 * Determine the multiplier from the current NTP tick length.
1905 * Avoid expensive division when the tick length doesn't change.
1907 if (likely(tk
->ntp_tick
== ntp_tick_length())) {
1908 mult
= tk
->tkr_mono
.mult
- tk
->ntp_err_mult
;
1910 tk
->ntp_tick
= ntp_tick_length();
1911 mult
= div64_u64((tk
->ntp_tick
>> tk
->ntp_error_shift
) -
1912 tk
->xtime_remainder
, tk
->cycle_interval
);
1916 * If the clock is behind the NTP time, increase the multiplier by 1
1917 * to catch up with it. If it's ahead and there was a remainder in the
1918 * tick division, the clock will slow down. Otherwise it will stay
1919 * ahead until the tick length changes to a non-divisible value.
1921 tk
->ntp_err_mult
= tk
->ntp_error
> 0 ? 1 : 0;
1922 mult
+= tk
->ntp_err_mult
;
1924 timekeeping_apply_adjustment(tk
, offset
, mult
- tk
->tkr_mono
.mult
);
1926 if (unlikely(tk
->tkr_mono
.clock
->maxadj
&&
1927 (abs(tk
->tkr_mono
.mult
- tk
->tkr_mono
.clock
->mult
)
1928 > tk
->tkr_mono
.clock
->maxadj
))) {
1929 printk_once(KERN_WARNING
1930 "Adjusting %s more than 11%% (%ld vs %ld)\n",
1931 tk
->tkr_mono
.clock
->name
, (long)tk
->tkr_mono
.mult
,
1932 (long)tk
->tkr_mono
.clock
->mult
+ tk
->tkr_mono
.clock
->maxadj
);
1936 * It may be possible that when we entered this function, xtime_nsec
1937 * was very small. Further, if we're slightly speeding the clocksource
1938 * in the code above, its possible the required corrective factor to
1939 * xtime_nsec could cause it to underflow.
1941 * Now, since we have already accumulated the second and the NTP
1942 * subsystem has been notified via second_overflow(), we need to skip
1945 if (unlikely((s64
)tk
->tkr_mono
.xtime_nsec
< 0)) {
1946 tk
->tkr_mono
.xtime_nsec
+= (u64
)NSEC_PER_SEC
<<
1949 tk
->skip_second_overflow
= 1;
1954 * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1956 * Helper function that accumulates the nsecs greater than a second
1957 * from the xtime_nsec field to the xtime_secs field.
1958 * It also calls into the NTP code to handle leapsecond processing.
1961 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper
*tk
)
1963 u64 nsecps
= (u64
)NSEC_PER_SEC
<< tk
->tkr_mono
.shift
;
1964 unsigned int clock_set
= 0;
1966 while (tk
->tkr_mono
.xtime_nsec
>= nsecps
) {
1969 tk
->tkr_mono
.xtime_nsec
-= nsecps
;
1973 * Skip NTP update if this second was accumulated before,
1974 * i.e. xtime_nsec underflowed in timekeeping_adjust()
1976 if (unlikely(tk
->skip_second_overflow
)) {
1977 tk
->skip_second_overflow
= 0;
1981 /* Figure out if its a leap sec and apply if needed */
1982 leap
= second_overflow(tk
->xtime_sec
);
1983 if (unlikely(leap
)) {
1984 struct timespec64 ts
;
1986 tk
->xtime_sec
+= leap
;
1990 tk_set_wall_to_mono(tk
,
1991 timespec64_sub(tk
->wall_to_monotonic
, ts
));
1993 __timekeeping_set_tai_offset(tk
, tk
->tai_offset
- leap
);
1995 clock_set
= TK_CLOCK_WAS_SET
;
2002 * logarithmic_accumulation - shifted accumulation of cycles
2004 * This functions accumulates a shifted interval of cycles into
2005 * into a shifted interval nanoseconds. Allows for O(log) accumulation
2008 * Returns the unconsumed cycles.
