blob: 73edd4074b500970920fe0daee28ad760f768c83 [file] [log] [blame]
john stultz85240702007-05-08 00:27:59 -07001/*
2 * linux/kernel/time/timekeeping.c
3 *
4 * Kernel timekeeping code and accessor functions
5 *
6 * This code was moved from linux/kernel/timer.c.
7 * Please see that file for copyright and history logs.
8 *
9 */
10
11#include <linux/module.h>
12#include <linux/interrupt.h>
13#include <linux/percpu.h>
14#include <linux/init.h>
15#include <linux/mm.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040016#include <linux/sched.h>
john stultz85240702007-05-08 00:27:59 -070017#include <linux/sysdev.h>
18#include <linux/clocksource.h>
19#include <linux/jiffies.h>
20#include <linux/time.h>
21#include <linux/tick.h>
Martin Schwidefsky75c51582009-08-14 15:47:30 +020022#include <linux/stop_machine.h>
john stultz85240702007-05-08 00:27:59 -070023
Martin Schwidefsky155ec602009-08-14 15:47:26 +020024/* Structure holding internal timekeeping values. */
25struct timekeeper {
26 /* Current clocksource used for timekeeping. */
27 struct clocksource *clock;
Martin Schwidefsky23ce7212009-08-14 15:47:27 +020028 /* The shift value of the current clocksource. */
29 int shift;
Martin Schwidefsky155ec602009-08-14 15:47:26 +020030
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. */
34 u64 xtime_interval;
35 /* Raw nano seconds accumulated per NTP interval. */
36 u32 raw_interval;
37
38 /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
39 u64 xtime_nsec;
40 /* Difference between accumulated time and NTP time in ntp
41 * shifted nano seconds. */
42 s64 ntp_error;
Martin Schwidefsky23ce7212009-08-14 15:47:27 +020043 /* Shift conversion between clock shifted nano seconds and
44 * ntp shifted nano seconds. */
45 int ntp_error_shift;
Martin Schwidefsky0a544192009-08-14 15:47:28 +020046 /* NTP adjusted clock multiplier */
47 u32 mult;
Martin Schwidefsky155ec602009-08-14 15:47:26 +020048};
49
50struct timekeeper timekeeper;
51
52/**
53 * timekeeper_setup_internals - Set up internals to use clocksource clock.
54 *
55 * @clock: Pointer to clocksource.
56 *
57 * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
58 * pair and interval request.
59 *
60 * Unless you're the timekeeping code, you should not be using this!
61 */
62static void timekeeper_setup_internals(struct clocksource *clock)
63{
64 cycle_t interval;
65 u64 tmp;
66
67 timekeeper.clock = clock;
68 clock->cycle_last = clock->read(clock);
69
70 /* Do the ns -> cycle conversion first, using original mult */
71 tmp = NTP_INTERVAL_LENGTH;
72 tmp <<= clock->shift;
Martin Schwidefsky0a544192009-08-14 15:47:28 +020073 tmp += clock->mult/2;
74 do_div(tmp, clock->mult);
Martin Schwidefsky155ec602009-08-14 15:47:26 +020075 if (tmp == 0)
76 tmp = 1;
77
78 interval = (cycle_t) tmp;
79 timekeeper.cycle_interval = interval;
80
81 /* Go back from cycles -> shifted ns */
82 timekeeper.xtime_interval = (u64) interval * clock->mult;
83 timekeeper.raw_interval =
Martin Schwidefsky0a544192009-08-14 15:47:28 +020084 ((u64) interval * clock->mult) >> clock->shift;
Martin Schwidefsky155ec602009-08-14 15:47:26 +020085
86 timekeeper.xtime_nsec = 0;
Martin Schwidefsky23ce7212009-08-14 15:47:27 +020087 timekeeper.shift = clock->shift;
Martin Schwidefsky155ec602009-08-14 15:47:26 +020088
89 timekeeper.ntp_error = 0;
Martin Schwidefsky23ce7212009-08-14 15:47:27 +020090 timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
Martin Schwidefsky0a544192009-08-14 15:47:28 +020091
92 /*
93 * The timekeeper keeps its own mult values for the currently
94 * active clocksource. These value will be adjusted via NTP
95 * to counteract clock drifting.
96 */
97 timekeeper.mult = clock->mult;
Martin Schwidefsky155ec602009-08-14 15:47:26 +020098}
john stultz85240702007-05-08 00:27:59 -070099
Martin Schwidefsky2ba2a302009-08-14 15:47:29 +0200100/* Timekeeper helper functions. */
101static inline s64 timekeeping_get_ns(void)
102{
103 cycle_t cycle_now, cycle_delta;
104 struct clocksource *clock;
105
106 /* read clocksource: */
107 clock = timekeeper.clock;
108 cycle_now = clock->read(clock);
109
110 /* calculate the delta since the last update_wall_time: */
111 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
112
113 /* return delta convert to nanoseconds using ntp adjusted mult. */
114 return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
115 timekeeper.shift);
116}
117
118static inline s64 timekeeping_get_ns_raw(void)
119{
120 cycle_t cycle_now, cycle_delta;
121 struct clocksource *clock;
122
123 /* read clocksource: */
124 clock = timekeeper.clock;
125 cycle_now = clock->read(clock);
126
127 /* calculate the delta since the last update_wall_time: */
128 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
129
130 /* return delta convert to nanoseconds using ntp adjusted mult. */
131 return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
132}
133
john stultz85240702007-05-08 00:27:59 -0700134/*
135 * This read-write spinlock protects us from races in SMP while
Thomas Gleixnerdce48a82009-04-11 10:43:41 +0200136 * playing with xtime.
