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 static struct timespec xtime_cache __attribute__ ((aligned (16)));
50 static inline void update_xtime_cache(u64 nsec)
53 timespec_add_ns(&xtime_cache, nsec);
56 static struct clocksource *clock; /* pointer to current clocksource */
59 #ifdef CONFIG_GENERIC_TIME
61 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
63 * private function, must hold xtime_lock lock when being
64 * called. Returns the number of nanoseconds since the
65 * last call to update_wall_time() (adjusted by NTP scaling)
67 static inline s64 __get_nsec_offset(void)
69 cycle_t cycle_now, cycle_delta;
72 /* read clocksource: */
73 cycle_now = clocksource_read(clock);
75 /* calculate the delta since the last update_wall_time: */
76 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
78 /* convert to nanoseconds: */
79 ns_offset = cyc2ns(clock, cycle_delta);
85 * __get_realtime_clock_ts - Returns the time of day in a timespec
86 * @ts: pointer to the timespec to be set
88 * Returns the time of day in a timespec. Used by
89 * do_gettimeofday() and get_realtime_clock_ts().
91 static inline void __get_realtime_clock_ts(struct timespec *ts)
97 seq = read_seqbegin(&xtime_lock);
100 nsecs = __get_nsec_offset();
102 } while (read_seqretry(&xtime_lock, seq));
104 timespec_add_ns(ts, nsecs);
108 * getnstimeofday - Returns the time of day in a timespec
109 * @ts: pointer to the timespec to be set
111 * Returns the time of day in a timespec.
113 void getnstimeofday(struct timespec *ts)
115 __get_realtime_clock_ts(ts);
118 EXPORT_SYMBOL(getnstimeofday);
121 * do_gettimeofday - Returns the time of day in a timeval
122 * @tv: pointer to the timeval to be set
124 * NOTE: Users should be converted to using get_realtime_clock_ts()
126 void do_gettimeofday(struct timeval *tv)
130 __get_realtime_clock_ts(&now);
131 tv->tv_sec = now.tv_sec;
132 tv->tv_usec = now.tv_nsec/1000;
135 EXPORT_SYMBOL(do_gettimeofday);
137 * do_settimeofday - Sets the time of day
138 * @tv: pointer to the timespec variable containing the new time
140 * Sets the time of day to the new time and update NTP and notify hrtimers
142 int do_settimeofday(struct timespec *tv)
145 time_t wtm_sec, sec = tv->tv_sec;
146 long wtm_nsec, nsec = tv->tv_nsec;
148 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
151 write_seqlock_irqsave(&xtime_lock, flags);
153 nsec -= __get_nsec_offset();
155 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
156 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
158 set_normalized_timespec(&xtime, sec, nsec);
159 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
164 update_vsyscall(&xtime, clock);
166 write_sequnlock_irqrestore(&xtime_lock, flags);
168 /* signal hrtimers about time change */
174 EXPORT_SYMBOL(do_settimeofday);
177 * change_clocksource - Swaps clocksources if a new one is available
179 * Accumulates current time interval and initializes new clocksource
181 static void change_clocksource(void)
183 struct clocksource *new;
187 new = clocksource_get_next();
192 now = clocksource_read(new);
193 nsec = __get_nsec_offset();
194 timespec_add_ns(&xtime, nsec);
197 clock->cycle_last = now;
200 clock->xtime_nsec = 0;
201 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
205 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
209 static inline void change_clocksource(void) { }
210 static inline s64 __get_nsec_offset(void) { return 0; }
214 * timekeeping_is_continuous - check to see if timekeeping is free running
216 int timekeeping_is_continuous(void)
222 seq = read_seqbegin(&xtime_lock);
224 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
226 } while (read_seqretry(&xtime_lock, seq));
232 * read_persistent_clock - Return time in seconds from the persistent clock.
234 * Weak dummy function for arches that do not yet support it.
235 * Returns seconds from epoch using the battery backed persistent clock.
236 * Returns zero if unsupported.
238 * XXX - Do be sure to remove it once all arches implement it.
240 unsigned long __attribute__((weak)) read_persistent_clock(void)
246 * timekeeping_init - Initializes the clocksource and common timekeeping values
248 void __init timekeeping_init(void)
251 unsigned long sec = read_persistent_clock();
253 write_seqlock_irqsave(&xtime_lock, flags);
257 clock = clocksource_get_next();
258 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
259 clock->cycle_last = clocksource_read(clock);
263 set_normalized_timespec(&wall_to_monotonic,
264 -xtime.tv_sec, -xtime.tv_nsec);
265 total_sleep_time = 0;
267 write_sequnlock_irqrestore(&xtime_lock, flags);
270 /* flag for if timekeeping is suspended */
271 static int timekeeping_suspended;
272 /* time in seconds when suspend began */
273 static unsigned long timekeeping_suspend_time;
274 /* xtime offset when we went into suspend */
275 static s64 timekeeping_suspend_nsecs;
278 * timekeeping_resume - Resumes the generic timekeeping subsystem.
281 * This is for the generic clocksource timekeeping.
282 * xtime/wall_to_monotonic/jiffies/etc are
283 * still managed by arch specific suspend/resume code.
