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/sysdev.h>
17 #include <linux/clocksource.h>
18 #include <linux/jiffies.h>
19 #include <linux/time.h>
20 #include <linux/tick.h>
24 * This read-write spinlock protects us from races in SMP while
25 * playing with xtime and avenrun.
27 __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
32 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
33 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
34 * at zero at system boot time, so wall_to_monotonic will be negative,
35 * however, we will ALWAYS keep the tv_nsec part positive so we can use
36 * the usual normalization.
38 * wall_to_monotonic is moved after resume from suspend for the monotonic
39 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
40 * to get the real boot based time offset.
42 * - wall_to_monotonic is no longer the boot time, getboottime must be
45 struct timespec xtime __attribute__ ((aligned (16)));
46 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
47 static unsigned long total_sleep_time; /* seconds */
49 /* flag for if timekeeping is suspended */
50 int __read_mostly timekeeping_suspended;
52 static struct timespec xtime_cache __attribute__ ((aligned (16)));
53 void update_xtime_cache(u64 nsec)
56 timespec_add_ns(&xtime_cache, nsec);
59 struct clocksource *clock;
62 #ifdef CONFIG_GENERIC_TIME
64 * clocksource_forward_now - update clock to the current time
66 * Forward the current clock to update its state since the last call to
67 * update_wall_time(). This is useful before significant clock changes,
68 * as it avoids having to deal with this time offset explicitly.
70 static void clocksource_forward_now(void)
72 cycle_t cycle_now, cycle_delta;
75 cycle_now = clocksource_read(clock);
76 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
77 clock->cycle_last = cycle_now;
79 nsec = cyc2ns(clock, cycle_delta);
80 timespec_add_ns(&xtime, nsec);
84 * getnstimeofday - Returns the time of day in a timespec
85 * @ts: pointer to the timespec to be set
87 * Returns the time of day in a timespec.
89 void getnstimeofday(struct timespec *ts)
91 cycle_t cycle_now, cycle_delta;
95 WARN_ON(timekeeping_suspended);
98 seq = read_seqbegin(&xtime_lock);
102 /* read clocksource: */
103 cycle_now = clocksource_read(clock);
105 /* calculate the delta since the last update_wall_time: */
106 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
108 /* convert to nanoseconds: */
109 nsecs = cyc2ns(clock, cycle_delta);
111 } while (read_seqretry(&xtime_lock, seq));
113 timespec_add_ns(ts, nsecs);
116 EXPORT_SYMBOL(getnstimeofday);
119 * do_gettimeofday - Returns the time of day in a timeval
120 * @tv: pointer to the timeval to be set
122 * NOTE: Users should be converted to using getnstimeofday()
124 void do_gettimeofday(struct timeval *tv)
128 getnstimeofday(&now);
129 tv->tv_sec = now.tv_sec;
130 tv->tv_usec = now.tv_nsec/1000;
133 EXPORT_SYMBOL(do_gettimeofday);
135 * do_settimeofday - Sets the time of day
136 * @tv: pointer to the timespec variable containing the new time
138 * Sets the time of day to the new time and update NTP and notify hrtimers
140 int do_settimeofday(struct timespec *tv)
142 struct timespec ts_delta;
145 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
148 write_seqlock_irqsave(&xtime_lock, flags);
150 clocksource_forward_now();
152 ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec;
153 ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec;
154 wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta);
158 update_xtime_cache(0);
163 update_vsyscall(&xtime, clock);
165 write_sequnlock_irqrestore(&xtime_lock, flags);
167 /* signal hrtimers about time change */
173 EXPORT_SYMBOL(do_settimeofday);
176 * change_clocksource - Swaps clocksources if a new one is available
178 * Accumulates current time interval and initializes new clocksource
180 static void change_clocksource(void)
182 struct clocksource *new;
184 new = clocksource_get_next();
189 clocksource_forward_now();
192 clock->cycle_last = 0;
193 clock->cycle_last = clocksource_read(new);
195 clock->xtime_nsec = 0;
196 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
201 * We're holding xtime lock and waking up klogd would deadlock
202 * us on enqueue. So no printing!
