~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/time/timekeeping.c

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/timekeeper_internal.h>
 12 #include <linux/module.h>
 13 #include <linux/interrupt.h>
 14 #include <linux/percpu.h>
 15 #include <linux/init.h>
 16 #include <linux/mm.h>
 17 #include <linux/sched.h>
 18 #include <linux/syscore_ops.h>
 19 #include <linux/clocksource.h>
 20 #include <linux/jiffies.h>
 21 #include <linux/time.h>
 22 #include <linux/tick.h>
 23 #include <linux/stop_machine.h>
 24 #include <linux/pvclock_gtod.h>
 25 
 26 #include "tick-internal.h"
 27 #include "ntp_internal.h"
 28 #include "timekeeping_internal.h"
 29 
 30 #define TK_CLEAR_NTP            (1 << 0)
 31 #define TK_MIRROR               (1 << 1)
 32 #define TK_CLOCK_WAS_SET        (1 << 2)
 33 
 34 static struct timekeeper timekeeper;
 35 static DEFINE_RAW_SPINLOCK(timekeeper_lock);
 36 static seqcount_t timekeeper_seq;
 37 static struct timekeeper shadow_timekeeper;
 38 
 39 /* flag for if timekeeping is suspended */
 40 int __read_mostly timekeeping_suspended;
 41 
 42 /* Flag for if there is a persistent clock on this platform */
 43 bool __read_mostly persistent_clock_exist = false;
 44 
 45 static inline void tk_normalize_xtime(struct timekeeper *tk)
 46 {
 47         while (tk->xtime_nsec >= ((u64)NSEC_PER_SEC << tk->shift)) {
 48                 tk->xtime_nsec -= (u64)NSEC_PER_SEC << tk->shift;
 49                 tk->xtime_sec++;
 50         }
 51 }
 52 
 53 static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
 54 {
 55         tk->xtime_sec = ts->tv_sec;
 56         tk->xtime_nsec = (u64)ts->tv_nsec << tk->shift;
 57 }
 58 
 59 static void tk_xtime_add(struct timekeeper *tk, const struct timespec *ts)
 60 {
 61         tk->xtime_sec += ts->tv_sec;
 62         tk->xtime_nsec += (u64)ts->tv_nsec << tk->shift;
 63         tk_normalize_xtime(tk);
 64 }
 65 
 66 static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec wtm)
 67 {
 68         struct timespec tmp;
 69 
 70         /*
 71          * Verify consistency of: offset_real = -wall_to_monotonic
 72          * before modifying anything
 73          */
 74         set_normalized_timespec(&tmp, -tk->wall_to_monotonic.tv_sec,
 75                                         -tk->wall_to_monotonic.tv_nsec);
 76         WARN_ON_ONCE(tk->offs_real.tv64 != timespec_to_ktime(tmp).tv64);
 77         tk->wall_to_monotonic = wtm;
 78         set_normalized_timespec(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
 79         tk->offs_real = timespec_to_ktime(tmp);
 80         tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
 81 }
 82 
 83 static void tk_set_sleep_time(struct timekeeper *tk, struct timespec t)
 84 {
 85         /* Verify consistency before modifying */
 86         WARN_ON_ONCE(tk->offs_boot.tv64 != timespec_to_ktime(tk->total_sleep_time).tv64);
 87 
 88         tk->total_sleep_time    = t;
 89         tk->offs_boot           = timespec_to_ktime(t);
 90 }
 91 
 92 /**
 93  * tk_setup_internals - Set up internals to use clocksource clock.
 94  *
 95  * @tk:         The target timekeeper to setup.
 96  * @clock:              Pointer to clocksource.
 97  *
 98  * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
 99  * pair and interval request.
100  *
101  * Unless you're the timekeeping code, you should not be using this!