john stultz85240702007-05-08 00:27:59 -0700137 */
Adrian Bunkba2a6312007-10-16 23:27:16 -0700138__cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
john stultz85240702007-05-08 00:27:59 -0700139
140
141/*
142 * The current time
143 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
144 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
145 * at zero at system boot time, so wall_to_monotonic will be negative,
146 * however, we will ALWAYS keep the tv_nsec part positive so we can use
147 * the usual normalization.
Tomas Janousek7c3f1a52007-07-15 23:39:41 -0700148 *
149 * wall_to_monotonic is moved after resume from suspend for the monotonic
150 * time not to jump. We need to add total_sleep_time to wall_to_monotonic
151 * to get the real boot based time offset.
152 *
153 * - wall_to_monotonic is no longer the boot time, getboottime must be
154 * used instead.
john stultz85240702007-05-08 00:27:59 -0700155 */
156struct timespec xtime __attribute__ ((aligned (16)));
157struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200158static struct timespec total_sleep_time;
john stultz85240702007-05-08 00:27:59 -0700159
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200160/*
161 * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock.
162 */
163struct timespec raw_time;
164
Thomas Gleixner1c5745a2008-12-22 23:05:28 +0100165/* flag for if timekeeping is suspended */
166int __read_mostly timekeeping_suspended;
167
John Stultz31089c12009-08-14 15:47:18 +0200168/* must hold xtime_lock */
169void timekeeping_leap_insert(int leapsecond)
170{
171 xtime.tv_sec += leapsecond;
172 wall_to_monotonic.tv_sec -= leapsecond;
Lin Ming0696b712009-11-17 13:49:50 +0800173 update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult);
John Stultz31089c12009-08-14 15:47:18 +0200174}
john stultz85240702007-05-08 00:27:59 -0700175
john stultz85240702007-05-08 00:27:59 -0700176/**
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200177 * timekeeping_forward_now - update clock to the current time
john stultz85240702007-05-08 00:27:59 -0700178 *
Roman Zippel9a055112008-08-20 16:37:28 -0700179 * Forward the current clock to update its state since the last call to
180 * update_wall_time(). This is useful before significant clock changes,
181 * as it avoids having to deal with this time offset explicitly.
john stultz85240702007-05-08 00:27:59 -0700182 */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200183static void timekeeping_forward_now(void)
john stultz85240702007-05-08 00:27:59 -0700184{
185 cycle_t cycle_now, cycle_delta;
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200186 struct clocksource *clock;
Roman Zippel9a055112008-08-20 16:37:28 -0700187 s64 nsec;
john stultz85240702007-05-08 00:27:59 -0700188
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200189 clock = timekeeper.clock;
Martin Schwidefskya0f7d482009-08-14 15:47:19 +0200190 cycle_now = clock->read(clock);
john stultz85240702007-05-08 00:27:59 -0700191 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
Roman Zippel9a055112008-08-20 16:37:28 -0700192 clock->cycle_last = cycle_now;
john stultz85240702007-05-08 00:27:59 -0700193
Martin Schwidefsky0a544192009-08-14 15:47:28 +0200194 nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult,
195 timekeeper.shift);
john stultz7d275582009-05-01 13:10:26 -0700196
197 /* If arch requires, add in gettimeoffset() */
198 nsec += arch_gettimeoffset();
199
Roman Zippel9a055112008-08-20 16:37:28 -0700200 timespec_add_ns(&xtime, nsec);
John Stultz2d422442008-08-20 16:37:30 -0700201
Martin Schwidefsky0a544192009-08-14 15:47:28 +0200202 nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200203 timespec_add_ns(&raw_time, nsec);
john stultz85240702007-05-08 00:27:59 -0700204}
205
206/**
Geert Uytterhoevenefd9ac82008-01-30 13:30:01 +0100207 * getnstimeofday - Returns the time of day in a timespec
john stultz85240702007-05-08 00:27:59 -0700208 * @ts: pointer to the timespec to be set
209 *
Geert Uytterhoevenefd9ac82008-01-30 13:30:01 +0100210 * Returns the time of day in a timespec.
