2 * linux/kernel/time/timekeeping.c
4 * Kernel timekeeping code and accessor functions
6 * This code was moved from linux/kernel/timer.c.
7 * Please see that file for copyright and history logs.
11 #include <linux/module.h>
12 #include <linux/interrupt.h>
13 #include <linux/percpu.h>
14 #include <linux/init.h>
16 #include <linux/sched.h>
17 #include <linux/syscore_ops.h>
18 #include <linux/clocksource.h>
19 #include <linux/jiffies.h>
20 #include <linux/time.h>
21 #include <linux/tick.h>
22 #include <linux/stop_machine.h>
24 /* Structure holding internal timekeeping values. */
26 /* Current clocksource used for timekeeping. */
27 struct clocksource
*clock
;
28 /* The shift value of the current clocksource. */
31 /* Number of clock cycles in one NTP interval. */
32 cycle_t cycle_interval
;
33 /* Number of clock shifted nano seconds in one NTP interval. */
35 /* shifted nano seconds left over when rounding cycle_interval */
37 /* Raw nano seconds accumulated per NTP interval. */
40 /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
42 /* Difference between accumulated time and NTP time in ntp
43 * shifted nano seconds. */
45 /* Shift conversion between clock shifted nano seconds and
46 * ntp shifted nano seconds. */
48 /* NTP adjusted clock multiplier */
51 /* The current time */
52 struct timespec xtime
;
54 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
55 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
56 * at zero at system boot time, so wall_to_monotonic will be negative,
57 * however, we will ALWAYS keep the tv_nsec part positive so we can use
58 * the usual normalization.
60 * wall_to_monotonic is moved after resume from suspend for the
61 * monotonic time not to jump. We need to add total_sleep_time to
62 * wall_to_monotonic to get the real boot based time offset.
64 * - wall_to_monotonic is no longer the boot time, getboottime must be
67 struct timespec wall_to_monotonic
;
68 /* time spent in suspend */
69 struct timespec total_sleep_time
;
70 /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */
71 struct timespec raw_time
;
73 /* Seqlock for all timekeeper values */
77 static struct timekeeper timekeeper
;
80 * This read-write spinlock protects us from races in SMP while
83 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(xtime_lock
);
86 /* flag for if timekeeping is suspended */
87 int __read_mostly timekeeping_suspended
;
92 * timekeeper_setup_internals - Set up internals to use clocksource clock.
94 * @clock: Pointer to clocksource.
96 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
97 * pair and interval request.
99 * Unless you're the timekeeping code, you should not be using this!
101 static void timekeeper_setup_internals(struct clocksource
*clock
)
104 u64 tmp
, ntpinterval
;
106 timekeeper
.clock
= clock
;
107 clock
->cycle_last
= clock
->read(clock
);
109 /* Do the ns -> cycle conversion first, using original mult */
110 tmp
= NTP_INTERVAL_LENGTH
;
111 tmp
<<= clock
->shift
;
113 tmp
+= clock
->mult
/2;
114 do_div(tmp
, clock
->mult
);
118 interval
= (cycle_t
) tmp
;
119 timekeeper
.cycle_interval
= interval
;
121 /* Go back from cycles -> shifted ns */
122 timekeeper
.xtime_interval
= (u64
) interval
* clock
->mult
;
123 timekeeper
.xtime_remainder
= ntpinterval
- timekeeper
.xtime_interval
;
124 timekeeper
.raw_interval
=
125 ((u64
) interval
* clock
->mult
) >> clock
->shift
;
127 timekeeper
.xtime_nsec
= 0;
128 timekeeper
.shift
= clock
->shift
;
130 timekeeper
.ntp_error
= 0;
131 timekeeper
.ntp_error_shift
= NTP_SCALE_SHIFT
- clock
->shift
;
134 * The timekeeper keeps its own mult values for the currently
135 * active clocksource. These value will be adjusted via NTP
136 * to counteract clock drifting.
138 timekeeper
.mult
= clock
->mult
;
141 /* Timekeeper helper functions. */
142 static inline s64
timekeeping_get_ns(void)
144 cycle_t cycle_now
, cycle_delta
;
145 struct clocksource
*clock
;
147 /* read clocksource: */
148 clock
= timekeeper
.clock
;
149 cycle_now
= clock
->read(clock
);
151 /* calculate the delta since the last update_wall_time: */
152 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
154 /* return delta convert to nanoseconds using ntp adjusted mult. */
155 return clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
159 static inline s64
timekeeping_get_ns_raw(void)
161 cycle_t cycle_now
, cycle_delta
;
162 struct clocksource
*clock
;
164 /* read clocksource: */
165 clock
= timekeeper
.clock
;
166 cycle_now
= clock
->read(clock
);
168 /* calculate the delta since the last update_wall_time: */
169 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
171 /* return delta convert to nanoseconds. */
172 return clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
175 /* must hold xtime_lock */
176 void timekeeping_leap_insert(int leapsecond
)
180 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
182 timekeeper
.xtime
.tv_sec
+= leapsecond
;
183 timekeeper
.wall_to_monotonic
.tv_sec
-= leapsecond
;
184 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
185 timekeeper
.clock
, timekeeper
.mult
);
187 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
192 * timekeeping_forward_now - update clock to the current time
194 * Forward the current clock to update its state since the last call to
195 * update_wall_time(). This is useful before significant clock changes,
196 * as it avoids having to deal with this time offset explicitly.