102  */
103 static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
104 {
105         cycle_t interval;
106         u64 tmp, ntpinterval;
107         struct clocksource *old_clock;
108 
109         old_clock = tk->clock;
110         tk->clock = clock;
111         tk->cycle_last = clock->cycle_last = clock->read(clock);
112 
113         /* Do the ns -> cycle conversion first, using original mult */
114         tmp = NTP_INTERVAL_LENGTH;
115         tmp <<= clock->shift;
116         ntpinterval = tmp;
117         tmp += clock->mult/2;
118         do_div(tmp, clock->mult);
119         if (tmp == 0)
120                 tmp = 1;
121 
122         interval = (cycle_t) tmp;
123         tk->cycle_interval = interval;
124 
125         /* Go back from cycles -> shifted ns */
126         tk->xtime_interval = (u64) interval * clock->mult;
127         tk->xtime_remainder = ntpinterval - tk->xtime_interval;
128         tk->raw_interval =
129                 ((u64) interval * clock->mult) >> clock->shift;
130 
131          /* if changing clocks, convert xtime_nsec shift units */
132         if (old_clock) {
133                 int shift_change = clock->shift - old_clock->shift;
134                 if (shift_change < 0)
135                         tk->xtime_nsec >>= -shift_change;
136                 else
137                         tk->xtime_nsec <<= shift_change;
138         }
139         tk->shift = clock->shift;
140 
141         tk->ntp_error = 0;
142         tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
143 
144         /*
145          * The timekeeper keeps its own mult values for the currently
146          * active clocksource. These value will be adjusted via NTP
147          * to counteract clock drifting.
148          */
149         tk->mult = clock->mult;
150 }
151 
152 /* Timekeeper helper functions. */
153 
154 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
155 u32 (*arch_gettimeoffset)(void);
156 
157 u32 get_arch_timeoffset(void)
158 {
159         if (likely(arch_gettimeoffset))
160                 return arch_gettimeoffset();
161         return 0;
162 }
163 #else
164 static inline u32 get_arch_timeoffset(void) { return 0; }
165 #endif
166 
167 static inline s64 timekeeping_get_ns(struct timekeeper *tk)
168 {
169         cycle_t cycle_now, cycle_delta;
170         struct clocksource *clock;
171         s64 nsec;
172 
173         /* read clocksource: */
174         clock = tk->clock;
175         cycle_now = clock->read(clock);
176 
177         /* calculate the delta since the last update_wall_time: */
178         cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
179 
180         nsec = cycle_delta * tk->mult + tk->xtime_nsec;
181         nsec >>= tk->shift;
182 
183         /* If arch requires, add in get_arch_timeoffset() */
184         return nsec + get_arch_timeoffset();
185 }
186 
187 static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
188 {
189         cycle_t cycle_now, cycle_delta;
190         struct clocksource *clock;
191         s64 nsec;
192 
193         /* read clocksource: */
194         clock = tk->clock;
195         cycle_now = clock->read(clock);
196 
197         /* calculate the delta since the last update_wall_time: */
198         cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
199 
200         /* convert delta to nanoseconds. */
201         nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
202 
203         /* If arch requires, add in get_arch_timeoffset() */
204         return nsec + get_arch_timeoffset();
205 }
206 
207 static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
208 
209 static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
210 {
211         raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
212 }
213 
214 /**
215  * pvclock_gtod_register_notifier - register a pvclock timedata update listener
216  */
217 int pvclock_gtod_register_notifier(struct notifier_block *nb)
218 {
219         struct timekeeper *tk = &timekeeper;
220         unsigned long flags;
221         int ret;
222 
223         raw_spin_lock_irqsave(&timekeeper_lock, flags);
224         ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
225         update_pvclock_gtod(tk, true);
226         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
227 
228         return ret;
229 }
230 EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
231 
232 /**
233  * pvclock_gtod_unregister_notifier - unregister a pvclock
234  * timedata update listener
235  */
236 int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
237 {
238         unsigned long flags;
239         int ret;
240 
241         raw_spin_lock_irqsave(&timekeeper_lock, flags);
242         ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
243         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
244 
245         return ret;
246 }
247 EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
248 
249 /* must hold timekeeper_lock */
250 static void timekeeping_update(struct timekeeper *tk, unsigned int action)
251 {
252         if (action & TK_CLEAR_NTP) {
253                 tk->ntp_error = 0;
254                 ntp_clear();
255         }
256         update_vsyscall(tk);
257         update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
258 
259         if (action & TK_MIRROR)
260                 memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
261 }
262 
263 /**
264  * timekeeping_forward_now - update clock to the current time
265  *
266  * Forward the current clock to update its state since the last call to
267  * update_wall_time(). This is useful before significant clock changes,
268  * as it avoids having to deal with this time offset explicitly.
