]>
Commit | Line | Data |
---|---|---|
8524070b JS |
1 | /* |
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> | |
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> | |
21 | ||
22 | ||
23 | /* | |
24 | * This read-write spinlock protects us from races in SMP while | |
25 | * playing with xtime and avenrun. | |
26 | */ | |
27 | __attribute__((weak)) __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock); | |
28 | ||
29 | EXPORT_SYMBOL(xtime_lock); | |
30 | ||
31 | ||
32 | /* | |
33 | * The current time | |
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. | |
7c3f1a57 TJ |
39 | * |
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. | |
43 | * | |
44 | * - wall_to_monotonic is no longer the boot time, getboottime must be | |
45 | * used instead. | |
8524070b JS |
46 | */ |
47 | struct timespec xtime __attribute__ ((aligned (16))); | |
48 | struct timespec wall_to_monotonic __attribute__ ((aligned (16))); | |
7c3f1a57 | 49 | static unsigned long total_sleep_time; /* seconds */ |
8524070b JS |
50 | EXPORT_SYMBOL(xtime); |
51 | ||
52 | ||
17c38b74 JS |
53 | #ifdef CONFIG_NO_HZ |
54 | static struct timespec xtime_cache __attribute__ ((aligned (16))); | |
55 | static inline void update_xtime_cache(u64 nsec) | |
56 | { | |
57 | xtime_cache = xtime; | |
58 | timespec_add_ns(&xtime_cache, nsec); | |
59 | } | |
60 | #else | |
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) | |
64 | #endif | |
65 | ||
8524070b JS |
66 | static struct clocksource *clock; /* pointer to current clocksource */ |
67 | ||
68 | ||
69 | #ifdef CONFIG_GENERIC_TIME | |
70 | /** | |
71 | * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook | |
72 | * | |
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) | |
76 | */ | |
77 | static inline s64 __get_nsec_offset(void) | |
78 | { | |
79 | cycle_t cycle_now, cycle_delta; | |
80 | s64 ns_offset; | |
81 | ||
82 | /* read clocksource: */ | |
83 | cycle_now = clocksource_read(clock); | |
84 | ||
85 | /* calculate the delta since the last update_wall_time: */ | |
86 | cycle_delta = (cycle_now - clock->cycle_last) & clock->mask; | |
87 | ||
88 | /* convert to nanoseconds: */ | |
89 | ns_offset = cyc2ns(clock, cycle_delta); | |
90 | ||
91 | return ns_offset; | |
92 | } | |
93 | ||
94 | /** | |
95 | * __get_realtime_clock_ts - Returns the time of day in a timespec | |
96 | * @ts: pointer to the timespec to be set | |
97 | * | |
98 | * Returns the time of day in a timespec. Used by | |
99 | * do_gettimeofday() and get_realtime_clock_ts(). | |
100 | */ | |
101 | static inline void __get_realtime_clock_ts(struct timespec *ts) | |
102 | { | |
103 | unsigned long seq; | |
104 | s64 nsecs; | |
105 | ||
106 | do { | |
107 | seq = read_seqbegin(&xtime_lock); | |
108 | ||
109 | *ts = xtime; | |
110 | nsecs = __get_nsec_offset(); | |
111 | ||
112 | } while (read_seqretry(&xtime_lock, seq)); | |
113 | ||
114 | timespec_add_ns(ts, nsecs); | |
115 | } | |
116 | ||
117 | /** | |
118 | * getnstimeofday - Returns the time of day in a timespec | |
119 | * @ts: pointer to the timespec to be set | |
120 | * | |
121 | * Returns the time of day in a timespec. | |
122 | */ | |
123 | void getnstimeofday(struct timespec *ts) | |
124 | { | |
125 | __get_realtime_clock_ts(ts); | |
126 | } | |
127 | ||
128 | EXPORT_SYMBOL(getnstimeofday); | |
129 | ||
130 | /** | |
131 | * do_gettimeofday - Returns the time of day in a timeval | |
132 | * @tv: pointer to the timeval to be set | |
133 | * | |
134 | * NOTE: Users should be converted to using get_realtime_clock_ts() | |
