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 struct timespec xtime __attribute__ ((aligned (16)));
41 struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
46 static struct clocksource *clock; /* pointer to current clocksource */
49 #ifdef CONFIG_GENERIC_TIME
51 * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
53 * private function, must hold xtime_lock lock when being
54 * called. Returns the number of nanoseconds since the
55 * last call to update_wall_time() (adjusted by NTP scaling)
57 static inline s64 __get_nsec_offset(void)
59 cycle_t cycle_now, cycle_delta;
62 /* read clocksource: */
63 cycle_now = clocksource_read(clock);
65 /* calculate the delta since the last update_wall_time: */
66 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
68 /* convert to nanoseconds: */
69 ns_offset = cyc2ns(clock, cycle_delta);
75 * __get_realtime_clock_ts - Returns the time of day in a timespec
76 * @ts: pointer to the timespec to be set
78 * Returns the time of day in a timespec. Used by
79 * do_gettimeofday() and get_realtime_clock_ts().
81 static inline void __get_realtime_clock_ts(struct timespec *ts)
87 seq = read_seqbegin(&xtime_lock);
90 nsecs = __get_nsec_offset();
92 } while (read_seqretry(&xtime_lock, seq));
94 timespec_add_ns(ts, nsecs);
98 * getnstimeofday - Returns the time of day in a timespec
99 * @ts: pointer to the timespec to be set
101 * Returns the time of day in a timespec.
103 void getnstimeofday(struct timespec *ts)
105 __get_realtime_clock_ts(ts);
108 EXPORT_SYMBOL(getnstimeofday);
111 * do_gettimeofday - Returns the time of day in a timeval
112 * @tv: pointer to the timeval to be set
114 * NOTE: Users should be converted to using get_realtime_clock_ts()
116 void do_gettimeofday(struct timeval *tv)
120 __get_realtime_clock_ts(&now);
121 tv->tv_sec = now.tv_sec;
122 tv->tv_usec = now.tv_nsec/1000;
125 EXPORT_SYMBOL(do_gettimeofday);
127 * do_settimeofday - Sets the time of day
128 * @tv: pointer to the timespec variable containing the new time
130 * Sets the time of day to the new time and update NTP and notify hrtimers
132 int do_settimeofday(struct timespec *tv)
135 time_t wtm_sec, sec = tv->tv_sec;
136 long wtm_nsec, nsec = tv->tv_nsec;
138 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
141 write_seqlock_irqsave(&xtime_lock, flags);
143 nsec -= __get_nsec_offset();
145 wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
146 wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
148 set_normalized_timespec(&xtime, sec, nsec);
149 set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
154 update_vsyscall(&xtime, clock);
156 write_sequnlock_irqrestore(&xtime_lock, flags);
158 /* signal hrtimers about time change */
164 EXPORT_SYMBOL(do_settimeofday);
167 * change_clocksource - Swaps clocksources if a new one is available
169 * Accumulates current time interval and initializes new clocksource
171 static void change_clocksource(void)
173 struct clocksource *new;
177 new = clocksource_get_next();
182 now = clocksource_read(new);
183 nsec = __get_nsec_offset();
184 timespec_add_ns(&xtime, nsec);
187 clock->cycle_last = now;
190 clock->xtime_nsec = 0;
191 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
195 printk(KERN_INFO "Time: %s clocksource has been installed.\n",
199 static inline void change_clocksource(void) { }
203 * timekeeping_is_continuous - check to see if timekeeping is free running
205 int timekeeping_is_continuous(void)
211 seq = read_seqbegin(&xtime_lock);
213 ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
215 } while (read_seqretry(&xtime_lock, seq));
221 * read_persistent_clock - Return time in seconds from the persistent clock.
223 * Weak dummy function for arches that do not yet support it.
224 * Returns seconds from epoch using the battery backed persistent clock.
225 * Returns zero if unsupported.
227 * XXX - Do be sure to remove it once all arches implement it.
229 unsigned long __attribute__((weak)) read_persistent_clock(void)
235 * timekeeping_init - Initializes the clocksource and common timekeeping values
237 void __init timekeeping_init(void)
240 unsigned long sec = read_persistent_clock();
242 write_seqlock_irqsave(&xtime_lock, flags);
246 clock = clocksource_get_next();
247 clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH);
248 clock->cycle_last = clocksource_read(clock);
252 set_normalized_timespec(&wall_to_monotonic,
253 -xtime.tv_sec, -xtime.tv_nsec);
255 write_sequnlock_irqrestore(&xtime_lock, flags);
258 /* flag for if timekeeping is suspended */
259 static int timekeeping_suspended;
260 /* time in seconds when suspend began */
261 static unsigned long timekeeping_suspend_time;
264 * timekeeping_resume - Resumes the generic timekeeping subsystem.
267 * This is for the generic clocksource timekeeping.