198 static void timekeeping_forward_now(void)
200 cycle_t cycle_now
, cycle_delta
;
201 struct clocksource
*clock
;
204 clock
= timekeeper
.clock
;
205 cycle_now
= clock
->read(clock
);
206 cycle_delta
= (cycle_now
- clock
->cycle_last
) & clock
->mask
;
207 clock
->cycle_last
= cycle_now
;
209 nsec
= clocksource_cyc2ns(cycle_delta
, timekeeper
.mult
,
212 /* If arch requires, add in gettimeoffset() */
213 nsec
+= arch_gettimeoffset();
215 timespec_add_ns(&timekeeper
.xtime
, nsec
);
217 nsec
= clocksource_cyc2ns(cycle_delta
, clock
->mult
, clock
->shift
);
218 timespec_add_ns(&timekeeper
.raw_time
, nsec
);
222 * getnstimeofday - Returns the time of day in a timespec
223 * @ts: pointer to the timespec to be set
225 * Returns the time of day in a timespec.
227 void getnstimeofday(struct timespec
*ts
)
232 WARN_ON(timekeeping_suspended
);
235 seq
= read_seqbegin(&timekeeper
.lock
);
237 *ts
= timekeeper
.xtime
;
238 nsecs
= timekeeping_get_ns();
240 /* If arch requires, add in gettimeoffset() */
241 nsecs
+= arch_gettimeoffset();
243 } while (read_seqretry(&timekeeper
.lock
, seq
));
245 timespec_add_ns(ts
, nsecs
);
248 EXPORT_SYMBOL(getnstimeofday
);
250 ktime_t
ktime_get(void)
255 WARN_ON(timekeeping_suspended
);
258 seq
= read_seqbegin(&timekeeper
.lock
);
259 secs
= timekeeper
.xtime
.tv_sec
+
260 timekeeper
.wall_to_monotonic
.tv_sec
;
261 nsecs
= timekeeper
.xtime
.tv_nsec
+
262 timekeeper
.wall_to_monotonic
.tv_nsec
;
263 nsecs
+= timekeeping_get_ns();
264 /* If arch requires, add in gettimeoffset() */
265 nsecs
+= arch_gettimeoffset();
267 } while (read_seqretry(&timekeeper
.lock
, seq
));
269 * Use ktime_set/ktime_add_ns to create a proper ktime on
270 * 32-bit architectures without CONFIG_KTIME_SCALAR.
272 return ktime_add_ns(ktime_set(secs
, 0), nsecs
);
274 EXPORT_SYMBOL_GPL(ktime_get
);
277 * ktime_get_ts - get the monotonic clock in timespec format
278 * @ts: pointer to timespec variable
280 * The function calculates the monotonic clock from the realtime
281 * clock and the wall_to_monotonic offset and stores the result
282 * in normalized timespec format in the variable pointed to by @ts.
284 void ktime_get_ts(struct timespec
*ts
)
286 struct timespec tomono
;
290 WARN_ON(timekeeping_suspended
);
293 seq
= read_seqbegin(&timekeeper
.lock
);
294 *ts
= timekeeper
.xtime
;
295 tomono
= timekeeper
.wall_to_monotonic
;
296 nsecs
= timekeeping_get_ns();
297 /* If arch requires, add in gettimeoffset() */
298 nsecs
+= arch_gettimeoffset();
300 } while (read_seqretry(&timekeeper
.lock
, seq
));
302 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
,
303 ts
->tv_nsec
+ tomono
.tv_nsec
+ nsecs
);
305 EXPORT_SYMBOL_GPL(ktime_get_ts
);
307 #ifdef CONFIG_NTP_PPS
310 * getnstime_raw_and_real - get day and raw monotonic time in timespec format
311 * @ts_raw: pointer to the timespec to be set to raw monotonic time
312 * @ts_real: pointer to the timespec to be set to the time of day
314 * This function reads both the time of day and raw monotonic time at the
315 * same time atomically and stores the resulting timestamps in timespec
318 void getnstime_raw_and_real(struct timespec
*ts_raw
, struct timespec
*ts_real
)
321 s64 nsecs_raw
, nsecs_real
;
323 WARN_ON_ONCE(timekeeping_suspended
);
328 seq
= read_seqbegin(&timekeeper
.lock
);
330 *ts_raw
= timekeeper
.raw_time
;
331 *ts_real
= timekeeper
.xtime
;
333 nsecs_raw
= timekeeping_get_ns_raw();
334 nsecs_real
= timekeeping_get_ns();
336 /* If arch requires, add in gettimeoffset() */
337 arch_offset
= arch_gettimeoffset();
338 nsecs_raw
+= arch_offset
;
339 nsecs_real
+= arch_offset
;
341 } while (read_seqretry(&timekeeper
.lock
, seq
));
343 timespec_add_ns(ts_raw
, nsecs_raw
);
344 timespec_add_ns(ts_real
, nsecs_real
);
346 EXPORT_SYMBOL(getnstime_raw_and_real
);
348 #endif /* CONFIG_NTP_PPS */
351 * do_gettimeofday - Returns the time of day in a timeval
352 * @tv: pointer to the timeval to be set
354 * NOTE: Users should be converted to using getnstimeofday()
356 void do_gettimeofday(struct timeval
*tv
)
360 getnstimeofday(&now
);
361 tv
->tv_sec
= now
.tv_sec
;
362 tv
->tv_usec
= now
.tv_nsec
/1000;
365 EXPORT_SYMBOL(do_gettimeofday
);
