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 __attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
29 EXPORT_SYMBOL(xtime_lock);
34 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
35 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
36 * at zero at system boot time, so wall_to_monotonic will be negative,
37 * however, we will ALWAYS keep the tv_nsec part positive so we can use
38 * the usual normalization.
40 * wall_to_monotonic is moved after resume from suspend for the monotonic
41 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
42 * to get the real boot based time offset.
44 * - wall_to_monotonic is no longer the boot time, getboottime must be
47 struct timespec xtime __attribute__ ((aligned (16)));
48 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
49 static unsigned long total_sleep_time; /* seconds */
54 static struct timespec xtime_cache __attribute__ ((aligned (16)));
55 static inline void update_xtime_cache(u64 nsec)
58 timespec_add_ns(&xtime_cache, nsec);
61 #define xtime_cache xtime
62 /* We do *not* want to evaluate the argument for this case */
63 #define update_xtime_cache(n) do { } while (0)
66 static struct clocksource *clock; /* pointer to current clocksource */
69 #ifdef CONFIG_GENERIC_TIME
71 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
73 * private function, must hold xtime_lock lock when being
74 * called. Returns the number of nanoseconds since the
75 * last call to update_wall_time() (adjusted by NTP scaling)
77 static inline s64 __get_nsec_offset(void)
79 cycle_t cycle_now, cycle_delta;
82 /* read clocksource: */
83 cycle_now = clocksource_read(clock);
85 /* calculate the delta since the last update_wall_time: */
86 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
88 /* convert to nanoseconds: */
89 ns_offset = cyc2ns(clock, cycle_delta);
95 * __get_realtime_clock_ts - Returns the time of day in a timespec
96 * @ts: pointer to the timespec to be set
98 * Returns the time of day in a timespec. Used by
99 * do_gettimeofday() and get_realtime_clock_ts().
101 static inline void __get_realtime_clock_ts(struct timespec *ts)
107 seq = read_seqbegin(&xtime_lock);
110 nsecs = __get_nsec_offset();
112 } while (read_seqretry(&xtime_lock, seq));
114 timespec_add_ns(ts, nsecs);
118 * getnstimeofday - Returns the time of day in a timespec
119 * @ts: pointer to the timespec to be set
121 * Returns the time of day in a timespec.
123 void getnstimeofday(struct timespec *ts)
125 __get_realtime_clock_ts(ts);
128 EXPORT_SYMBOL(getnstimeofday);
131 * do_gettimeofday - Returns the time of day in a timeval
132 * @tv: pointer to the timeval to be set
134 * NOTE: Users should be converted to using get_realtime_clock_ts()
136 void do_gettimeofday(struct timeval *tv)
140 __get_realtime_clock_ts(&now);
141 tv->tv_sec = now.tv_sec;
142 tv->tv_usec = now.tv_nsec/1000;
145 EXPORT_SYMBOL(do_gettimeofday);
147 * do_settimeofday - Sets the time of day
148 * @tv: pointer to the timespec variable containing the new time
150 * Sets the time of day to the new time and update NTP and notify hrtimers
152 int do_settimeofday(struct timespec *tv)
155 time_t wtm_sec, sec = tv->tv_sec;
156 long wtm_nsec, nsec = tv->tv_nsec;
158 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
161 write_seqlock_irqsave(&xtime_lock, flags);
163 nsec -= __get_nsec_offset();
165 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
166 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
168 set_normalized_timespec(&xtime, sec, nsec);
169 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
174 update_vsyscall(&xtime, clock);
176 write_sequnlock_irqrestore(&xtime_lock, flags);
178 /* signal hrtimers about time change */
184 EXPORT_SYMBOL(do_settimeofday);
187 * change_clocksource - Swaps clocksources if a new one is available
189 * Accumulates current time interval and initializes new clocksource
191 static void change_clocksource(void)
193 struct clocksource *new;
197 new = clocksource_get_next();
202 now = clocksource_read(new);
203 nsec = __get_nsec_offset();
204 timespec_add_ns(&xtime, nsec);
207 clock->cycle_last = now;
210 clock->xtime_nsec = 0;
211 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
215 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
219 static inline void change_clocksource(void) { }
220 static inline s64 __get_nsec_offset(void) { return 0; }
224 * timekeeping_is_continuous - check to see if timekeeping is free running
226 int timekeeping_is_continuous(void)
232 seq = read_seqbegin(&xtime_lock);
234 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
236 } while (read_seqretry(&xtime_lock, seq));
242 * read_persistent_clock - Return time in seconds from the persistent clock.