john stultz85240702007-05-08 00:27:59 -0700211 */
Geert Uytterhoevenefd9ac82008-01-30 13:30:01 +0100212void getnstimeofday(struct timespec *ts)
john stultz85240702007-05-08 00:27:59 -0700213{
214 unsigned long seq;
215 s64 nsecs;
216
Thomas Gleixner1c5745a2008-12-22 23:05:28 +0100217 WARN_ON(timekeeping_suspended);
218
john stultz85240702007-05-08 00:27:59 -0700219 do {
220 seq = read_seqbegin(&xtime_lock);
221
222 *ts = xtime;
Martin Schwidefsky2ba2a302009-08-14 15:47:29 +0200223 nsecs = timekeeping_get_ns();
john stultz85240702007-05-08 00:27:59 -0700224
john stultz7d275582009-05-01 13:10:26 -0700225 /* If arch requires, add in gettimeoffset() */
226 nsecs += arch_gettimeoffset();
227
john stultz85240702007-05-08 00:27:59 -0700228 } while (read_seqretry(&xtime_lock, seq));
229
230 timespec_add_ns(ts, nsecs);
231}
232
john stultz85240702007-05-08 00:27:59 -0700233EXPORT_SYMBOL(getnstimeofday);
234
Martin Schwidefsky951ed4d2009-07-07 11:27:28 +0200235ktime_t ktime_get(void)
236{
Martin Schwidefsky951ed4d2009-07-07 11:27:28 +0200237 unsigned int seq;
238 s64 secs, nsecs;
239
240 WARN_ON(timekeeping_suspended);
241
242 do {
243 seq = read_seqbegin(&xtime_lock);
244 secs = xtime.tv_sec + wall_to_monotonic.tv_sec;
245 nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec;
Martin Schwidefsky2ba2a302009-08-14 15:47:29 +0200246 nsecs += timekeeping_get_ns();
Martin Schwidefsky951ed4d2009-07-07 11:27:28 +0200247
248 } while (read_seqretry(&xtime_lock, seq));
249 /*
250 * Use ktime_set/ktime_add_ns to create a proper ktime on
251 * 32-bit architectures without CONFIG_KTIME_SCALAR.
252 */
253 return ktime_add_ns(ktime_set(secs, 0), nsecs);
254}
255EXPORT_SYMBOL_GPL(ktime_get);
256
257/**
258 * ktime_get_ts - get the monotonic clock in timespec format
259 * @ts: pointer to timespec variable
260 *
261 * The function calculates the monotonic clock from the realtime
262 * clock and the wall_to_monotonic offset and stores the result
263 * in normalized timespec format in the variable pointed to by @ts.
264 */
265void ktime_get_ts(struct timespec *ts)
266{
Martin Schwidefsky951ed4d2009-07-07 11:27:28 +0200267 struct timespec tomono;
268 unsigned int seq;
269 s64 nsecs;
270
271 WARN_ON(timekeeping_suspended);
272
273 do {
274 seq = read_seqbegin(&xtime_lock);
275 *ts = xtime;
276 tomono = wall_to_monotonic;
Martin Schwidefsky2ba2a302009-08-14 15:47:29 +0200277 nsecs = timekeeping_get_ns();
Martin Schwidefsky951ed4d2009-07-07 11:27:28 +0200278
279 } while (read_seqretry(&xtime_lock, seq));
280
281 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
282 ts->tv_nsec + tomono.tv_nsec + nsecs);
283}
284EXPORT_SYMBOL_GPL(ktime_get_ts);
285
john stultz85240702007-05-08 00:27:59 -0700286/**
287 * do_gettimeofday - Returns the time of day in a timeval
288 * @tv: pointer to the timeval to be set
289 *
Geert Uytterhoevenefd9ac82008-01-30 13:30:01 +0100290 * NOTE: Users should be converted to using getnstimeofday()
john stultz85240702007-05-08 00:27:59 -0700291 */
292void do_gettimeofday(struct timeval *tv)
293{
294 struct timespec now;
295
Geert Uytterhoevenefd9ac82008-01-30 13:30:01 +0100296 getnstimeofday(&now);
john stultz85240702007-05-08 00:27:59 -0700297 tv->tv_sec = now.tv_sec;
298 tv->tv_usec = now.tv_nsec/1000;
299}
300
301EXPORT_SYMBOL(do_gettimeofday);
302/**
303 * do_settimeofday - Sets the time of day
304 * @tv: pointer to the timespec variable containing the new time
305 *
306 * Sets the time of day to the new time and update NTP and notify hrtimers
307 */
308int do_settimeofday(struct timespec *tv)
309{
Roman Zippel9a055112008-08-20 16:37:28 -0700310 struct timespec ts_delta;
john stultz85240702007-05-08 00:27:59 -0700311 unsigned long flags;
john stultz85240702007-05-08 00:27:59 -0700312
313 if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
314 return -EINVAL;
315
316 write_seqlock_irqsave(&xtime_lock, flags);
317
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200318 timekeeping_forward_now();
john stultz85240702007-05-08 00:27:59 -0700319
Roman Zippel9a055112008-08-20 16:37:28 -0700320 ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec;
321 ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec;
322 wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta);
john stultz85240702007-05-08 00:27:59 -0700323
Roman Zippel9a055112008-08-20 16:37:28 -0700324 xtime = *tv;
325
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200326 timekeeper.ntp_error = 0;
john stultz85240702007-05-08 00:27:59 -0700327 ntp_clear();
328
Lin Ming0696b712009-11-17 13:49:50 +0800329 update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult);