203 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
208 static inline void clocksource_forward_now(void) { }
209 static inline void change_clocksource(void) { }
213 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
215 int timekeeping_valid_for_hres(void)
221 seq = read_seqbegin(&xtime_lock);
223 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
225 } while (read_seqretry(&xtime_lock, seq));
231 * read_persistent_clock - Return time in seconds from the persistent clock.
233 * Weak dummy function for arches that do not yet support it.
234 * Returns seconds from epoch using the battery backed persistent clock.
235 * Returns zero if unsupported.
237 * XXX - Do be sure to remove it once all arches implement it.
239 unsigned long __attribute__((weak)) read_persistent_clock(void)
245 * timekeeping_init - Initializes the clocksource and common timekeeping values
247 void __init timekeeping_init(void)
250 unsigned long sec = read_persistent_clock();
252 write_seqlock_irqsave(&xtime_lock, flags);
256 clock = clocksource_get_next();
257 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
258 clock->cycle_last = clocksource_read(clock);
262 set_normalized_timespec(&wall_to_monotonic,
263 -xtime.tv_sec, -xtime.tv_nsec);
264 update_xtime_cache(0);
265 total_sleep_time = 0;
266 write_sequnlock_irqrestore(&xtime_lock, flags);
269 /* time in seconds when suspend began */
270 static unsigned long timekeeping_suspend_time;
273 * timekeeping_resume - Resumes the generic timekeeping subsystem.
276 * This is for the generic clocksource timekeeping.
277 * xtime/wall_to_monotonic/jiffies/etc are
278 * still managed by arch specific suspend/resume code.
280 static int timekeeping_resume(struct sys_device *dev)
283 unsigned long now = read_persistent_clock();
285 clocksource_resume();
287 write_seqlock_irqsave(&xtime_lock, flags);
289 if (now && (now > timekeeping_suspend_time)) {
290 unsigned long sleep_length = now - timekeeping_suspend_time;
292 xtime.tv_sec += sleep_length;
293 wall_to_monotonic.tv_sec -= sleep_length;
294 total_sleep_time += sleep_length;
296 update_xtime_cache(0);
297 /* re-base the last cycle value */
298 clock->cycle_last = 0;
299 clock->cycle_last = clocksource_read(clock);
301 timekeeping_suspended = 0;
302 write_sequnlock_irqrestore(&xtime_lock, flags);
304 touch_softlockup_watchdog();
306 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
308 /* Resume hrtimers */
309 hres_timers_resume();
314 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
318 timekeeping_suspend_time = read_persistent_clock();
320 write_seqlock_irqsave(&xtime_lock, flags);
321 clocksource_forward_now();
322 timekeeping_suspended = 1;
323 write_sequnlock_irqrestore(&xtime_lock, flags);
325 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
330 /* sysfs resume/suspend bits for timekeeping */
331 static struct sysdev_class timekeeping_sysclass = {
332 .name = "timekeeping",
333 .resume = timekeeping_resume,
334 .suspend = timekeeping_suspend,
337 static struct sys_device device_timer = {
339 .cls = &timekeeping_sysclass,
342 static int __init timekeeping_init_device(void)
344 int error = sysdev_class_register(&timekeeping_sysclass);
346 error = sysdev_register(&device_timer);
350 device_initcall(timekeeping_init_device);
353 * If the error is already larger, we look ahead even further
354 * to compensate for late or lost adjustments.
356 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
364 * Use the current error value to determine how much to look ahead.
365 * The larger the error the slower we adjust for it to avoid problems
366 * with losing too many ticks, otherwise we would overadjust and
367 * produce an even larger error. The smaller the adjustment the
368 * faster we try to adjust for it, as lost ticks can do less harm
369 * here. This is tuned so that an error of about 1 msec is adjusted
370 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
372 error2 = clock->error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
373 error2 = abs(error2);
374 for (look_ahead = 0; error2 > 0; look_ahead++)
378 * Now calculate the error in (1 << look_ahead) ticks, but first
379 * remove the single look ahead already included in the error.
381 tick_error = tick_length >> (NTP_SCALE_SHIFT - clock->shift + 1);
382 tick_error -= clock->xtime_interval >> 1;
383 error = ((error - tick_error) >> look_ahead) + tick_error;
385 /* Finally calculate the adjustment shift value. */
390 *interval = -*interval;
394 for (adj = 0; error > i; adj++)
403 * Adjust the multiplier to reduce the error value,
404 * this is optimized for the most common adjustments of -1,0,1,
405 * for other values we can do a bit more work.