244 * Weak dummy function for arches that do not yet support it.
245 * Returns seconds from epoch using the battery backed persistent clock.
246 * Returns zero if unsupported.
248 * XXX - Do be sure to remove it once all arches implement it.
250 unsigned long __attribute__((weak)) read_persistent_clock(void)
256 * timekeeping_init - Initializes the clocksource and common timekeeping values
258 void __init timekeeping_init(void)
261 unsigned long sec = read_persistent_clock();
263 write_seqlock_irqsave(&xtime_lock, flags);
267 clock = clocksource_get_next();
268 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
269 clock->cycle_last = clocksource_read(clock);
273 set_normalized_timespec(&wall_to_monotonic,
274 -xtime.tv_sec, -xtime.tv_nsec);
275 total_sleep_time = 0;
277 write_sequnlock_irqrestore(&xtime_lock, flags);
280 /* flag for if timekeeping is suspended */
281 static int timekeeping_suspended;
282 /* time in seconds when suspend began */
283 static unsigned long timekeeping_suspend_time;
284 /* xtime offset when we went into suspend */
285 static s64 timekeeping_suspend_nsecs;
288 * timekeeping_resume - Resumes the generic timekeeping subsystem.
291 * This is for the generic clocksource timekeeping.
292 * xtime/wall_to_monotonic/jiffies/etc are
293 * still managed by arch specific suspend/resume code.
295 static int timekeeping_resume(struct sys_device *dev)
298 unsigned long now = read_persistent_clock();
300 clocksource_resume();
302 write_seqlock_irqsave(&xtime_lock, flags);
304 if (now && (now > timekeeping_suspend_time)) {
305 unsigned long sleep_length = now - timekeeping_suspend_time;
307 xtime.tv_sec += sleep_length;
308 wall_to_monotonic.tv_sec -= sleep_length;
309 total_sleep_time += sleep_length;
311 /* Make sure that we have the correct xtime reference */
312 timespec_add_ns(&xtime, timekeeping_suspend_nsecs);
313 /* re-base the last cycle value */
314 clock->cycle_last = clocksource_read(clock);
316 timekeeping_suspended = 0;
317 write_sequnlock_irqrestore(&xtime_lock, flags);
319 touch_softlockup_watchdog();
321 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
323 /* Resume hrtimers */
324 hres_timers_resume();
329 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
333 timekeeping_suspend_time = read_persistent_clock();
335 write_seqlock_irqsave(&xtime_lock, flags);
336 /* Get the current xtime offset */
337 timekeeping_suspend_nsecs = __get_nsec_offset();
338 timekeeping_suspended = 1;
339 write_sequnlock_irqrestore(&xtime_lock, flags);
341 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
346 /* sysfs resume/suspend bits for timekeeping */
347 static struct sysdev_class timekeeping_sysclass = {
348 .resume = timekeeping_resume,
349 .suspend = timekeeping_suspend,
350 set_kset_name("timekeeping"),
353 static struct sys_device device_timer = {
355 .cls = &timekeeping_sysclass,
358 static int __init timekeeping_init_device(void)
360 int error = sysdev_class_register(&timekeeping_sysclass);
362 error = sysdev_register(&device_timer);
366 device_initcall(timekeeping_init_device);
369 * If the error is already larger, we look ahead even further
370 * to compensate for late or lost adjustments.
372 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
380 * Use the current error value to determine how much to look ahead.