john stultz85240702007-05-08 00:27:59 -0700330
331 write_sequnlock_irqrestore(&xtime_lock, flags);
332
333 /* signal hrtimers about time change */
334 clock_was_set();
335
336 return 0;
337}
338
339EXPORT_SYMBOL(do_settimeofday);
340
341/**
342 * change_clocksource - Swaps clocksources if a new one is available
343 *
344 * Accumulates current time interval and initializes new clocksource
345 */
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200346static int change_clocksource(void *data)
john stultz85240702007-05-08 00:27:59 -0700347{
Magnus Damm4614e6a2009-04-21 12:24:02 -0700348 struct clocksource *new, *old;
john stultz85240702007-05-08 00:27:59 -0700349
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200350 new = (struct clocksource *) data;
john stultz85240702007-05-08 00:27:59 -0700351
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200352 timekeeping_forward_now();
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200353 if (!new->enable || new->enable(new) == 0) {
354 old = timekeeper.clock;
355 timekeeper_setup_internals(new);
356 if (old->disable)
357 old->disable(old);
358 }
359 return 0;
360}
john stultz85240702007-05-08 00:27:59 -0700361
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200362/**
363 * timekeeping_notify - Install a new clock source
364 * @clock: pointer to the clock source
365 *
366 * This function is called from clocksource.c after a new, better clock
367 * source has been registered. The caller holds the clocksource_mutex.
368 */
369void timekeeping_notify(struct clocksource *clock)
370{
371 if (timekeeper.clock == clock)
Magnus Damm4614e6a2009-04-21 12:24:02 -0700372 return;
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200373 stop_machine(change_clocksource, clock, NULL);
john stultz85240702007-05-08 00:27:59 -0700374 tick_clock_notify();
john stultz85240702007-05-08 00:27:59 -0700375}
Martin Schwidefsky75c51582009-08-14 15:47:30 +0200376
Thomas Gleixnera40f2622009-07-07 13:00:31 +0200377/**
378 * ktime_get_real - get the real (wall-) time in ktime_t format
379 *
380 * returns the time in ktime_t format
381 */
382ktime_t ktime_get_real(void)
383{
384 struct timespec now;
385
386 getnstimeofday(&now);
387
388 return timespec_to_ktime(now);
389}
390EXPORT_SYMBOL_GPL(ktime_get_real);
john stultz85240702007-05-08 00:27:59 -0700391
392/**
John Stultz2d422442008-08-20 16:37:30 -0700393 * getrawmonotonic - Returns the raw monotonic time in a timespec
394 * @ts: pointer to the timespec to be set
395 *
396 * Returns the raw monotonic time (completely un-modified by ntp)
397 */
398void getrawmonotonic(struct timespec *ts)
399{
400 unsigned long seq;
401 s64 nsecs;
John Stultz2d422442008-08-20 16:37:30 -0700402
403 do {
404 seq = read_seqbegin(&xtime_lock);
Martin Schwidefsky2ba2a302009-08-14 15:47:29 +0200405 nsecs = timekeeping_get_ns_raw();
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200406 *ts = raw_time;
John Stultz2d422442008-08-20 16:37:30 -0700407
408 } while (read_seqretry(&xtime_lock, seq));
409
410 timespec_add_ns(ts, nsecs);
411}
412EXPORT_SYMBOL(getrawmonotonic);
413
414
415/**
Li Zefancf4fc6c2008-02-08 04:19:24 -0800416 * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
john stultz85240702007-05-08 00:27:59 -0700417 */
Li Zefancf4fc6c2008-02-08 04:19:24 -0800418int timekeeping_valid_for_hres(void)
john stultz85240702007-05-08 00:27:59 -0700419{
420 unsigned long seq;
421 int ret;
422
423 do {
424 seq = read_seqbegin(&xtime_lock);
425
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200426 ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
john stultz85240702007-05-08 00:27:59 -0700427
428 } while (read_seqretry(&xtime_lock, seq));
429
430 return ret;
431}
432
433/**
Jon Hunter98962462009-08-18 12:45:10 -0500434 * timekeeping_max_deferment - Returns max time the clocksource can be deferred
435 *
436 * Caller must observe xtime_lock via read_seqbegin/read_seqretry to
437 * ensure that the clocksource does not change!
438 */
439u64 timekeeping_max_deferment(void)
440{
441 return timekeeper.clock->max_idle_ns;
442}
443
444/**
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200445 * read_persistent_clock - Return time from the persistent clock.
john stultz85240702007-05-08 00:27:59 -0700446 *
447 * Weak dummy function for arches that do not yet support it.
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200448 * Reads the time from the battery backed persistent clock.
449 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
john stultz85240702007-05-08 00:27:59 -0700450 *
451 * XXX - Do be sure to remove it once all arches implement it.