2010 static u64
logarithmic_accumulation(struct timekeeper
*tk
, u64 offset
,
2011 u32 shift
, unsigned int *clock_set
)
2013 u64 interval
= tk
->cycle_interval
<< shift
;
2016 /* If the offset is smaller than a shifted interval, do nothing */
2017 if (offset
< interval
)
2020 /* Accumulate one shifted interval */
2022 tk
->tkr_mono
.cycle_last
+= interval
;
2023 tk
->tkr_raw
.cycle_last
+= interval
;
2025 tk
->tkr_mono
.xtime_nsec
+= tk
->xtime_interval
<< shift
;
2026 *clock_set
|= accumulate_nsecs_to_secs(tk
);
2028 /* Accumulate raw time */
2029 tk
->tkr_raw
.xtime_nsec
+= tk
->raw_interval
<< shift
;
2030 snsec_per_sec
= (u64
)NSEC_PER_SEC
<< tk
->tkr_raw
.shift
;
2031 while (tk
->tkr_raw
.xtime_nsec
>= snsec_per_sec
) {
2032 tk
->tkr_raw
.xtime_nsec
-= snsec_per_sec
;
2036 /* Accumulate error between NTP and clock interval */
2037 tk
->ntp_error
+= tk
->ntp_tick
<< shift
;
2038 tk
->ntp_error
-= (tk
->xtime_interval
+ tk
->xtime_remainder
) <<
2039 (tk
->ntp_error_shift
+ shift
);
2045 * timekeeping_advance - Updates the timekeeper to the current time and
2046 * current NTP tick length
2048 static void timekeeping_advance(enum timekeeping_adv_mode mode
)
2050 struct timekeeper
*real_tk
= &tk_core
.timekeeper
;
2051 struct timekeeper
*tk
= &shadow_timekeeper
;
2053 int shift
= 0, maxshift
;
2054 unsigned int clock_set
= 0;
2055 unsigned long flags
;
2057 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2059 /* Make sure we're fully resumed: */
2060 if (unlikely(timekeeping_suspended
))
2063 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
2064 offset
= real_tk
->cycle_interval
;
2066 if (mode
!= TK_ADV_TICK
)
2069 offset
= clocksource_delta(tk_clock_read(&tk
->tkr_mono
),
2070 tk
->tkr_mono
.cycle_last
, tk
->tkr_mono
.mask
);
2072 /* Check if there's really nothing to do */
2073 if (offset
< real_tk
->cycle_interval
&& mode
== TK_ADV_TICK
)
2077 /* Do some additional sanity checking */
2078 timekeeping_check_update(tk
, offset
);
2081 * With NO_HZ we may have to accumulate many cycle_intervals
2082 * (think "ticks") worth of time at once. To do this efficiently,
2083 * we calculate the largest doubling multiple of cycle_intervals
2084 * that is smaller than the offset. We then accumulate that
2085 * chunk in one go, and then try to consume the next smaller
2088 shift
= ilog2(offset
) - ilog2(tk
->cycle_interval
);
2089 shift
= max(0, shift
);
2090 /* Bound shift to one less than what overflows tick_length */
2091 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
2092 shift
= min(shift
, maxshift
);
2093 while (offset
>= tk
->cycle_interval
) {
2094 offset
= logarithmic_accumulation(tk
, offset
, shift
,
2096 if (offset
< tk
->cycle_interval
<<shift
)
2100 /* Adjust the multiplier to correct NTP error */
2101 timekeeping_adjust(tk
, offset
);
2104 * Finally, make sure that after the rounding
2105 * xtime_nsec isn't larger than NSEC_PER_SEC
2107 clock_set
|= accumulate_nsecs_to_secs(tk
);
2109 write_seqcount_begin(&tk_core
.seq
);
2111 * Update the real timekeeper.
2113 * We could avoid this memcpy by switching pointers, but that
2114 * requires changes to all other timekeeper usage sites as
2115 * well, i.e. move the timekeeper pointer getter into the
2116 * spinlocked/seqcount protected sections. And we trade this
2117 * memcpy under the tk_core.seq against one before we start
2120 timekeeping_update(tk
, clock_set
);
2121 memcpy(real_tk
, tk
, sizeof(*tk
));
2122 /* The memcpy must come last. Do not put anything here! */
2123 write_seqcount_end(&tk_core
.seq
);
2125 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2127 /* Have to call _delayed version, since in irq context*/
2128 clock_was_set_delayed();
2132 * update_wall_time - Uses the current clocksource to increment the wall time
2135 void update_wall_time(void)
2137 timekeeping_advance(TK_ADV_TICK
);
2141 * getboottime64 - Return the real time of system boot.