367 * do_settimeofday - Sets the time of day
368 * @tv: pointer to the timespec variable containing the new time
370 * Sets the time of day to the new time and update NTP and notify hrtimers
372 int do_settimeofday(const struct timespec
*tv
)
374 struct timespec ts_delta
;
375 unsigned long flags1
,flags2
;
377 if ((unsigned long)tv
->tv_nsec
>= NSEC_PER_SEC
)
380 write_seqlock_irqsave(&xtime_lock
, flags1
);
381 write_seqlock_irqsave(&timekeeper
.lock
, flags2
);
383 timekeeping_forward_now();
385 ts_delta
.tv_sec
= tv
->tv_sec
- timekeeper
.xtime
.tv_sec
;
386 ts_delta
.tv_nsec
= tv
->tv_nsec
- timekeeper
.xtime
.tv_nsec
;
387 timekeeper
.wall_to_monotonic
=
388 timespec_sub(timekeeper
.wall_to_monotonic
, ts_delta
);
390 timekeeper
.xtime
= *tv
;
392 timekeeper
.ntp_error
= 0;
395 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
396 timekeeper
.clock
, timekeeper
.mult
);
398 write_sequnlock_irqrestore(&timekeeper
.lock
, flags2
);
399 write_sequnlock_irqrestore(&xtime_lock
, flags1
);
401 /* signal hrtimers about time change */
407 EXPORT_SYMBOL(do_settimeofday
);
411 * timekeeping_inject_offset - Adds or subtracts from the current time.
412 * @tv: pointer to the timespec variable containing the offset
414 * Adds or subtracts an offset value from the current time.
416 int timekeeping_inject_offset(struct timespec
*ts
)
418 unsigned long flags1
,flags2
;
420 if ((unsigned long)ts
->tv_nsec
>= NSEC_PER_SEC
)
423 write_seqlock_irqsave(&xtime_lock
, flags1
);
424 write_seqlock_irqsave(&timekeeper
.lock
, flags2
);
426 timekeeping_forward_now();
428 timekeeper
.xtime
= timespec_add(timekeeper
.xtime
, *ts
);
429 timekeeper
.wall_to_monotonic
=
430 timespec_sub(timekeeper
.wall_to_monotonic
, *ts
);
432 timekeeper
.ntp_error
= 0;
435 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
436 timekeeper
.clock
, timekeeper
.mult
);
438 write_sequnlock_irqrestore(&timekeeper
.lock
, flags2
);
439 write_sequnlock_irqrestore(&xtime_lock
, flags1
);
441 /* signal hrtimers about time change */
446 EXPORT_SYMBOL(timekeeping_inject_offset
);
449 * change_clocksource - Swaps clocksources if a new one is available
451 * Accumulates current time interval and initializes new clocksource
453 static int change_clocksource(void *data
)
455 struct clocksource
*new, *old
;
457 new = (struct clocksource
*) data
;
459 timekeeping_forward_now();
460 if (!new->enable
|| new->enable(new) == 0) {
461 old
= timekeeper
.clock
;
462 timekeeper_setup_internals(new);
470 * timekeeping_notify - Install a new clock source
471 * @clock: pointer to the clock source
473 * This function is called from clocksource.c after a new, better clock
474 * source has been registered. The caller holds the clocksource_mutex.
476 void timekeeping_notify(struct clocksource
*clock
)
478 if (timekeeper
.clock
== clock
)
480 stop_machine(change_clocksource
, clock
, NULL
);
485 * ktime_get_real - get the real (wall-) time in ktime_t format
487 * returns the time in ktime_t format
489 ktime_t
ktime_get_real(void)
493 getnstimeofday(&now
);
495 return timespec_to_ktime(now
);
497 EXPORT_SYMBOL_GPL(ktime_get_real
);
500 * getrawmonotonic - Returns the raw monotonic time in a timespec
501 * @ts: pointer to the timespec to be set
503 * Returns the raw monotonic time (completely un-modified by ntp)
505 void getrawmonotonic(struct timespec
*ts
)
511 seq
= read_seqbegin(&timekeeper
.lock
);
512 nsecs
= timekeeping_get_ns_raw();
513 *ts
= timekeeper
.raw_time
;
515 } while (read_seqretry(&timekeeper
.lock
, seq
));
517 timespec_add_ns(ts
, nsecs
);
519 EXPORT_SYMBOL(getrawmonotonic
);
523 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
525 int timekeeping_valid_for_hres(void)
531 seq
= read_seqbegin(&timekeeper
.lock
);
533 ret
= timekeeper
.clock
->flags
& CLOCK_SOURCE_VALID_FOR_HRES
;
535 } while (read_seqretry(&timekeeper
.lock
, seq
));
541 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
543 u64
timekeeping_max_deferment(void)
548 seq
= read_seqbegin(&timekeeper
.lock
);
550 ret
= timekeeper
.clock
->max_idle_ns
;
552 } while (read_seqretry(&timekeeper
.lock
, seq
));
558 * read_persistent_clock - Return time from the persistent clock.