452 */
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200453void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
john stultz85240702007-05-08 00:27:59 -0700454{
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200455 ts->tv_sec = 0;
456 ts->tv_nsec = 0;
john stultz85240702007-05-08 00:27:59 -0700457}
458
Martin Schwidefsky23970e32009-08-14 15:47:32 +0200459/**
460 * read_boot_clock - Return time of the system start.
461 *
462 * Weak dummy function for arches that do not yet support it.
463 * Function to read the exact time the system has been started.
464 * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
465 *
466 * XXX - Do be sure to remove it once all arches implement it.
467 */
468void __attribute__((weak)) read_boot_clock(struct timespec *ts)
469{
470 ts->tv_sec = 0;
471 ts->tv_nsec = 0;
472}
473
john stultz85240702007-05-08 00:27:59 -0700474/*
475 * timekeeping_init - Initializes the clocksource and common timekeeping values
476 */
477void __init timekeeping_init(void)
478{
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200479 struct clocksource *clock;
john stultz85240702007-05-08 00:27:59 -0700480 unsigned long flags;
Martin Schwidefsky23970e32009-08-14 15:47:32 +0200481 struct timespec now, boot;
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200482
483 read_persistent_clock(&now);
Martin Schwidefsky23970e32009-08-14 15:47:32 +0200484 read_boot_clock(&boot);
john stultz85240702007-05-08 00:27:59 -0700485
486 write_seqlock_irqsave(&xtime_lock, flags);
487
Roman Zippel7dffa3c2008-05-01 04:34:41 -0700488 ntp_init();
john stultz85240702007-05-08 00:27:59 -0700489
Martin Schwidefskyf1b82742009-08-14 15:47:21 +0200490 clock = clocksource_default_clock();
Martin Schwidefskya0f7d482009-08-14 15:47:19 +0200491 if (clock->enable)
492 clock->enable(clock);
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200493 timekeeper_setup_internals(clock);
john stultz85240702007-05-08 00:27:59 -0700494
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200495 xtime.tv_sec = now.tv_sec;
496 xtime.tv_nsec = now.tv_nsec;
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200497 raw_time.tv_sec = 0;
498 raw_time.tv_nsec = 0;
Martin Schwidefsky23970e32009-08-14 15:47:32 +0200499 if (boot.tv_sec == 0 && boot.tv_nsec == 0) {
500 boot.tv_sec = xtime.tv_sec;
501 boot.tv_nsec = xtime.tv_nsec;
502 }
john stultz85240702007-05-08 00:27:59 -0700503 set_normalized_timespec(&wall_to_monotonic,
Martin Schwidefsky23970e32009-08-14 15:47:32 +0200504 -boot.tv_sec, -boot.tv_nsec);
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200505 total_sleep_time.tv_sec = 0;
506 total_sleep_time.tv_nsec = 0;
john stultz85240702007-05-08 00:27:59 -0700507 write_sequnlock_irqrestore(&xtime_lock, flags);
508}
509
john stultz85240702007-05-08 00:27:59 -0700510/* time in seconds when suspend began */
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200511static struct timespec timekeeping_suspend_time;
john stultz85240702007-05-08 00:27:59 -0700512
513/**
514 * timekeeping_resume - Resumes the generic timekeeping subsystem.
515 * @dev: unused
516 *
517 * This is for the generic clocksource timekeeping.
518 * xtime/wall_to_monotonic/jiffies/etc are
519 * still managed by arch specific suspend/resume code.
520 */
521static int timekeeping_resume(struct sys_device *dev)
522{
523 unsigned long flags;
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200524 struct timespec ts;
525
526 read_persistent_clock(&ts);
john stultz85240702007-05-08 00:27:59 -0700527
Thomas Gleixnerd10ff3f2007-05-14 11:10:02 +0200528 clocksource_resume();
529
john stultz85240702007-05-08 00:27:59 -0700530 write_seqlock_irqsave(&xtime_lock, flags);
531
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200532 if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
533 ts = timespec_sub(ts, timekeeping_suspend_time);
John Stultzce3bf7a2010-07-13 17:56:19 -0700534 xtime = timespec_add(xtime, ts);
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200535 wall_to_monotonic = timespec_sub(wall_to_monotonic, ts);
John Stultzce3bf7a2010-07-13 17:56:19 -0700536 total_sleep_time = timespec_add(total_sleep_time, ts);
john stultz85240702007-05-08 00:27:59 -0700537 }
538 /* re-base the last cycle value */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200539 timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock);
540 timekeeper.ntp_error = 0;
john stultz85240702007-05-08 00:27:59 -0700541 timekeeping_suspended = 0;
542 write_sequnlock_irqrestore(&xtime_lock, flags);
543
544 touch_softlockup_watchdog();
545
546 clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
547
548 /* Resume hrtimers */
549 hres_timers_resume();
550
551 return 0;
552}
553
554static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
555{
556 unsigned long flags;
557
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200558 read_persistent_clock(&timekeeping_suspend_time);
Thomas Gleixner3be90952007-09-16 15:36:43 +0200559
john stultz85240702007-05-08 00:27:59 -0700560 write_seqlock_irqsave(&xtime_lock, flags);
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200561 timekeeping_forward_now();
john stultz85240702007-05-08 00:27:59 -0700562 timekeeping_suspended = 1;
john stultz85240702007-05-08 00:27:59 -0700563 write_sequnlock_irqrestore(&xtime_lock, flags);
564
565 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
Magnus Dammc54a42b2010-02-02 14:41:41 -0800566 clocksource_suspend();
john stultz85240702007-05-08 00:27:59 -0700567
568 return 0;
569}
570
571/* sysfs resume/suspend bits for timekeeping */
572static struct sysdev_class timekeeping_sysclass = {
Kay Sieversaf5ca3f2007-12-20 02:09:39 +0100573 .name = "timekeeping",
john stultz85240702007-05-08 00:27:59 -0700574 .resume = timekeeping_resume,
575 .suspend = timekeeping_suspend,
john stultz85240702007-05-08 00:27:59 -0700576};
577
578static struct sys_device device_timer = {
579 .id = 0,
580 .cls = &timekeeping_sysclass,
581};
582
583static int __init timekeeping_init_device(void)
584{
585 int error = sysdev_class_register(&timekeeping_sysclass);
586 if (!error)
587 error = sysdev_register(&device_timer);
588 return error;
589}
590
591device_initcall(timekeeping_init_device);
592
593/*
594 * If the error is already larger, we look ahead even further
595 * to compensate for late or lost adjustments.