135 | */ | |
136 | void do_gettimeofday(struct timeval *tv) | |
137 | { | |
138 | struct timespec now; | |
139 | ||
140 | __get_realtime_clock_ts(&now); | |
141 | tv->tv_sec = now.tv_sec; | |
142 | tv->tv_usec = now.tv_nsec/1000; | |
143 | } | |
144 | ||
145 | EXPORT_SYMBOL(do_gettimeofday); | |
146 | /** | |
147 | * do_settimeofday - Sets the time of day | |
148 | * @tv: pointer to the timespec variable containing the new time | |
149 | * | |
150 | * Sets the time of day to the new time and update NTP and notify hrtimers | |
151 | */ | |
152 | int do_settimeofday(struct timespec *tv) | |
153 | { | |
154 | unsigned long flags; | |
155 | time_t wtm_sec, sec = tv->tv_sec; | |
156 | long wtm_nsec, nsec = tv->tv_nsec; | |
157 | ||
158 | if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC) | |
159 | return -EINVAL; | |
160 | ||
161 | write_seqlock_irqsave(&xtime_lock, flags); | |
162 | ||
163 | nsec -= __get_nsec_offset(); | |
164 | ||
165 | wtm_sec = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec); | |
166 | wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec); | |
167 | ||
168 | set_normalized_timespec(&xtime, sec, nsec); | |
169 | set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec); | |
170 | ||
171 | clock->error = 0; | |
172 | ntp_clear(); | |
173 | ||
174 | update_vsyscall(&xtime, clock); | |
175 | ||
176 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
177 | ||
178 | /* signal hrtimers about time change */ | |
179 | clock_was_set(); | |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
184 | EXPORT_SYMBOL(do_settimeofday); | |
185 | ||
186 | /** | |
187 | * change_clocksource - Swaps clocksources if a new one is available | |
188 | * | |
189 | * Accumulates current time interval and initializes new clocksource | |
190 | */ | |
191 | static void change_clocksource(void) | |
192 | { | |
193 | struct clocksource *new; | |
194 | cycle_t now; | |
195 | u64 nsec; | |
196 | ||
197 | new = clocksource_get_next(); | |
198 | ||
199 | if (clock == new) | |
200 | return; | |
201 | ||
202 | now = clocksource_read(new); | |
203 | nsec = __get_nsec_offset(); | |
204 | timespec_add_ns(&xtime, nsec); | |
205 | ||
206 | clock = new; | |
207 | clock->cycle_last = now; | |
208 | ||
209 | clock->error = 0; | |
210 | clock->xtime_nsec = 0; | |
211 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | |
212 | ||
213 | tick_clock_notify(); | |
214 | ||
215 | printk(KERN_INFO "Time: %s clocksource has been installed.\n", | |
216 | clock->name); | |
217 | } | |
218 | #else | |
219 | static inline void change_clocksource(void) { } | |
220 | #endif | |
221 | ||
222 | /** | |
223 | * timekeeping_is_continuous - check to see if timekeeping is free running | |
224 | */ | |
225 | int timekeeping_is_continuous(void) | |
226 | { | |
227 | unsigned long seq; | |
228 | int ret; | |
229 | ||
230 | do { | |
231 | seq = read_seqbegin(&xtime_lock); | |
232 | ||
233 | ret = clock->flags & CLOCK_SOURCE_VALID_FOR_HRES; | |
234 | ||
235 | } while (read_seqretry(&xtime_lock, seq)); | |
236 | ||
237 | return ret; | |
238 | } | |
239 | ||
240 | /** | |
241 | * read_persistent_clock - Return time in seconds from the persistent clock. | |
242 | * | |
243 | * Weak dummy function for arches that do not yet support it. | |
244 | * Returns seconds from epoch using the battery backed persistent clock. | |
245 | * Returns zero if unsupported. | |
246 | * | |
247 | * XXX - Do be sure to remove it once all arches implement it. | |
248 | */ | |