269  */
270 static void timekeeping_forward_now(struct timekeeper *tk)
271 {
272         cycle_t cycle_now, cycle_delta;
273         struct clocksource *clock;
274         s64 nsec;
275 
276         clock = tk->clock;
277         cycle_now = clock->read(clock);
278         cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
279         tk->cycle_last = clock->cycle_last = cycle_now;
280 
281         tk->xtime_nsec += cycle_delta * tk->mult;
282 
283         /* If arch requires, add in get_arch_timeoffset() */
284         tk->xtime_nsec += (u64)get_arch_timeoffset() << tk->shift;
285 
286         tk_normalize_xtime(tk);
287 
288         nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
289         timespec_add_ns(&tk->raw_time, nsec);
290 }
291 
292 /**
293  * __getnstimeofday - Returns the time of day in a timespec.
294  * @ts:         pointer to the timespec to be set
295  *
296  * Updates the time of day in the timespec.
297  * Returns 0 on success, or -ve when suspended (timespec will be undefined).
298  */
299 int __getnstimeofday(struct timespec *ts)
300 {
301         struct timekeeper *tk = &timekeeper;
302         unsigned long seq;
303         s64 nsecs = 0;
304 
305         do {
306                 seq = read_seqcount_begin(&timekeeper_seq);
307 
308                 ts->tv_sec = tk->xtime_sec;
309                 nsecs = timekeeping_get_ns(tk);
310 
311         } while (read_seqcount_retry(&timekeeper_seq, seq));
312 
313         ts->tv_nsec = 0;
314         timespec_add_ns(ts, nsecs);
315 
316         /*
317          * Do not bail out early, in case there were callers still using
318          * the value, even in the face of the WARN_ON.
319          */
320         if (unlikely(timekeeping_suspended))
321                 return -EAGAIN;
322         return 0;
323 }
324 EXPORT_SYMBOL(__getnstimeofday);
325 
326 /**
327  * getnstimeofday - Returns the time of day in a timespec.
328  * @ts:         pointer to the timespec to be set
329  *
330  * Returns the time of day in a timespec (WARN if suspended).
331  */
332 void getnstimeofday(struct timespec *ts)
333 {
334         WARN_ON(__getnstimeofday(ts));
335 }
336 EXPORT_SYMBOL(getnstimeofday);
337 
338 ktime_t ktime_get(void)
339 {
340         struct timekeeper *tk = &timekeeper;
341         unsigned int seq;
342         s64 secs, nsecs;
343 
344         WARN_ON(timekeeping_suspended);
345 
346         do {
347                 seq = read_seqcount_begin(&timekeeper_seq);
348                 secs = tk->xtime_sec + tk->wall_to_monotonic.tv_sec;
349                 nsecs = timekeeping_get_ns(tk) + tk->wall_to_monotonic.tv_nsec;
350 
351         } while (read_seqcount_retry(&timekeeper_seq, seq));
352         /*
353          * Use ktime_set/ktime_add_ns to create a proper ktime on
354          * 32-bit architectures without CONFIG_KTIME_SCALAR.
355          */
356         return ktime_add_ns(ktime_set(secs, 0), nsecs);
357 }
358 EXPORT_SYMBOL_GPL(ktime_get);
359 
360 /**
361  * ktime_get_ts - get the monotonic clock in timespec format
362  * @ts:         pointer to timespec variable
363  *
364  * The function calculates the monotonic clock from the realtime
365  * clock and the wall_to_monotonic offset and stores the result
366  * in normalized timespec format in the variable pointed to by @ts.