596 */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200597static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval,
john stultz85240702007-05-08 00:27:59 -0700598 s64 *offset)
599{
600 s64 tick_error, i;
601 u32 look_ahead, adj;
602 s32 error2, mult;
603
604 /*
605 * Use the current error value to determine how much to look ahead.
606 * The larger the error the slower we adjust for it to avoid problems
607 * with losing too many ticks, otherwise we would overadjust and
608 * produce an even larger error. The smaller the adjustment the
609 * faster we try to adjust for it, as lost ticks can do less harm
Li Zefan3eb05672008-02-08 04:19:25 -0800610 * here. This is tuned so that an error of about 1 msec is adjusted
john stultz85240702007-05-08 00:27:59 -0700611 * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
612 */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200613 error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
john stultz85240702007-05-08 00:27:59 -0700614 error2 = abs(error2);
615 for (look_ahead = 0; error2 > 0; look_ahead++)
616 error2 >>= 2;
617
618 /*
619 * Now calculate the error in (1 << look_ahead) ticks, but first
620 * remove the single look ahead already included in the error.
621 */
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200622 tick_error = tick_length >> (timekeeper.ntp_error_shift + 1);
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200623 tick_error -= timekeeper.xtime_interval >> 1;
john stultz85240702007-05-08 00:27:59 -0700624 error = ((error - tick_error) >> look_ahead) + tick_error;
625
626 /* Finally calculate the adjustment shift value. */
627 i = *interval;
628 mult = 1;
629 if (error < 0) {
630 error = -error;
631 *interval = -*interval;
632 *offset = -*offset;
633 mult = -1;
634 }
635 for (adj = 0; error > i; adj++)
636 error >>= 1;
637
638 *interval <<= adj;
639 *offset <<= adj;
640 return mult << adj;
641}
642
643/*
644 * Adjust the multiplier to reduce the error value,
645 * this is optimized for the most common adjustments of -1,0,1,
646 * for other values we can do a bit more work.
647 */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200648static void timekeeping_adjust(s64 offset)
john stultz85240702007-05-08 00:27:59 -0700649{
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200650 s64 error, interval = timekeeper.cycle_interval;
john stultz85240702007-05-08 00:27:59 -0700651 int adj;
652
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200653 error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1);
john stultz85240702007-05-08 00:27:59 -0700654 if (error > interval) {
655 error >>= 2;
656 if (likely(error <= interval))
657 adj = 1;
658 else
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200659 adj = timekeeping_bigadjust(error, &interval, &offset);
john stultz85240702007-05-08 00:27:59 -0700660 } else if (error < -interval) {
661 error >>= 2;
662 if (likely(error >= -interval)) {
663 adj = -1;
664 interval = -interval;
665 offset = -offset;
666 } else
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200667 adj = timekeeping_bigadjust(error, &interval, &offset);
john stultz85240702007-05-08 00:27:59 -0700668 } else
669 return;
670
Martin Schwidefsky0a544192009-08-14 15:47:28 +0200671 timekeeper.mult += adj;
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200672 timekeeper.xtime_interval += interval;
673 timekeeper.xtime_nsec -= offset;
674 timekeeper.ntp_error -= (interval - offset) <<
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200675 timekeeper.ntp_error_shift;
john stultz85240702007-05-08 00:27:59 -0700676}
677
Linus Torvalds83f57a12009-12-22 14:10:37 -0800678
john stultz85240702007-05-08 00:27:59 -0700679/**
john stultza092ff02009-10-02 16:17:53 -0700680 * logarithmic_accumulation - shifted accumulation of cycles
681 *
682 * This functions accumulates a shifted interval of cycles into
683 * into a shifted interval nanoseconds. Allows for O(log) accumulation
684 * loop.