560 * Weak dummy function for arches that do not yet support it.
561 * Reads the time from the battery backed persistent clock.
562 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
564 * XXX - Do be sure to remove it once all arches implement it.
566 void __attribute__((weak
)) read_persistent_clock(struct timespec
*ts
)
573 * read_boot_clock - Return time of the system start.
575 * Weak dummy function for arches that do not yet support it.
576 * Function to read the exact time the system has been started.
577 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
579 * XXX - Do be sure to remove it once all arches implement it.
581 void __attribute__((weak
)) read_boot_clock(struct timespec
*ts
)
588 * timekeeping_init - Initializes the clocksource and common timekeeping values
590 void __init
timekeeping_init(void)
592 struct clocksource
*clock
;
594 struct timespec now
, boot
;
596 read_persistent_clock(&now
);
597 read_boot_clock(&boot
);
599 seqlock_init(&timekeeper
.lock
);
601 write_seqlock_irqsave(&xtime_lock
, flags
);
603 write_sequnlock_irqrestore(&xtime_lock
, flags
);
605 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
606 clock
= clocksource_default_clock();
608 clock
->enable(clock
);
609 timekeeper_setup_internals(clock
);
611 timekeeper
.xtime
.tv_sec
= now
.tv_sec
;
612 timekeeper
.xtime
.tv_nsec
= now
.tv_nsec
;
613 timekeeper
.raw_time
.tv_sec
= 0;
614 timekeeper
.raw_time
.tv_nsec
= 0;
615 if (boot
.tv_sec
== 0 && boot
.tv_nsec
== 0) {
616 boot
.tv_sec
= timekeeper
.xtime
.tv_sec
;
617 boot
.tv_nsec
= timekeeper
.xtime
.tv_nsec
;
619 set_normalized_timespec(&timekeeper
.wall_to_monotonic
,
620 -boot
.tv_sec
, -boot
.tv_nsec
);
621 timekeeper
.total_sleep_time
.tv_sec
= 0;
622 timekeeper
.total_sleep_time
.tv_nsec
= 0;
623 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
626 /* time in seconds when suspend began */
627 static struct timespec timekeeping_suspend_time
;
630 * __timekeeping_inject_sleeptime - Internal function to add sleep interval
631 * @delta: pointer to a timespec delta value
633 * Takes a timespec offset measuring a suspend interval and properly
634 * adds the sleep offset to the timekeeping variables.
636 static void __timekeeping_inject_sleeptime(struct timespec
*delta
)
638 if (!timespec_valid(delta
)) {
639 printk(KERN_WARNING
"__timekeeping_inject_sleeptime: Invalid "
640 "sleep delta value!\n");
644 timekeeper
.xtime
= timespec_add(timekeeper
.xtime
, *delta
);
645 timekeeper
.wall_to_monotonic
=
646 timespec_sub(timekeeper
.wall_to_monotonic
, *delta
);
647 timekeeper
.total_sleep_time
= timespec_add(
648 timekeeper
.total_sleep_time
, *delta
);
653 * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
654 * @delta: pointer to a timespec delta value
656 * This hook is for architectures that cannot support read_persistent_clock
657 * because their RTC/persistent clock is only accessible when irqs are enabled.
659 * This function should only be called by rtc_resume(), and allows
660 * a suspend offset to be injected into the timekeeping values.
662 void timekeeping_inject_sleeptime(struct timespec
*delta
)
664 unsigned long flags1
,flags2
;
667 /* Make sure we don't set the clock twice */
668 read_persistent_clock(&ts
);
669 if (!(ts
.tv_sec
== 0 && ts
.tv_nsec
== 0))
672 write_seqlock_irqsave(&xtime_lock
, flags1
);
673 write_seqlock_irqsave(&timekeeper
.lock
, flags2
);
675 timekeeping_forward_now();
677 __timekeeping_inject_sleeptime(delta
);
679 timekeeper
.ntp_error
= 0;
681 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
682 timekeeper
.clock
, timekeeper
.mult
);
684 write_sequnlock_irqrestore(&timekeeper
.lock
, flags2
);
685 write_sequnlock_irqrestore(&xtime_lock
, flags1
);
687 /* signal hrtimers about time change */
693 * timekeeping_resume - Resumes the generic timekeeping subsystem.
695 * This is for the generic clocksource timekeeping.
696 * xtime/wall_to_monotonic/jiffies/etc are
697 * still managed by arch specific suspend/resume code.