367  */
368 void ktime_get_ts(struct timespec *ts)
369 {
370         struct timekeeper *tk = &timekeeper;
371         struct timespec tomono;
372         s64 nsec;
373         unsigned int seq;
374 
375         WARN_ON(timekeeping_suspended);
376 
377         do {
378                 seq = read_seqcount_begin(&timekeeper_seq);
379                 ts->tv_sec = tk->xtime_sec;
380                 nsec = timekeeping_get_ns(tk);
381                 tomono = tk->wall_to_monotonic;
382 
383         } while (read_seqcount_retry(&timekeeper_seq, seq));
384 
385         ts->tv_sec += tomono.tv_sec;
386         ts->tv_nsec = 0;
387         timespec_add_ns(ts, nsec + tomono.tv_nsec);
388 }
389 EXPORT_SYMBOL_GPL(ktime_get_ts);
390 
391 
392 /**
393  * timekeeping_clocktai - Returns the TAI time of day in a timespec
394  * @ts:         pointer to the timespec to be set
395  *
396  * Returns the time of day in a timespec.
397  */
398 void timekeeping_clocktai(struct timespec *ts)
399 {
400         struct timekeeper *tk = &timekeeper;
401         unsigned long seq;
402         u64 nsecs;
403 
404         WARN_ON(timekeeping_suspended);
405 
406         do {
407                 seq = read_seqcount_begin(&timekeeper_seq);
408 
409                 ts->tv_sec = tk->xtime_sec + tk->tai_offset;
410                 nsecs = timekeeping_get_ns(tk);
411 
412         } while (read_seqcount_retry(&timekeeper_seq, seq));
413 
414         ts->tv_nsec = 0;
415         timespec_add_ns(ts, nsecs);
416 
417 }
418 EXPORT_SYMBOL(timekeeping_clocktai);
419 
420 
421 /**
422  * ktime_get_clocktai - Returns the TAI time of day in a ktime
423  *
424  * Returns the time of day in a ktime.
425  */
426 ktime_t ktime_get_clocktai(void)
427 {
428         struct timespec ts;
429 
430         timekeeping_clocktai(&ts);
431         return timespec_to_ktime(ts);
432 }
433 EXPORT_SYMBOL(ktime_get_clocktai);
434 
435 #ifdef CONFIG_NTP_PPS
436 
437 /**
438  * getnstime_raw_and_real - get day and raw monotonic time in timespec format
439  * @ts_raw:     pointer to the timespec to be set to raw monotonic time
440  * @ts_real:    pointer to the timespec to be set to the time of day
441  *
442  * This function reads both the time of day and raw monotonic time at the
443  * same time atomically and stores the resulting timestamps in timespec
444  * format.
445  */
446 void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
447 {
448         struct timekeeper *tk = &timekeeper;
449         unsigned long seq;
450         s64 nsecs_raw, nsecs_real;
451 
452         WARN_ON_ONCE(timekeeping_suspended);
453 
454         do {
455                 seq = read_seqcount_begin(&timekeeper_seq);
456 
457                 *ts_raw = tk->raw_time;
458                 ts_real->tv_sec = tk->xtime_sec;
459                 ts_real->tv_nsec = 0;
460 
461                 nsecs_raw = timekeeping_get_ns_raw(tk);
462                 nsecs_real = timekeeping_get_ns(tk);
463 
464         } while (read_seqcount_retry(&timekeeper_seq, seq));
465 
466         timespec_add_ns(ts_raw, nsecs_raw);
467         timespec_add_ns(ts_real, nsecs_real);
468 }
469 EXPORT_SYMBOL(getnstime_raw_and_real);
470 
471 #endif /* CONFIG_NTP_PPS */
472 
473 /**
474  * do_gettimeofday - Returns the time of day in a timeval
475  * @tv:         pointer to the timeval to be set
476  *
477  * NOTE: Users should be converted to using getnstimeofday()
478  */
479 void do_gettimeofday(struct timeval *tv)
480 {
481         struct timespec now;
482 
483         getnstimeofday(&now);
484         tv->tv_sec = now.tv_sec;
485         tv->tv_usec = now.tv_nsec/1000;
486 }
487 EXPORT_SYMBOL(do_gettimeofday);
488 
489 /**
490  * do_settimeofday - Sets the time of day
491  * @tv:         pointer to the timespec variable containing the new time
492  *
493  * Sets the time of day to the new time and update NTP and notify hrtimers
494  */
495 int do_settimeofday(const struct timespec *tv)
496 {
497         struct timekeeper *tk = &timekeeper;
498         struct timespec ts_delta, xt;
499         unsigned long flags;
500 
501         if (!timespec_valid_strict(tv))
502                 return -EINVAL;
503 
504         raw_spin_lock_irqsave(&timekeeper_lock, flags);
505         write_seqcount_begin(&timekeeper_seq);