285 static int timekeeping_resume(struct sys_device *dev)
288 unsigned long now = read_persistent_clock();
290 clocksource_resume();
292 write_seqlock_irqsave(&xtime_lock, flags);
294 if (now && (now > timekeeping_suspend_time)) {
295 unsigned long sleep_length = now - timekeeping_suspend_time;
297 xtime.tv_sec += sleep_length;
298 wall_to_monotonic.tv_sec -= sleep_length;
299 total_sleep_time += sleep_length;
301 /* Make sure that we have the correct xtime reference */
302 timespec_add_ns(&xtime, timekeeping_suspend_nsecs);
303 /* re-base the last cycle value */
304 clock->cycle_last = clocksource_read(clock);
306 timekeeping_suspended = 0;
307 write_sequnlock_irqrestore(&xtime_lock, flags);
309 touch_softlockup_watchdog();
311 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
313 /* Resume hrtimers */
314 hres_timers_resume();
319 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
323 timekeeping_suspend_time = read_persistent_clock();
325 write_seqlock_irqsave(&xtime_lock, flags);
326 /* Get the current xtime offset */
327 timekeeping_suspend_nsecs = __get_nsec_offset();
328 timekeeping_suspended = 1;
329 write_sequnlock_irqrestore(&xtime_lock, flags);
331 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
336 /* sysfs resume/suspend bits for timekeeping */
337 static struct sysdev_class timekeeping_sysclass = {
338 .name = "timekeeping",
339 .resume = timekeeping_resume,
340 .suspend = timekeeping_suspend,
343 static struct sys_device device_timer = {
345 .cls = &timekeeping_sysclass,
348 static int __init timekeeping_init_device(void)
350 int error = sysdev_class_register(&timekeeping_sysclass);
352 error = sysdev_register(&device_timer);
356 device_initcall(timekeeping_init_device);
359 * If the error is already larger, we look ahead even further
360 * to compensate for late or lost adjustments.
362 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
370 * Use the current error value to determine how much to look ahead.
371 * The larger the error the slower we adjust for it to avoid problems
372 * with losing too many ticks, otherwise we would overadjust and
373 * produce an even larger error. The smaller the adjustment the
374 * faster we try to adjust for it, as lost ticks can do less harm
375 * here. This is tuned so that an error of about 1 msec is adusted
376 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
378 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
379 error2 = abs(error2);
380 for (look_ahead = 0; error2 > 0; look_ahead++)
384 * Now calculate the error in (1 << look_ahead) ticks, but first
385 * remove the single look ahead already included in the error.
387 tick_error = current_tick_length() >>
388 (TICK_LENGTH_SHIFT - clock->shift + 1);
389 tick_error -= clock->xtime_interval >> 1;
390 error = ((error - tick_error) >> look_ahead) + tick_error;
392 /* Finally calculate the adjustment shift value. */
397 *interval = -*interval;
401 for (adj = 0; error > i; adj++)
410 * Adjust the multiplier to reduce the error value,
411 * this is optimized for the most common adjustments of -1,0,1,
412 * for other values we can do a bit more work.
414 static void clocksource_adjust(s64 offset)
416 s64 error, interval = clock->cycle_interval;
419 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
420 if (error > interval) {
422 if (likely(error <= interval))
425 adj = clocksource_bigadjust(error, &interval, &offset);
426 } else if (error < -interval) {
428 if (likely(error >= -interval)) {
430 interval = -interval;
433 adj = clocksource_bigadjust(error, &interval, &offset);
438 clock->xtime_interval += interval;
439 clock->xtime_nsec -= offset;
440 clock->error -= (interval - offset) <<
441 (TICK_LENGTH_SHIFT - clock->shift);
445 * update_wall_time - Uses the current clocksource to increment the wall time
447 * Called from the timer interrupt, must hold a write on xtime_lock.
449 void update_wall_time(void)
453 /* Make sure we're fully resumed: */
454 if (unlikely(timekeeping_suspended))
457 #ifdef CONFIG_GENERIC_TIME
458 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
460 offset = clock->cycle_interval;
462 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
464 /* normally this loop will run just once, however in the
465 * case of lost or late ticks, it will accumulate correctly.
467 while (offset >= clock->cycle_interval) {
468 /* accumulate one interval */
469 clock->xtime_nsec += clock->xtime_interval;
470 clock->cycle_last += clock->cycle_interval;
471 offset -= clock->cycle_interval;
473 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
474 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
479 /* accumulate error between NTP and clock interval */
480 clock->error += current_tick_length();
481 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
484 /* correct the clock when NTP error is too big */
485 clocksource_adjust(offset);
487 /* store full nanoseconds into xtime */
488 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
489 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
491 update_xtime_cache(cyc2ns(clock, offset));
493 /* check to see if there is a new clocksource to use */
494 change_clocksource();
495 update_vsyscall(&xtime, clock);
499 * getboottime - Return the real time of system boot.
500 * @ts: pointer to the timespec to be set
502 * Returns the time of day in a timespec.
504 * This is based on the wall_to_monotonic offset and the total suspend
505 * time. Calls to settimeofday will affect the value returned (which
506 * basically means that however wrong your real time clock is at boot time,
507 * you get the right time here).
509 void getboottime(struct timespec *ts)
511 set_normalized_timespec(ts,
512 - (wall_to_monotonic.tv_sec + total_sleep_time),
513 - wall_to_monotonic.tv_nsec);
517 * monotonic_to_bootbased - Convert the monotonic time to boot based.
518 * @ts: pointer to the timespec to be converted
520 void monotonic_to_bootbased(struct timespec *ts)
522 ts->tv_sec += total_sleep_time;
525 unsigned long get_seconds(void)
527 return xtime_cache.tv_sec;
529 EXPORT_SYMBOL(get_seconds);
532 struct timespec current_kernel_time(void)
538 seq = read_seqbegin(&xtime_lock);
541 } while (read_seqretry(&xtime_lock, seq));
545 EXPORT_SYMBOL(current_kernel_time);