685 *
686 * Returns the unconsumed cycles.
687 */
688static cycle_t logarithmic_accumulation(cycle_t offset, int shift)
689{
690 u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift;
691
692 /* If the offset is smaller then a shifted interval, do nothing */
693 if (offset < timekeeper.cycle_interval<<shift)
694 return offset;
695
696 /* Accumulate one shifted interval */
697 offset -= timekeeper.cycle_interval << shift;
698 timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift;
699
700 timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
701 while (timekeeper.xtime_nsec >= nsecps) {
702 timekeeper.xtime_nsec -= nsecps;
703 xtime.tv_sec++;
704 second_overflow();
705 }
706
707 /* Accumulate into raw time */
708 raw_time.tv_nsec += timekeeper.raw_interval << shift;;
709 while (raw_time.tv_nsec >= NSEC_PER_SEC) {
710 raw_time.tv_nsec -= NSEC_PER_SEC;
711 raw_time.tv_sec++;
712 }
713
714 /* Accumulate error between NTP and clock interval */
715 timekeeper.ntp_error += tick_length << shift;
716 timekeeper.ntp_error -= timekeeper.xtime_interval <<
717 (timekeeper.ntp_error_shift + shift);
718
719 return offset;
720}
721
Linus Torvalds83f57a12009-12-22 14:10:37 -0800722
john stultz85240702007-05-08 00:27:59 -0700723/**
724 * update_wall_time - Uses the current clocksource to increment the wall time
725 *
726 * Called from the timer interrupt, must hold a write on xtime_lock.
727 */
728void update_wall_time(void)
729{
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200730 struct clocksource *clock;
john stultz85240702007-05-08 00:27:59 -0700731 cycle_t offset;
john stultza092ff02009-10-02 16:17:53 -0700732 int shift = 0, maxshift;
john stultz85240702007-05-08 00:27:59 -0700733
734 /* Make sure we're fully resumed: */
735 if (unlikely(timekeeping_suspended))
736 return;
737
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200738 clock = timekeeper.clock;
John Stultz592913e2010-07-13 17:56:20 -0700739
740#ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200741 offset = timekeeper.cycle_interval;
John Stultz592913e2010-07-13 17:56:20 -0700742#else
743 offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
john stultz85240702007-05-08 00:27:59 -0700744#endif
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200745 timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift;
john stultz85240702007-05-08 00:27:59 -0700746
john stultza092ff02009-10-02 16:17:53 -0700747 /*
748 * With NO_HZ we may have to accumulate many cycle_intervals
749 * (think "ticks") worth of time at once. To do this efficiently,
750 * we calculate the largest doubling multiple of cycle_intervals
751 * that is smaller then the offset. We then accumulate that
752 * chunk in one go, and then try to consume the next smaller
753 * doubled multiple.
john stultz85240702007-05-08 00:27:59 -0700754 */
john stultza092ff02009-10-02 16:17:53 -0700755 shift = ilog2(offset) - ilog2(timekeeper.cycle_interval);
756 shift = max(0, shift);
757 /* Bound shift to one less then what overflows tick_length */
758 maxshift = (8*sizeof(tick_length) - (ilog2(tick_length)+1)) - 1;
759 shift = min(shift, maxshift);
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200760 while (offset >= timekeeper.cycle_interval) {
john stultza092ff02009-10-02 16:17:53 -0700761 offset = logarithmic_accumulation(offset, shift);
John Stultz830ec042010-03-18 14:47:30 -0700762 if(offset < timekeeper.cycle_interval<<shift)
763 shift--;
john stultz85240702007-05-08 00:27:59 -0700764 }
765
766 /* correct the clock when NTP error is too big */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200767 timekeeping_adjust(offset);
john stultz85240702007-05-08 00:27:59 -0700768
john stultz6c9bacb2008-12-01 18:34:41 -0800769 /*
770 * Since in the loop above, we accumulate any amount of time
771 * in xtime_nsec over a second into xtime.tv_sec, its possible for
772 * xtime_nsec to be fairly small after the loop. Further, if we're
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200773 * slightly speeding the clocksource up in timekeeping_adjust(),
john stultz6c9bacb2008-12-01 18:34:41 -0800774 * its possible the required corrective factor to xtime_nsec could
775 * cause it to underflow.
776 *
777 * Now, we cannot simply roll the accumulated second back, since
778 * the NTP subsystem has been notified via second_overflow. So
779 * instead we push xtime_nsec forward by the amount we underflowed,
780 * and add that amount into the error.
781 *
782 * We'll correct this error next time through this function, when
783 * xtime_nsec is not as small.
784 */
Martin Schwidefsky155ec602009-08-14 15:47:26 +0200785 if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
786 s64 neg = -(s64)timekeeper.xtime_nsec;
787 timekeeper.xtime_nsec = 0;
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200788 timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
john stultz6c9bacb2008-12-01 18:34:41 -0800789 }
790
John Stultz6a867a32010-04-06 14:30:51 -0700791
792 /*
793 * Store full nanoseconds into xtime after rounding it up and
Roman Zippel5cd1c9c2008-09-22 14:42:43 -0700794 * add the remainder to the error difference.