506 
507         timekeeping_forward_now(tk);
508 
509         xt = tk_xtime(tk);
510         ts_delta.tv_sec = tv->tv_sec - xt.tv_sec;
511         ts_delta.tv_nsec = tv->tv_nsec - xt.tv_nsec;
512 
513         tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, ts_delta));
514 
515         tk_set_xtime(tk, tv);
516 
517         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
518 
519         write_seqcount_end(&timekeeper_seq);
520         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
521 
522         /* signal hrtimers about time change */
523         clock_was_set();
524 
525         return 0;
526 }
527 EXPORT_SYMBOL(do_settimeofday);
528 
529 /**
530  * timekeeping_inject_offset - Adds or subtracts from the current time.
531  * @tv:         pointer to the timespec variable containing the offset
532  *
533  * Adds or subtracts an offset value from the current time.
534  */
535 int timekeeping_inject_offset(struct timespec *ts)
536 {
537         struct timekeeper *tk = &timekeeper;
538         unsigned long flags;
539         struct timespec tmp;
540         int ret = 0;
541 
542         if ((unsigned long)ts->tv_nsec >= NSEC_PER_SEC)
543                 return -EINVAL;
544 
545         raw_spin_lock_irqsave(&timekeeper_lock, flags);
546         write_seqcount_begin(&timekeeper_seq);
547 
548         timekeeping_forward_now(tk);
549 
550         /* Make sure the proposed value is valid */
551         tmp = timespec_add(tk_xtime(tk),  *ts);
552         if (!timespec_valid_strict(&tmp)) {
553                 ret = -EINVAL;
554                 goto error;
555         }
556 
557         tk_xtime_add(tk, ts);
558         tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *ts));
559 
560 error: /* even if we error out, we forwarded the time, so call update */
561         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
562 
563         write_seqcount_end(&timekeeper_seq);
564         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
565 
566         /* signal hrtimers about time change */
567         clock_was_set();
568 
569         return ret;
570 }
571 EXPORT_SYMBOL(timekeeping_inject_offset);
572 
573 
574 /**
575  * timekeeping_get_tai_offset - Returns current TAI offset from UTC
576  *
577  */
578 s32 timekeeping_get_tai_offset(void)
579 {
580         struct timekeeper *tk = &timekeeper;
581         unsigned int seq;
582         s32 ret;
583 
584         do {
585                 seq = read_seqcount_begin(&timekeeper_seq);
586                 ret = tk->tai_offset;
587         } while (read_seqcount_retry(&timekeeper_seq, seq));
588 
589         return ret;
590 }
591 
592 /**
593  * __timekeeping_set_tai_offset - Lock free worker function
594  *
595  */
596 static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
597 {
598         tk->tai_offset = tai_offset;
599         tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
600 }
601 
602 /**
603  * timekeeping_set_tai_offset - Sets the current TAI offset from UTC
604  *
605  */
606 void timekeeping_set_tai_offset(s32 tai_offset)
607 {
608         struct timekeeper *tk = &timekeeper;
609         unsigned long flags;
610 
611         raw_spin_lock_irqsave(&timekeeper_lock, flags);
612         write_seqcount_begin(&timekeeper_seq);
613         __timekeeping_set_tai_offset(tk, tai_offset);
614         timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
615         write_seqcount_end(&timekeeper_seq);
616         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
617         clock_was_set();
618 }
619 
620 /**
621  * change_clocksource - Swaps clocksources if a new one is available
622  *
623  * Accumulates current time interval and initializes new clocksource
624  */
625 static int change_clocksource(void *data)
626 {
627         struct timekeeper *tk = &timekeeper;
628         struct clocksource *new, *old;
629         unsigned long flags;
630 
631         new = (struct clocksource *) data;
632 
633         raw_spin_lock_irqsave(&timekeeper_lock, flags);
634         write_seqcount_begin(&timekeeper_seq);
635 
636         timekeeping_forward_now(tk);
637         /*
638          * If the cs is in module, get a module reference. Succeeds
639          * for built-in code (owner == NULL) as well.