249 | unsigned long __attribute__((weak)) read_persistent_clock(void) | |
250 | { | |
251 | return 0; | |
252 | } | |
253 | ||
254 | /* | |
255 | * timekeeping_init - Initializes the clocksource and common timekeeping values | |
256 | */ | |
257 | void __init timekeeping_init(void) | |
258 | { | |
259 | unsigned long flags; | |
260 | unsigned long sec = read_persistent_clock(); | |
261 | ||
262 | write_seqlock_irqsave(&xtime_lock, flags); | |
263 | ||
264 | ntp_clear(); | |
265 | ||
266 | clock = clocksource_get_next(); | |
267 | clocksource_calculate_interval(clock, NTP_INTERVAL_LENGTH); | |
268 | clock->cycle_last = clocksource_read(clock); | |
269 | ||
270 | xtime.tv_sec = sec; | |
271 | xtime.tv_nsec = 0; | |
272 | set_normalized_timespec(&wall_to_monotonic, | |
273 | -xtime.tv_sec, -xtime.tv_nsec); | |
7c3f1a57 | 274 | total_sleep_time = 0; |
8524070b JS |
275 | |
276 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
277 | } | |
278 | ||
279 | /* flag for if timekeeping is suspended */ | |
280 | static int timekeeping_suspended; | |
281 | /* time in seconds when suspend began */ | |
282 | static unsigned long timekeeping_suspend_time; | |
283 | ||
284 | /** | |
285 | * timekeeping_resume - Resumes the generic timekeeping subsystem. | |
286 | * @dev: unused | |
287 | * | |
288 | * This is for the generic clocksource timekeeping. | |
289 | * xtime/wall_to_monotonic/jiffies/etc are | |
290 | * still managed by arch specific suspend/resume code. | |
291 | */ | |
292 | static int timekeeping_resume(struct sys_device *dev) | |
293 | { | |
294 | unsigned long flags; | |
295 | unsigned long now = read_persistent_clock(); | |
296 | ||
d10ff3fb TG |
297 | clocksource_resume(); |
298 | ||
8524070b JS |
299 | write_seqlock_irqsave(&xtime_lock, flags); |
300 | ||
301 | if (now && (now > timekeeping_suspend_time)) { | |
302 | unsigned long sleep_length = now - timekeeping_suspend_time; | |
303 | ||
304 | xtime.tv_sec += sleep_length; | |
305 | wall_to_monotonic.tv_sec -= sleep_length; | |
7c3f1a57 | 306 | total_sleep_time += sleep_length; |
8524070b JS |
307 | } |
308 | /* re-base the last cycle value */ | |
309 | clock->cycle_last = clocksource_read(clock); | |
310 | clock->error = 0; | |
311 | timekeeping_suspended = 0; | |
312 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
313 | ||
314 | touch_softlockup_watchdog(); | |
315 | ||
316 | clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL); | |
317 | ||
318 | /* Resume hrtimers */ | |
319 | hres_timers_resume(); | |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
324 | static int timekeeping_suspend(struct sys_device *dev, pm_message_t state) | |
325 | { | |
326 | unsigned long flags; | |
327 | ||
328 | write_seqlock_irqsave(&xtime_lock, flags); | |
329 | timekeeping_suspended = 1; | |
330 | timekeeping_suspend_time = read_persistent_clock(); | |
331 | write_sequnlock_irqrestore(&xtime_lock, flags); | |
332 | ||
333 | clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); | |
334 | ||
335 | return 0; | |
336 | } | |
337 | ||
338 | /* sysfs resume/suspend bits for timekeeping */ | |
339 | static struct sysdev_class timekeeping_sysclass = { | |
340 | .resume = timekeeping_resume, | |
341 | .suspend = timekeeping_suspend, | |
342 | set_kset_name("timekeeping"), | |
343 | }; | |
344 | ||
345 | static struct sys_device device_timer = { | |
346 | .id = 0, | |
347 | .cls = &timekeeping_sysclass, | |
348 | }; | |
349 | ||
350 | static int __init timekeeping_init_device(void) | |
351 | { | |