2142 * @ts: pointer to the timespec64 to be set
2144 * Returns the wall-time of boot in a timespec64.
2146 * This is based on the wall_to_monotonic offset and the total suspend
2147 * time. Calls to settimeofday will affect the value returned (which
2148 * basically means that however wrong your real time clock is at boot time,
2149 * you get the right time here).
2151 void getboottime64(struct timespec64
*ts
)
2153 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2154 ktime_t t
= ktime_sub(tk
->offs_real
, tk
->offs_boot
);
2156 *ts
= ktime_to_timespec64(t
);
2158 EXPORT_SYMBOL_GPL(getboottime64
);
2160 void ktime_get_coarse_real_ts64(struct timespec64
*ts
)
2162 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2166 seq
= read_seqcount_begin(&tk_core
.seq
);
2169 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2171 EXPORT_SYMBOL(ktime_get_coarse_real_ts64
);
2173 void ktime_get_coarse_ts64(struct timespec64
*ts
)
2175 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2176 struct timespec64 now
, mono
;
2180 seq
= read_seqcount_begin(&tk_core
.seq
);
2183 mono
= tk
->wall_to_monotonic
;
2184 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2186 set_normalized_timespec64(ts
, now
.tv_sec
+ mono
.tv_sec
,
2187 now
.tv_nsec
+ mono
.tv_nsec
);
2189 EXPORT_SYMBOL(ktime_get_coarse_ts64
);
2192 * Must hold jiffies_lock
2194 void do_timer(unsigned long ticks
)
2196 jiffies_64
+= ticks
;
2197 calc_global_load(ticks
);
2201 * ktime_get_update_offsets_now - hrtimer helper
2202 * @cwsseq: pointer to check and store the clock was set sequence number
2203 * @offs_real: pointer to storage for monotonic -> realtime offset
2204 * @offs_boot: pointer to storage for monotonic -> boottime offset
2205 * @offs_tai: pointer to storage for monotonic -> clock tai offset
2207 * Returns current monotonic time and updates the offsets if the
2208 * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
2211 * Called from hrtimer_interrupt() or retrigger_next_event()
2213 ktime_t
ktime_get_update_offsets_now(unsigned int *cwsseq
, ktime_t
*offs_real
,
2214 ktime_t
*offs_boot
, ktime_t
*offs_tai
)
2216 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2222 seq
= read_seqcount_begin(&tk_core
.seq
);
2224 base
= tk
->tkr_mono
.base
;
2225 nsecs
= timekeeping_get_ns(&tk
->tkr_mono
);
2226 base
= ktime_add_ns(base
, nsecs
);
2228 if (*cwsseq
!= tk
->clock_was_set_seq
) {
2229 *cwsseq
= tk
->clock_was_set_seq
;
2230 *offs_real
= tk
->offs_real
;
2231 *offs_boot
= tk
->offs_boot
;
2232 *offs_tai
= tk
->offs_tai
;
2235 /* Handle leapsecond insertion adjustments */
2236 if (unlikely(base
>= tk
->next_leap_ktime
))
2237 *offs_real
= ktime_sub(tk
->offs_real
, ktime_set(1, 0));
2239 } while (read_seqcount_retry(&tk_core
.seq
, seq
));
2245 * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
2247 static int timekeeping_validate_timex(const struct __kernel_timex
*txc
)
2249 if (txc
->modes
& ADJ_ADJTIME
) {
2250 /* singleshot must not be used with any other mode bits */
2251 if (!(txc
->modes
& ADJ_OFFSET_SINGLESHOT
))
2253 if (!(txc
->modes
& ADJ_OFFSET_READONLY
) &&
2254 !capable(CAP_SYS_TIME
))
2257 /* In order to modify anything, you gotta be super-user! */
2258 if (txc
->modes
&& !capable(CAP_SYS_TIME
))
2261 * if the quartz is off by more than 10% then
2262 * something is VERY wrong!