640          */
641         if (try_module_get(new->owner)) {
642                 if (!new->enable || new->enable(new) == 0) {
643                         old = tk->clock;
644                         tk_setup_internals(tk, new);
645                         if (old->disable)
646                                 old->disable(old);
647                         module_put(old->owner);
648                 } else {
649                         module_put(new->owner);
650                 }
651         }
652         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
653 
654         write_seqcount_end(&timekeeper_seq);
655         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
656 
657         return 0;
658 }
659 
660 /**
661  * timekeeping_notify - Install a new clock source
662  * @clock:              pointer to the clock source
663  *
664  * This function is called from clocksource.c after a new, better clock
665  * source has been registered. The caller holds the clocksource_mutex.
666  */
667 int timekeeping_notify(struct clocksource *clock)
668 {
669         struct timekeeper *tk = &timekeeper;
670 
671         if (tk->clock == clock)
672                 return 0;
673         stop_machine(change_clocksource, clock, NULL);
674         tick_clock_notify();
675         return tk->clock == clock ? 0 : -1;
676 }
677 
678 /**
679  * ktime_get_real - get the real (wall-) time in ktime_t format
680  *
681  * returns the time in ktime_t format
682  */
683 ktime_t ktime_get_real(void)
684 {
685         struct timespec now;
686 
687         getnstimeofday(&now);
688 
689         return timespec_to_ktime(now);
690 }
691 EXPORT_SYMBOL_GPL(ktime_get_real);
692 
693 /**
694  * getrawmonotonic - Returns the raw monotonic time in a timespec
695  * @ts:         pointer to the timespec to be set
696  *
697  * Returns the raw monotonic time (completely un-modified by ntp)
698  */
699 void getrawmonotonic(struct timespec *ts)
700 {
701         struct timekeeper *tk = &timekeeper;
702         unsigned long seq;
703         s64 nsecs;
704 
705         do {
706                 seq = read_seqcount_begin(&timekeeper_seq);
707                 nsecs = timekeeping_get_ns_raw(tk);
708                 *ts = tk->raw_time;
709 
710         } while (read_seqcount_retry(&timekeeper_seq, seq));
711 
712         timespec_add_ns(ts, nsecs);
713 }
714 EXPORT_SYMBOL(getrawmonotonic);
715 
716 /**
717  * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
718  */
719 int timekeeping_valid_for_hres(void)
720 {
721         struct timekeeper *tk = &timekeeper;
722         unsigned long seq;
723         int ret;
724 
725         do {
726                 seq = read_seqcount_begin(&timekeeper_seq);
727 
728                 ret = tk->clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
729 
730         } while (read_seqcount_retry(&timekeeper_seq, seq));
731 
732         return ret;
733 }
734 
735 /**
736  * timekeeping_max_deferment - Returns max time the clocksource can be deferred
737  */
738 u64 timekeeping_max_deferment(void)
739 {
740         struct timekeeper *tk = &timekeeper;
741         unsigned long seq;
742         u64 ret;
743 
744         do {
745                 seq = read_seqcount_begin(&timekeeper_seq);
746 
747                 ret = tk->clock->max_idle_ns;
748 
749         } while (read_seqcount_retry(&timekeeper_seq, seq));
750 
751         return ret;
752 }
753 
754 /**
755  * read_persistent_clock -  Return time from the persistent clock.