699 static void timekeeping_resume(void)
701 unsigned long flags1
,flags2
;
704 read_persistent_clock(&ts
);
706 clocksource_resume();
708 write_seqlock_irqsave(&xtime_lock
, flags1
);
709 write_seqlock_irqsave(&timekeeper
.lock
, flags2
);
711 if (timespec_compare(&ts
, &timekeeping_suspend_time
) > 0) {
712 ts
= timespec_sub(ts
, timekeeping_suspend_time
);
713 __timekeeping_inject_sleeptime(&ts
);
715 /* re-base the last cycle value */
716 timekeeper
.clock
->cycle_last
= timekeeper
.clock
->read(timekeeper
.clock
);
717 timekeeper
.ntp_error
= 0;
718 timekeeping_suspended
= 0;
719 write_sequnlock_irqrestore(&timekeeper
.lock
, flags2
);
720 write_sequnlock_irqrestore(&xtime_lock
, flags1
);
722 touch_softlockup_watchdog();
724 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME
, NULL
);
726 /* Resume hrtimers */
730 static int timekeeping_suspend(void)
732 unsigned long flags1
,flags2
;
733 struct timespec delta
, delta_delta
;
734 static struct timespec old_delta
;
736 read_persistent_clock(&timekeeping_suspend_time
);
738 write_seqlock_irqsave(&xtime_lock
, flags1
);
739 write_seqlock_irqsave(&timekeeper
.lock
, flags2
);
740 timekeeping_forward_now();
741 timekeeping_suspended
= 1;
744 * To avoid drift caused by repeated suspend/resumes,
745 * which each can add ~1 second drift error,
746 * try to compensate so the difference in system time
747 * and persistent_clock time stays close to constant.
749 delta
= timespec_sub(timekeeper
.xtime
, timekeeping_suspend_time
);
750 delta_delta
= timespec_sub(delta
, old_delta
);
751 if (abs(delta_delta
.tv_sec
) >= 2) {
753 * if delta_delta is too large, assume time correction
754 * has occured and set old_delta to the current delta.
758 /* Otherwise try to adjust old_system to compensate */
759 timekeeping_suspend_time
=
760 timespec_add(timekeeping_suspend_time
, delta_delta
);
762 write_sequnlock_irqrestore(&timekeeper
.lock
, flags2
);
763 write_sequnlock_irqrestore(&xtime_lock
, flags1
);
765 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND
, NULL
);
766 clocksource_suspend();
771 /* sysfs resume/suspend bits for timekeeping */
772 static struct syscore_ops timekeeping_syscore_ops
= {
773 .resume
= timekeeping_resume
,
774 .suspend
= timekeeping_suspend
,
777 static int __init
timekeeping_init_ops(void)
779 register_syscore_ops(&timekeeping_syscore_ops
);
783 device_initcall(timekeeping_init_ops
);
786 * If the error is already larger, we look ahead even further
787 * to compensate for late or lost adjustments.
789 static __always_inline
int timekeeping_bigadjust(s64 error
, s64
*interval
,
797 * Use the current error value to determine how much to look ahead.
798 * The larger the error the slower we adjust for it to avoid problems
799 * with losing too many ticks, otherwise we would overadjust and
800 * produce an even larger error. The smaller the adjustment the
801 * faster we try to adjust for it, as lost ticks can do less harm
802 * here. This is tuned so that an error of about 1 msec is adjusted
803 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
805 error2
= timekeeper
.ntp_error
>> (NTP_SCALE_SHIFT
+ 22 - 2 * SHIFT_HZ
);
806 error2
= abs(error2
);
807 for (look_ahead
= 0; error2
> 0; look_ahead
++)
811 * Now calculate the error in (1 << look_ahead) ticks, but first
812 * remove the single look ahead already included in the error.
814 tick_error
= ntp_tick_length() >> (timekeeper
.ntp_error_shift
+ 1);
815 tick_error
-= timekeeper
.xtime_interval
>> 1;
816 error
= ((error
- tick_error
) >> look_ahead
) + tick_error
;
818 /* Finally calculate the adjustment shift value. */
823 *interval
= -*interval
;
827 for (adj
= 0; error
> i
; adj
++)
836 * Adjust the multiplier to reduce the error value,
837 * this is optimized for the most common adjustments of -1,0,1,
838 * for other values we can do a bit more work.
840 static void timekeeping_adjust(s64 offset
)
842 s64 error
, interval
= timekeeper
.cycle_interval
;
846 * The point of this is to check if the error is greater then half
849 * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
851 * Note we subtract one in the shift, so that error is really error*2.
852 * This "saves" dividing(shifting) interval twice, but keeps the
853 * (error > interval) comparison as still measuring if error is
854 * larger then half an interval.
856 * Note: It does not "save" on aggravation when reading the code.
858 error
= timekeeper
.ntp_error
>> (timekeeper
.ntp_error_shift
- 1);
859 if (error
> interval
) {
861 * We now divide error by 4(via shift), which checks if
862 * the error is greater then twice the interval.