2264 if (txc
->modes
& ADJ_TICK
&&
2265 (txc
->tick
< 900000/USER_HZ
||
2266 txc
->tick
> 1100000/USER_HZ
))
2270 if (txc
->modes
& ADJ_SETOFFSET
) {
2271 /* In order to inject time, you gotta be super-user! */
2272 if (!capable(CAP_SYS_TIME
))
2276 * Validate if a timespec/timeval used to inject a time
2277 * offset is valid. Offsets can be postive or negative, so
2278 * we don't check tv_sec. The value of the timeval/timespec
2279 * is the sum of its fields,but *NOTE*:
2280 * The field tv_usec/tv_nsec must always be non-negative and
2281 * we can't have more nanoseconds/microseconds than a second.
2283 if (txc
->time
.tv_usec
< 0)
2286 if (txc
->modes
& ADJ_NANO
) {
2287 if (txc
->time
.tv_usec
>= NSEC_PER_SEC
)
2290 if (txc
->time
.tv_usec
>= USEC_PER_SEC
)
2296 * Check for potential multiplication overflows that can
2297 * only happen on 64-bit systems:
2299 if ((txc
->modes
& ADJ_FREQUENCY
) && (BITS_PER_LONG
== 64)) {
2300 if (LLONG_MIN
/ PPM_SCALE
> txc
->freq
)
2302 if (LLONG_MAX
/ PPM_SCALE
< txc
->freq
)
2311 * do_adjtimex() - Accessor function to NTP __do_adjtimex function
2313 int do_adjtimex(struct __kernel_timex
*txc
)
2315 struct timekeeper
*tk
= &tk_core
.timekeeper
;
2316 struct audit_ntp_data ad
;
2317 unsigned long flags
;
2318 struct timespec64 ts
;
2322 /* Validate the data before disabling interrupts */
2323 ret
= timekeeping_validate_timex(txc
);
2327 if (txc
->modes
& ADJ_SETOFFSET
) {
2328 struct timespec64 delta
;
2329 delta
.tv_sec
= txc
->time
.tv_sec
;
2330 delta
.tv_nsec
= txc
->time
.tv_usec
;
2331 if (!(txc
->modes
& ADJ_NANO
))
2332 delta
.tv_nsec
*= 1000;
2333 ret
= timekeeping_inject_offset(&delta
);
2337 audit_tk_injoffset(delta
);
2340 audit_ntp_init(&ad
);
2342 ktime_get_real_ts64(&ts
);
2344 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2345 write_seqcount_begin(&tk_core
.seq
);
2347 orig_tai
= tai
= tk
->tai_offset
;
2348 ret
= __do_adjtimex(txc
, &ts
, &tai
, &ad
);
2350 if (tai
!= orig_tai
) {
2351 __timekeeping_set_tai_offset(tk
, tai
);
2352 timekeeping_update(tk
, TK_MIRROR
| TK_CLOCK_WAS_SET
);
2354 tk_update_leap_state(tk
);
2356 write_seqcount_end(&tk_core
.seq
);
2357 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2361 /* Update the multiplier immediately if frequency was set directly */
2362 if (txc
->modes
& (ADJ_FREQUENCY
| ADJ_TICK
))
2363 timekeeping_advance(TK_ADV_FREQ
);
2365 if (tai
!= orig_tai
)
2368 ntp_notify_cmos_timer();
2373 #ifdef CONFIG_NTP_PPS
2375 * hardpps() - Accessor function to NTP __hardpps function
2377 void hardpps(const struct timespec64
*phase_ts
, const struct timespec64
*raw_ts
)
2379 unsigned long flags
;
2381 raw_spin_lock_irqsave(&timekeeper_lock
, flags
);
2382 write_seqcount_begin(&tk_core
.seq
);
2384 __hardpps(phase_ts
, raw_ts
);
2386 write_seqcount_end(&tk_core
.seq
);
2387 raw_spin_unlock_irqrestore(&timekeeper_lock
, flags
);
2389 EXPORT_SYMBOL(hardpps
);
2390 #endif /* CONFIG_NTP_PPS */
2393 * xtime_update() - advances the timekeeping infrastructure
2394 * @ticks: number of ticks, that have elapsed since the last call.
2396 * Must be called with interrupts disabled.
2398 void xtime_update(unsigned long ticks
)
2400 write_seqlock(&jiffies_lock
);
2402 write_sequnlock(&jiffies_lock
);