407 static void clocksource_adjust(s64 offset)
409 s64 error, interval = clock->cycle_interval;
412 error = clock->error >> (NTP_SCALE_SHIFT - clock->shift - 1);
413 if (error > interval) {
415 if (likely(error <= interval))
418 adj = clocksource_bigadjust(error, &interval, &offset);
419 } else if (error < -interval) {
421 if (likely(error >= -interval)) {
423 interval = -interval;
426 adj = clocksource_bigadjust(error, &interval, &offset);
431 clock->xtime_interval += interval;
432 clock->xtime_nsec -= offset;
433 clock->error -= (interval - offset) <<
434 (NTP_SCALE_SHIFT - clock->shift);
438 * update_wall_time - Uses the current clocksource to increment the wall time
440 * Called from the timer interrupt, must hold a write on xtime_lock.
442 void update_wall_time(void)
446 /* Make sure we're fully resumed: */
447 if (unlikely(timekeeping_suspended))
450 #ifdef CONFIG_GENERIC_TIME
451 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
453 offset = clock->cycle_interval;
455 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
457 /* normally this loop will run just once, however in the
458 * case of lost or late ticks, it will accumulate correctly.
460 while (offset >= clock->cycle_interval) {
461 /* accumulate one interval */
462 offset -= clock->cycle_interval;
463 clock->cycle_last += clock->cycle_interval;
465 clock->xtime_nsec += clock->xtime_interval;
466 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
467 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
472 /* accumulate error between NTP and clock interval */
473 clock->error += tick_length;
474 clock->error -= clock->xtime_interval << (NTP_SCALE_SHIFT - clock->shift);
477 /* correct the clock when NTP error is too big */
478 clocksource_adjust(offset);
481 * Since in the loop above, we accumulate any amount of time
482 * in xtime_nsec over a second into xtime.tv_sec, its possible for
483 * xtime_nsec to be fairly small after the loop. Further, if we're
484 * slightly speeding the clocksource up in clocksource_adjust(),
485 * its possible the required corrective factor to xtime_nsec could
486 * cause it to underflow.
488 * Now, we cannot simply roll the accumulated second back, since
489 * the NTP subsystem has been notified via second_overflow. So
490 * instead we push xtime_nsec forward by the amount we underflowed,
491 * and add that amount into the error.
493 * We'll correct this error next time through this function, when
494 * xtime_nsec is not as small.
496 if (unlikely((s64)clock->xtime_nsec < 0)) {
497 s64 neg = -(s64)clock->xtime_nsec;
498 clock->xtime_nsec = 0;
499 clock->error += neg << (NTP_SCALE_SHIFT - clock->shift);
502 /* store full nanoseconds into xtime */
503 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
504 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
506 update_xtime_cache(cyc2ns(clock, offset));
508 /* check to see if there is a new clocksource to use */
509 change_clocksource();
510 update_vsyscall(&xtime, clock);
514 * getboottime - Return the real time of system boot.
515 * @ts: pointer to the timespec to be set
517 * Returns the time of day in a timespec.
519 * This is based on the wall_to_monotonic offset and the total suspend
520 * time. Calls to settimeofday will affect the value returned (which
521 * basically means that however wrong your real time clock is at boot time,
522 * you get the right time here).
524 void getboottime(struct timespec *ts)
526 set_normalized_timespec(ts,
527 - (wall_to_monotonic.tv_sec + total_sleep_time),
528 - wall_to_monotonic.tv_nsec);
532 * monotonic_to_bootbased - Convert the monotonic time to boot based.
533 * @ts: pointer to the timespec to be converted
535 void monotonic_to_bootbased(struct timespec *ts)
537 ts->tv_sec += total_sleep_time;
540 unsigned long get_seconds(void)
542 return xtime_cache.tv_sec;
544 EXPORT_SYMBOL(get_seconds);
547 struct timespec current_kernel_time(void)
553 seq = read_seqbegin(&xtime_lock);
556 } while (read_seqretry(&xtime_lock, seq));
560 EXPORT_SYMBOL(current_kernel_time);