863 * If it is greater, we need a bigadjust, if its smaller,
864 * we can adjust by 1.
868 * XXX - In update_wall_time, we round up to the next
869 * nanosecond, and store the amount rounded up into
870 * the error. This causes the likely below to be unlikely.
872 * The proper fix is to avoid rounding up by using
873 * the high precision timekeeper.xtime_nsec instead of
874 * xtime.tv_nsec everywhere. Fixing this will take some
877 if (likely(error
<= interval
))
880 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
881 } else if (error
< -interval
) {
882 /* See comment above, this is just switched for the negative */
884 if (likely(error
>= -interval
)) {
886 interval
= -interval
;
889 adj
= timekeeping_bigadjust(error
, &interval
, &offset
);
890 } else /* No adjustment needed */
893 WARN_ONCE(timekeeper
.clock
->maxadj
&&
894 (timekeeper
.mult
+ adj
> timekeeper
.clock
->mult
+
895 timekeeper
.clock
->maxadj
),
896 "Adjusting %s more then 11%% (%ld vs %ld)\n",
897 timekeeper
.clock
->name
, (long)timekeeper
.mult
+ adj
,
898 (long)timekeeper
.clock
->mult
+
899 timekeeper
.clock
->maxadj
);
901 * So the following can be confusing.
903 * To keep things simple, lets assume adj == 1 for now.
905 * When adj != 1, remember that the interval and offset values
906 * have been appropriately scaled so the math is the same.
908 * The basic idea here is that we're increasing the multiplier
909 * by one, this causes the xtime_interval to be incremented by
910 * one cycle_interval. This is because:
911 * xtime_interval = cycle_interval * mult
912 * So if mult is being incremented by one:
913 * xtime_interval = cycle_interval * (mult + 1)
915 * xtime_interval = (cycle_interval * mult) + cycle_interval
916 * Which can be shortened to:
917 * xtime_interval += cycle_interval
919 * So offset stores the non-accumulated cycles. Thus the current
920 * time (in shifted nanoseconds) is:
921 * now = (offset * adj) + xtime_nsec
922 * Now, even though we're adjusting the clock frequency, we have
923 * to keep time consistent. In other words, we can't jump back
924 * in time, and we also want to avoid jumping forward in time.
926 * So given the same offset value, we need the time to be the same
927 * both before and after the freq adjustment.
928 * now = (offset * adj_1) + xtime_nsec_1
929 * now = (offset * adj_2) + xtime_nsec_2
931 * (offset * adj_1) + xtime_nsec_1 =
932 * (offset * adj_2) + xtime_nsec_2
936 * (offset * adj_1) + xtime_nsec_1 =
937 * (offset * (adj_1+1)) + xtime_nsec_2
938 * (offset * adj_1) + xtime_nsec_1 =
939 * (offset * adj_1) + offset + xtime_nsec_2
940 * Canceling the sides:
941 * xtime_nsec_1 = offset + xtime_nsec_2
943 * xtime_nsec_2 = xtime_nsec_1 - offset
944 * Which simplfies to:
945 * xtime_nsec -= offset
947 * XXX - TODO: Doc ntp_error calculation.
949 timekeeper
.mult
+= adj
;
950 timekeeper
.xtime_interval
+= interval
;
951 timekeeper
.xtime_nsec
-= offset
;
952 timekeeper
.ntp_error
-= (interval
- offset
) <<
953 timekeeper
.ntp_error_shift
;
958 * logarithmic_accumulation - shifted accumulation of cycles
960 * This functions accumulates a shifted interval of cycles into
961 * into a shifted interval nanoseconds. Allows for O(log) accumulation
964 * Returns the unconsumed cycles.
966 static cycle_t
logarithmic_accumulation(cycle_t offset
, int shift
)
968 u64 nsecps
= (u64
)NSEC_PER_SEC
<< timekeeper
.shift
;
971 /* If the offset is smaller then a shifted interval, do nothing */
972 if (offset
< timekeeper
.cycle_interval
<<shift
)
975 /* Accumulate one shifted interval */
976 offset
-= timekeeper
.cycle_interval
<< shift
;
977 timekeeper
.clock
->cycle_last
+= timekeeper
.cycle_interval
<< shift
;
979 timekeeper
.xtime_nsec
+= timekeeper
.xtime_interval
<< shift
;
980 while (timekeeper
.xtime_nsec
>= nsecps
) {
981 timekeeper
.xtime_nsec
-= nsecps
;
982 timekeeper
.xtime
.tv_sec
++;
986 /* Accumulate raw time */
987 raw_nsecs
= timekeeper
.raw_interval
<< shift
;
988 raw_nsecs
+= timekeeper
.raw_time
.tv_nsec
;
989 if (raw_nsecs
>= NSEC_PER_SEC
) {
990 u64 raw_secs
= raw_nsecs
;
991 raw_nsecs
= do_div(raw_secs
, NSEC_PER_SEC
);
992 timekeeper
.raw_time
.tv_sec
+= raw_secs
;
994 timekeeper
.raw_time
.tv_nsec
= raw_nsecs
;
996 /* Accumulate error between NTP and clock interval */
997 timekeeper
.ntp_error
+= ntp_tick_length() << shift
;
998 timekeeper
.ntp_error
-=
999 (timekeeper
.xtime_interval
+ timekeeper
.xtime_remainder
) <<
1000 (timekeeper
.ntp_error_shift
+ shift
);
1007 * update_wall_time - Uses the current clocksource to increment the wall time
1009 * Called from the timer interrupt, must hold a write on xtime_lock.