352 | int error = sysdev_class_register(&timekeeping_sysclass); | |
353 | if (!error) | |
354 | error = sysdev_register(&device_timer); | |
355 | return error; | |
356 | } | |
357 | ||
358 | device_initcall(timekeeping_init_device); | |
359 | ||
360 | /* | |
361 | * If the error is already larger, we look ahead even further | |
362 | * to compensate for late or lost adjustments. | |
363 | */ | |
364 | static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, | |
365 | s64 *offset) | |
366 | { | |
367 | s64 tick_error, i; | |
368 | u32 look_ahead, adj; | |
369 | s32 error2, mult; | |
370 | ||
371 | /* | |
372 | * Use the current error value to determine how much to look ahead. | |
373 | * The larger the error the slower we adjust for it to avoid problems | |
374 | * with losing too many ticks, otherwise we would overadjust and | |
375 | * produce an even larger error. The smaller the adjustment the | |
376 | * faster we try to adjust for it, as lost ticks can do less harm | |
377 | * here. This is tuned so that an error of about 1 msec is adusted | |
378 | * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks). | |
379 | */ | |
380 | error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ); | |
381 | error2 = abs(error2); | |
382 | for (look_ahead = 0; error2 > 0; look_ahead++) | |
383 | error2 >>= 2; | |
384 | ||
385 | /* | |
386 | * Now calculate the error in (1 << look_ahead) ticks, but first | |
387 | * remove the single look ahead already included in the error. | |
388 | */ | |
389 | tick_error = current_tick_length() >> | |
390 | (TICK_LENGTH_SHIFT - clock->shift + 1); | |
391 | tick_error -= clock->xtime_interval >> 1; | |
392 | error = ((error - tick_error) >> look_ahead) + tick_error; | |
393 | ||
394 | /* Finally calculate the adjustment shift value. */ | |
395 | i = *interval; | |
396 | mult = 1; | |
397 | if (error < 0) { | |
398 | error = -error; | |
399 | *interval = -*interval; | |
400 | *offset = -*offset; | |
401 | mult = -1; | |
402 | } | |
403 | for (adj = 0; error > i; adj++) | |
404 | error >>= 1; | |
405 | ||
406 | *interval <<= adj; | |
407 | *offset <<= adj; | |
408 | return mult << adj; | |
409 | } | |
410 | ||
411 | /* | |
412 | * Adjust the multiplier to reduce the error value, | |
413 | * this is optimized for the most common adjustments of -1,0,1, | |
414 | * for other values we can do a bit more work. | |
415 | */ | |
71120f18 | 416 | static void clocksource_adjust(s64 offset) |
8524070b JS |
417 | { |
418 | s64 error, interval = clock->cycle_interval; | |
419 | int adj; | |
420 | ||
421 | error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1); | |
422 | if (error > interval) { | |
423 | error >>= 2; | |
424 | if (likely(error <= interval)) | |
425 | adj = 1; | |
426 | else | |
427 | adj = clocksource_bigadjust(error, &interval, &offset); | |
428 | } else if (error < -interval) { | |
429 | error >>= 2; | |
430 | if (likely(error >= -interval)) { | |
431 | adj = -1; | |
432 | interval = -interval; | |
433 | offset = -offset; | |
434 | } else | |
435 | adj = clocksource_bigadjust(error, &interval, &offset); | |
436 | } else | |
437 | return; | |
438 | ||
439 | clock->mult += adj; | |
440 | clock->xtime_interval += interval; | |
441 | clock->xtime_nsec -= offset; | |
442 | clock->error -= (interval - offset) << | |
443 | (TICK_LENGTH_SHIFT - clock->shift); | |
444 | } | |
445 | ||
446 | /** | |
447 | * update_wall_time - Uses the current clocksource to increment the wall time | |
448 | * | |
449 | * Called from the timer interrupt, must hold a write on xtime_lock. | |