381 * The larger the error the slower we adjust for it to avoid problems
382 * with losing too many ticks, otherwise we would overadjust and
383 * produce an even larger error. The smaller the adjustment the
384 * faster we try to adjust for it, as lost ticks can do less harm
385 * here. This is tuned so that an error of about 1 msec is adusted
386 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
388 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
389 error2 = abs(error2);
390 for (look_ahead = 0; error2 > 0; look_ahead++)
394 * Now calculate the error in (1 << look_ahead) ticks, but first
395 * remove the single look ahead already included in the error.
397 tick_error = current_tick_length() >>
398 (TICK_LENGTH_SHIFT - clock->shift + 1);
399 tick_error -= clock->xtime_interval >> 1;
400 error = ((error - tick_error) >> look_ahead) + tick_error;
402 /* Finally calculate the adjustment shift value. */
407 *interval = -*interval;
411 for (adj = 0; error > i; adj++)
420 * Adjust the multiplier to reduce the error value,
421 * this is optimized for the most common adjustments of -1,0,1,
422 * for other values we can do a bit more work.
424 static void clocksource_adjust(s64 offset)
426 s64 error, interval = clock->cycle_interval;
429 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
430 if (error > interval) {
432 if (likely(error <= interval))
435 adj = clocksource_bigadjust(error, &interval, &offset);
436 } else if (error < -interval) {
438 if (likely(error >= -interval)) {
440 interval = -interval;
443 adj = clocksource_bigadjust(error, &interval, &offset);
448 clock->xtime_interval += interval;
449 clock->xtime_nsec -= offset;
450 clock->error -= (interval - offset) <<
451 (TICK_LENGTH_SHIFT - clock->shift);
455 * update_wall_time - Uses the current clocksource to increment the wall time
457 * Called from the timer interrupt, must hold a write on xtime_lock.
459 void update_wall_time(void)
463 /* Make sure we're fully resumed: */
464 if (unlikely(timekeeping_suspended))
467 #ifdef CONFIG_GENERIC_TIME
468 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
470 offset = clock->cycle_interval;
472 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
474 /* normally this loop will run just once, however in the
475 * case of lost or late ticks, it will accumulate correctly.
477 while (offset >= clock->cycle_interval) {
478 /* accumulate one interval */
479 clock->xtime_nsec += clock->xtime_interval;
480 clock->cycle_last += clock->cycle_interval;
481 offset -= clock->cycle_interval;
483 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
484 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
489 /* accumulate error between NTP and clock interval */
490 clock->error += current_tick_length();
491 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
494 /* correct the clock when NTP error is too big */
495 clocksource_adjust(offset);
497 /* store full nanoseconds into xtime */
498 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
499 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
501 update_xtime_cache(cyc2ns(clock, offset));
503 /* check to see if there is a new clocksource to use */
504 change_clocksource();
505 update_vsyscall(&xtime, clock);
509 * getboottime - Return the real time of system boot.
510 * @ts: pointer to the timespec to be set
512 * Returns the time of day in a timespec.
514 * This is based on the wall_to_monotonic offset and the total suspend
515 * time. Calls to settimeofday will affect the value returned (which
516 * basically means that however wrong your real time clock is at boot time,
517 * you get the right time here).
519 void getboottime(struct timespec *ts)
521 set_normalized_timespec(ts,
522 - (wall_to_monotonic.tv_sec + total_sleep_time),
523 - wall_to_monotonic.tv_nsec);
527 * monotonic_to_bootbased - Convert the monotonic time to boot based.
528 * @ts: pointer to the timespec to be converted
530 void monotonic_to_bootbased(struct timespec *ts)
532 ts->tv_sec += total_sleep_time;
535 unsigned long get_seconds(void)
537 return xtime_cache.tv_sec;
539 EXPORT_SYMBOL(get_seconds);
542 struct timespec current_kernel_time(void)
548 seq = read_seqbegin(&xtime_lock);
551 } while (read_seqretry(&xtime_lock, seq));
555 EXPORT_SYMBOL(current_kernel_time);