1011 static void update_wall_time(void)
1013 struct clocksource
*clock
;
1015 int shift
= 0, maxshift
;
1016 unsigned long flags
;
1018 write_seqlock_irqsave(&timekeeper
.lock
, flags
);
1020 /* Make sure we're fully resumed: */
1021 if (unlikely(timekeeping_suspended
))
1024 clock
= timekeeper
.clock
;
1026 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1027 offset
= timekeeper
.cycle_interval
;
1029 offset
= (clock
->read(clock
) - clock
->cycle_last
) & clock
->mask
;
1031 timekeeper
.xtime_nsec
= (s64
)timekeeper
.xtime
.tv_nsec
<<
1035 * With NO_HZ we may have to accumulate many cycle_intervals
1036 * (think "ticks") worth of time at once. To do this efficiently,
1037 * we calculate the largest doubling multiple of cycle_intervals
1038 * that is smaller then the offset. We then accumulate that
1039 * chunk in one go, and then try to consume the next smaller
1042 shift
= ilog2(offset
) - ilog2(timekeeper
.cycle_interval
);
1043 shift
= max(0, shift
);
1044 /* Bound shift to one less then what overflows tick_length */
1045 maxshift
= (64 - (ilog2(ntp_tick_length())+1)) - 1;
1046 shift
= min(shift
, maxshift
);
1047 while (offset
>= timekeeper
.cycle_interval
) {
1048 offset
= logarithmic_accumulation(offset
, shift
);
1049 if(offset
< timekeeper
.cycle_interval
<<shift
)
1053 /* correct the clock when NTP error is too big */
1054 timekeeping_adjust(offset
);
1057 * Since in the loop above, we accumulate any amount of time
1058 * in xtime_nsec over a second into xtime.tv_sec, its possible for
1059 * xtime_nsec to be fairly small after the loop. Further, if we're
1060 * slightly speeding the clocksource up in timekeeping_adjust(),
1061 * its possible the required corrective factor to xtime_nsec could
1062 * cause it to underflow.
1064 * Now, we cannot simply roll the accumulated second back, since
1065 * the NTP subsystem has been notified via second_overflow. So
1066 * instead we push xtime_nsec forward by the amount we underflowed,
1067 * and add that amount into the error.
1069 * We'll correct this error next time through this function, when
1070 * xtime_nsec is not as small.
1072 if (unlikely((s64
)timekeeper
.xtime_nsec
< 0)) {
1073 s64 neg
= -(s64
)timekeeper
.xtime_nsec
;
1074 timekeeper
.xtime_nsec
= 0;
1075 timekeeper
.ntp_error
+= neg
<< timekeeper
.ntp_error_shift
;
1080 * Store full nanoseconds into xtime after rounding it up and
1081 * add the remainder to the error difference.
1083 timekeeper
.xtime
.tv_nsec
= ((s64
)timekeeper
.xtime_nsec
>>
1084 timekeeper
.shift
) + 1;
1085 timekeeper
.xtime_nsec
-= (s64
)timekeeper
.xtime
.tv_nsec
<<
1087 timekeeper
.ntp_error
+= timekeeper
.xtime_nsec
<<
1088 timekeeper
.ntp_error_shift
;
1091 * Finally, make sure that after the rounding
1092 * xtime.tv_nsec isn't larger then NSEC_PER_SEC
1094 if (unlikely(timekeeper
.xtime
.tv_nsec
>= NSEC_PER_SEC
)) {
1095 timekeeper
.xtime
.tv_nsec
-= NSEC_PER_SEC
;
1096 timekeeper
.xtime
.tv_sec
++;
1100 /* check to see if there is a new clocksource to use */
1101 update_vsyscall(&timekeeper
.xtime
, &timekeeper
.wall_to_monotonic
,
1102 timekeeper
.clock
, timekeeper
.mult
);
1105 write_sequnlock_irqrestore(&timekeeper
.lock
, flags
);
1110 * getboottime - Return the real time of system boot.
1111 * @ts: pointer to the timespec to be set
1113 * Returns the wall-time of boot in a timespec.
1115 * This is based on the wall_to_monotonic offset and the total suspend
1116 * time. Calls to settimeofday will affect the value returned (which
1117 * basically means that however wrong your real time clock is at boot time,
1118 * you get the right time here).
1120 void getboottime(struct timespec
*ts
)
1122 struct timespec boottime
= {
1123 .tv_sec
= timekeeper
.wall_to_monotonic
.tv_sec
+
1124 timekeeper
.total_sleep_time
.tv_sec
,
1125 .tv_nsec
= timekeeper
.wall_to_monotonic
.tv_nsec
+
1126 timekeeper
.total_sleep_time
.tv_nsec
1129 set_normalized_timespec(ts
, -boottime
.tv_sec
, -boottime
.tv_nsec
);
1131 EXPORT_SYMBOL_GPL(getboottime
);
1135 * get_monotonic_boottime - Returns monotonic time since boot
1136 * @ts: pointer to the timespec to be set
1138 * Returns the monotonic time since boot in a timespec.