450 | */ | |
451 | void update_wall_time(void) | |
452 | { | |
453 | cycle_t offset; | |
454 | ||
455 | /* Make sure we're fully resumed: */ | |
456 | if (unlikely(timekeeping_suspended)) | |
457 | return; | |
458 | ||
459 | #ifdef CONFIG_GENERIC_TIME | |
460 | offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask; | |
461 | #else | |
462 | offset = clock->cycle_interval; | |
463 | #endif | |
464 | clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift; | |
465 | ||
466 | /* normally this loop will run just once, however in the | |
467 | * case of lost or late ticks, it will accumulate correctly. | |
468 | */ | |
469 | while (offset >= clock->cycle_interval) { | |
470 | /* accumulate one interval */ | |
471 | clock->xtime_nsec += clock->xtime_interval; | |
472 | clock->cycle_last += clock->cycle_interval; | |
473 | offset -= clock->cycle_interval; | |
474 | ||
475 | if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) { | |
476 | clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift; | |
477 | xtime.tv_sec++; | |
478 | second_overflow(); | |
479 | } | |
480 | ||
8524070b JS |
481 | /* accumulate error between NTP and clock interval */ |
482 | clock->error += current_tick_length(); | |
483 | clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift); | |
484 | } | |
485 | ||
486 | /* correct the clock when NTP error is too big */ | |
71120f18 | 487 | clocksource_adjust(offset); |
8524070b JS |
488 | |
489 | /* store full nanoseconds into xtime */ | |
490 | xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift; | |
491 | clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift; | |
492 | ||
17c38b74 JS |
493 | update_xtime_cache(cyc2ns(clock, offset)); |
494 | ||
8524070b JS |
495 | /* check to see if there is a new clocksource to use */ |
496 | change_clocksource(); | |
497 | update_vsyscall(&xtime, clock); | |
498 | } | |
7c3f1a57 TJ |
499 | |
500 | /** | |
501 | * getboottime - Return the real time of system boot. | |
502 | * @ts: pointer to the timespec to be set | |
503 | * | |
504 | * Returns the time of day in a timespec. | |
505 | * | |
506 | * This is based on the wall_to_monotonic offset and the total suspend | |
507 | * time. Calls to settimeofday will affect the value returned (which | |
508 | * basically means that however wrong your real time clock is at boot time, | |
509 | * you get the right time here). | |
510 | */ | |
511 | void getboottime(struct timespec *ts) | |
512 | { | |
513 | set_normalized_timespec(ts, | |
514 | - (wall_to_monotonic.tv_sec + total_sleep_time), | |
515 | - wall_to_monotonic.tv_nsec); | |
516 | } | |
517 | ||
518 | /** | |
519 | * monotonic_to_bootbased - Convert the monotonic time to boot based. | |
520 | * @ts: pointer to the timespec to be converted | |
521 | */ | |
522 | void monotonic_to_bootbased(struct timespec *ts) | |
523 | { | |
524 | ts->tv_sec += total_sleep_time; | |
525 | } | |
2c6b47de | 526 | |
17c38b74 JS |
527 | unsigned long get_seconds(void) |
528 | { | |
529 | return xtime_cache.tv_sec; | |
530 | } | |
531 | EXPORT_SYMBOL(get_seconds); | |
532 | ||
533 | ||
2c6b47de JS |
534 | struct timespec current_kernel_time(void) |
535 | { | |
536 | struct timespec now; | |
537 | unsigned long seq; | |
538 | ||
539 | do { | |
540 | seq = read_seqbegin(&xtime_lock); | |
541 | ||
17c38b74 | 542 | now = xtime_cache; |
2c6b47de JS |
543 | } while (read_seqretry(&xtime_lock, seq)); |
544 | ||
545 | return now; | |
546 | } | |
2c6b47de | 547 | EXPORT_SYMBOL(current_kernel_time); |