268 * xtime/wall_to_monotonic/jiffies/etc are
269 * still managed by arch specific suspend/resume code.
271 static int timekeeping_resume(struct sys_device *dev)
274 unsigned long now = read_persistent_clock();
276 clocksource_resume();
278 write_seqlock_irqsave(&xtime_lock, flags);
280 if (now && (now > timekeeping_suspend_time)) {
281 unsigned long sleep_length = now - timekeeping_suspend_time;
283 xtime.tv_sec += sleep_length;
284 wall_to_monotonic.tv_sec -= sleep_length;
286 /* re-base the last cycle value */
287 clock->cycle_last = clocksource_read(clock);
289 timekeeping_suspended = 0;
290 write_sequnlock_irqrestore(&xtime_lock, flags);
292 touch_softlockup_watchdog();
294 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
296 /* Resume hrtimers */
297 hres_timers_resume();
302 static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
306 write_seqlock_irqsave(&xtime_lock, flags);
307 timekeeping_suspended = 1;
308 timekeeping_suspend_time = read_persistent_clock();
309 write_sequnlock_irqrestore(&xtime_lock, flags);
311 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
316 /* sysfs resume/suspend bits for timekeeping */
317 static struct sysdev_class timekeeping_sysclass = {
318 .resume = timekeeping_resume,
319 .suspend = timekeeping_suspend,
320 set_kset_name("timekeeping"),
323 static struct sys_device device_timer = {
325 .cls = &timekeeping_sysclass,
328 static int __init timekeeping_init_device(void)
330 int error = sysdev_class_register(&timekeeping_sysclass);
332 error = sysdev_register(&device_timer);
336 device_initcall(timekeeping_init_device);
339 * If the error is already larger, we look ahead even further
340 * to compensate for late or lost adjustments.
342 static __always_inline int clocksource_bigadjust(s64 error, s64 *interval,
350 * Use the current error value to determine how much to look ahead.
351 * The larger the error the slower we adjust for it to avoid problems
352 * with losing too many ticks, otherwise we would overadjust and
353 * produce an even larger error. The smaller the adjustment the
354 * faster we try to adjust for it, as lost ticks can do less harm
355 * here. This is tuned so that an error of about 1 msec is adusted
356 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
358 error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
359 error2 = abs(error2);
360 for (look_ahead = 0; error2 > 0; look_ahead++)
364 * Now calculate the error in (1 << look_ahead) ticks, but first
365 * remove the single look ahead already included in the error.
367 tick_error = current_tick_length() >>
368 (TICK_LENGTH_SHIFT - clock->shift + 1);
369 tick_error -= clock->xtime_interval >> 1;
370 error = ((error - tick_error) >> look_ahead) + tick_error;
372 /* Finally calculate the adjustment shift value. */
377 *interval = -*interval;
381 for (adj = 0; error > i; adj++)
390 * Adjust the multiplier to reduce the error value,
391 * this is optimized for the most common adjustments of -1,0,1,
392 * for other values we can do a bit more work.
394 static void clocksource_adjust(struct clocksource *clock, s64 offset)
396 s64 error, interval = clock->cycle_interval;
399 error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
400 if (error > interval) {
402 if (likely(error <= interval))
405 adj = clocksource_bigadjust(error, &interval, &offset);
406 } else if (error < -interval) {
408 if (likely(error >= -interval)) {
410 interval = -interval;
413 adj = clocksource_bigadjust(error, &interval, &offset);
418 clock->xtime_interval += interval;
419 clock->xtime_nsec -= offset;
420 clock->error -= (interval - offset) <<
421 (TICK_LENGTH_SHIFT - clock->shift);
425 * update_wall_time - Uses the current clocksource to increment the wall time
427 * Called from the timer interrupt, must hold a write on xtime_lock.
429 void update_wall_time(void)
433 /* Make sure we're fully resumed: */
434 if (unlikely(timekeeping_suspended))
437 #ifdef CONFIG_GENERIC_TIME
438 offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
440 offset = clock->cycle_interval;
442 clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
444 /* normally this loop will run just once, however in the
445 * case of lost or late ticks, it will accumulate correctly.
447 while (offset >= clock->cycle_interval) {
448 /* accumulate one interval */
449 clock->xtime_nsec += clock->xtime_interval;
450 clock->cycle_last += clock->cycle_interval;
451 offset -= clock->cycle_interval;
453 if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
454 clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
459 /* interpolator bits */
460 time_interpolator_update(clock->xtime_interval
463 /* accumulate error between NTP and clock interval */
464 clock->error += current_tick_length();
465 clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
468 /* correct the clock when NTP error is too big */
469 clocksource_adjust(clock, offset);
471 /* store full nanoseconds into xtime */
472 xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
473 clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
475 /* check to see if there is a new clocksource to use */
476 change_clocksource();
477 update_vsyscall(&xtime, clock);