1140 * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1141 * includes the time spent in suspend.
1143 void get_monotonic_boottime(struct timespec
*ts
)
1145 struct timespec tomono
, sleep
;
1149 WARN_ON(timekeeping_suspended
);
1152 seq
= read_seqbegin(&timekeeper
.lock
);
1153 *ts
= timekeeper
.xtime
;
1154 tomono
= timekeeper
.wall_to_monotonic
;
1155 sleep
= timekeeper
.total_sleep_time
;
1156 nsecs
= timekeeping_get_ns();
1158 } while (read_seqretry(&timekeeper
.lock
, seq
));
1160 set_normalized_timespec(ts
, ts
->tv_sec
+ tomono
.tv_sec
+ sleep
.tv_sec
,
1161 ts
->tv_nsec
+ tomono
.tv_nsec
+ sleep
.tv_nsec
+ nsecs
);
1163 EXPORT_SYMBOL_GPL(get_monotonic_boottime
);
1166 * ktime_get_boottime - Returns monotonic time since boot in a ktime
1168 * Returns the monotonic time since boot in a ktime
1170 * This is similar to CLOCK_MONTONIC/ktime_get, but also
1171 * includes the time spent in suspend.
1173 ktime_t
ktime_get_boottime(void)
1177 get_monotonic_boottime(&ts
);
1178 return timespec_to_ktime(ts
);
1180 EXPORT_SYMBOL_GPL(ktime_get_boottime
);
1183 * monotonic_to_bootbased - Convert the monotonic time to boot based.
1184 * @ts: pointer to the timespec to be converted
1186 void monotonic_to_bootbased(struct timespec
*ts
)
1188 *ts
= timespec_add(*ts
, timekeeper
.total_sleep_time
);
1190 EXPORT_SYMBOL_GPL(monotonic_to_bootbased
);
1192 unsigned long get_seconds(void)
1194 return timekeeper
.xtime
.tv_sec
;
1196 EXPORT_SYMBOL(get_seconds
);
1198 struct timespec
__current_kernel_time(void)
1200 return timekeeper
.xtime
;
1203 struct timespec
current_kernel_time(void)
1205 struct timespec now
;
1209 seq
= read_seqbegin(&timekeeper
.lock
);
1211 now
= timekeeper
.xtime
;
1212 } while (read_seqretry(&timekeeper
.lock
, seq
));
1216 EXPORT_SYMBOL(current_kernel_time
);
1218 struct timespec
get_monotonic_coarse(void)
1220 struct timespec now
, mono
;
1224 seq
= read_seqbegin(&timekeeper
.lock
);
1226 now
= timekeeper
.xtime
;
1227 mono
= timekeeper
.wall_to_monotonic
;
1228 } while (read_seqretry(&timekeeper
.lock
, seq
));
1230 set_normalized_timespec(&now
, now
.tv_sec
+ mono
.tv_sec
,
1231 now
.tv_nsec
+ mono
.tv_nsec
);
1236 * The 64-bit jiffies value is not atomic - you MUST NOT read it
1237 * without sampling the sequence number in xtime_lock.
1238 * jiffies is defined in the linker script...
1240 void do_timer(unsigned long ticks
)
1242 jiffies_64
+= ticks
;
1244 calc_global_load(ticks
);
1248 * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1249 * and sleep offsets.
1250 * @xtim: pointer to timespec to be set with xtime
1251 * @wtom: pointer to timespec to be set with wall_to_monotonic
1252 * @sleep: pointer to timespec to be set with time in suspend
1254 void get_xtime_and_monotonic_and_sleep_offset(struct timespec
*xtim
,
1255 struct timespec
*wtom
, struct timespec
*sleep
)
1260 seq
= read_seqbegin(&timekeeper
.lock
);
1261 *xtim
= timekeeper
.xtime
;
1262 *wtom
= timekeeper
.wall_to_monotonic
;
1263 *sleep
= timekeeper
.total_sleep_time
;
1264 } while (read_seqretry(&timekeeper
.lock
, seq
));
1268 * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1270 ktime_t
ktime_get_monotonic_offset(void)
1273 struct timespec wtom
;
1276 seq
= read_seqbegin(&timekeeper
.lock
);
1277 wtom
= timekeeper
.wall_to_monotonic
;
1278 } while (read_seqretry(&timekeeper
.lock
, seq
));
1280 return timespec_to_ktime(wtom
);
1284 * xtime_update() - advances the timekeeping infrastructure
1285 * @ticks: number of ticks, that have elapsed since the last call.
1287 * Must be called with interrupts disabled.
1289 void xtime_update(unsigned long ticks
)
1291 write_seqlock(&xtime_lock
);
1293 write_sequnlock(&xtime_lock
);