795 */
Martin Schwidefsky23ce7212009-08-14 15:47:27 +0200796 xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1;
797 timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift;
798 timekeeper.ntp_error += timekeeper.xtime_nsec <<
799 timekeeper.ntp_error_shift;
john stultz85240702007-05-08 00:27:59 -0700800
John Stultz6a867a32010-04-06 14:30:51 -0700801 /*
802 * Finally, make sure that after the rounding
803 * xtime.tv_nsec isn't larger then NSEC_PER_SEC
804 */
805 if (unlikely(xtime.tv_nsec >= NSEC_PER_SEC)) {
806 xtime.tv_nsec -= NSEC_PER_SEC;
807 xtime.tv_sec++;
808 second_overflow();
809 }
Linus Torvalds83f57a12009-12-22 14:10:37 -0800810
john stultz85240702007-05-08 00:27:59 -0700811 /* check to see if there is a new clocksource to use */
Lin Ming0696b712009-11-17 13:49:50 +0800812 update_vsyscall(&xtime, timekeeper.clock, timekeeper.mult);
john stultz85240702007-05-08 00:27:59 -0700813}
Tomas Janousek7c3f1a52007-07-15 23:39:41 -0700814
815/**
816 * getboottime - Return the real time of system boot.
817 * @ts: pointer to the timespec to be set
818 *
819 * Returns the time of day in a timespec.
820 *
821 * This is based on the wall_to_monotonic offset and the total suspend
822 * time. Calls to settimeofday will affect the value returned (which
823 * basically means that however wrong your real time clock is at boot time,
824 * you get the right time here).
825 */
826void getboottime(struct timespec *ts)
827{
Hiroshi Shimamoto36d47482009-08-25 15:08:30 +0900828 struct timespec boottime = {
829 .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec,
830 .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec
831 };
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200832
Martin Schwidefskyd4f587c2009-08-14 15:47:31 +0200833 set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
Tomas Janousek7c3f1a52007-07-15 23:39:41 -0700834}
Jason Wangc93d89f2010-01-27 19:13:40 +0800835EXPORT_SYMBOL_GPL(getboottime);
Tomas Janousek7c3f1a52007-07-15 23:39:41 -0700836
837/**
838 * monotonic_to_bootbased - Convert the monotonic time to boot based.
839 * @ts: pointer to the timespec to be converted
840 */
841void monotonic_to_bootbased(struct timespec *ts)
842{
John Stultzce3bf7a2010-07-13 17:56:19 -0700843 *ts = timespec_add(*ts, total_sleep_time);
Tomas Janousek7c3f1a52007-07-15 23:39:41 -0700844}
Jason Wangc93d89f2010-01-27 19:13:40 +0800845EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
john stultz2c6b47d2007-07-24 17:47:43 -0700846
john stultz17c38b72007-07-24 18:38:34 -0700847unsigned long get_seconds(void)
848{
John Stultz6a867a32010-04-06 14:30:51 -0700849 return xtime.tv_sec;
john stultz17c38b72007-07-24 18:38:34 -0700850}
851EXPORT_SYMBOL(get_seconds);
852
john stultzda15cfd2009-08-19 19:13:34 -0700853struct timespec __current_kernel_time(void)
854{
John Stultz6a867a32010-04-06 14:30:51 -0700855 return xtime;
john stultzda15cfd2009-08-19 19:13:34 -0700856}
john stultz17c38b72007-07-24 18:38:34 -0700857
john stultz2c6b47d2007-07-24 17:47:43 -0700858struct timespec current_kernel_time(void)
859{
860 struct timespec now;
861 unsigned long seq;
862
863 do {
864 seq = read_seqbegin(&xtime_lock);
Linus Torvalds83f57a12009-12-22 14:10:37 -0800865
John Stultz6a867a32010-04-06 14:30:51 -0700866 now = xtime;
john stultz2c6b47d2007-07-24 17:47:43 -0700867 } while (read_seqretry(&xtime_lock, seq));
868
869 return now;
870}
john stultz2c6b47d2007-07-24 17:47:43 -0700871EXPORT_SYMBOL(current_kernel_time);
john stultzda15cfd2009-08-19 19:13:34 -0700872
873struct timespec get_monotonic_coarse(void)
874{
875 struct timespec now, mono;
876 unsigned long seq;
877
878 do {
879 seq = read_seqbegin(&xtime_lock);
Linus Torvalds83f57a12009-12-22 14:10:37 -0800880
John Stultz6a867a32010-04-06 14:30:51 -0700881 now = xtime;
john stultzda15cfd2009-08-19 19:13:34 -0700882 mono = wall_to_monotonic;
883 } while (read_seqretry(&xtime_lock, seq));
884
885 set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
886 now.tv_nsec + mono.tv_nsec);
887 return now;
888}