756  *
757  * Weak dummy function for arches that do not yet support it.
758  * Reads the time from the battery backed persistent clock.
759  * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
760  *
761  *  XXX - Do be sure to remove it once all arches implement it.
762  */
763 void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
764 {
765         ts->tv_sec = 0;
766         ts->tv_nsec = 0;
767 }
768 
769 /**
770  * read_boot_clock -  Return time of the system start.
771  *
772  * Weak dummy function for arches that do not yet support it.
773  * Function to read the exact time the system has been started.
774  * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
775  *
776  *  XXX - Do be sure to remove it once all arches implement it.
777  */
778 void __attribute__((weak)) read_boot_clock(struct timespec *ts)
779 {
780         ts->tv_sec = 0;
781         ts->tv_nsec = 0;
782 }
783 
784 /*
785  * timekeeping_init - Initializes the clocksource and common timekeeping values
786  */
787 void __init timekeeping_init(void)
788 {
789         struct timekeeper *tk = &timekeeper;
790         struct clocksource *clock;
791         unsigned long flags;
792         struct timespec now, boot, tmp;
793 
794         read_persistent_clock(&now);
795 
796         if (!timespec_valid_strict(&now)) {
797                 pr_warn("WARNING: Persistent clock returned invalid value!\n"
798                         "         Check your CMOS/BIOS settings.\n");
799                 now.tv_sec = 0;
800                 now.tv_nsec = 0;
801         } else if (now.tv_sec || now.tv_nsec)
802                 persistent_clock_exist = true;
803 
804         read_boot_clock(&boot);
805         if (!timespec_valid_strict(&boot)) {
806                 pr_warn("WARNING: Boot clock returned invalid value!\n"
807                         "         Check your CMOS/BIOS settings.\n");
808                 boot.tv_sec = 0;
809                 boot.tv_nsec = 0;
810         }
811 
812         raw_spin_lock_irqsave(&timekeeper_lock, flags);
813         write_seqcount_begin(&timekeeper_seq);
814         ntp_init();
815 
816         clock = clocksource_default_clock();
817         if (clock->enable)
818                 clock->enable(clock);
819         tk_setup_internals(tk, clock);
820 
821         tk_set_xtime(tk, &now);
822         tk->raw_time.tv_sec = 0;
823         tk->raw_time.tv_nsec = 0;
824         if (boot.tv_sec == 0 && boot.tv_nsec == 0)
825                 boot = tk_xtime(tk);
826 
827         set_normalized_timespec(&tmp, -boot.tv_sec, -boot.tv_nsec);
828         tk_set_wall_to_mono(tk, tmp);
829 
830         tmp.tv_sec = 0;
831         tmp.tv_nsec = 0;
832         tk_set_sleep_time(tk, tmp);
833 
834         memcpy(&shadow_timekeeper, &timekeeper, sizeof(timekeeper));
835 
836         write_seqcount_end(&timekeeper_seq);
837         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
838 }
839 
840 /* time in seconds when suspend began */
841 static struct timespec timekeeping_suspend_time;
842 
843 /**
844  * __timekeeping_inject_sleeptime - Internal function to add sleep interval
845  * @delta: pointer to a timespec delta value
846  *
847  * Takes a timespec offset measuring a suspend interval and properly
848  * adds the sleep offset to the timekeeping variables.
849  */
850 static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
851                                                         struct timespec *delta)
852 {
853         if (!timespec_valid_strict(delta)) {
854                 printk(KERN_WARNING "__timekeeping_inject_sleeptime: Invalid "
855                                         "sleep delta value!\n");
856                 return;
857         }
858         tk_xtime_add(tk, delta);
859         tk_set_wall_to_mono(tk, timespec_sub(tk->wall_to_monotonic, *delta));
860         tk_set_sleep_time(tk, timespec_add(tk->total_sleep_time, *delta));
861         tk_debug_account_sleep_time(delta);
862 }
863 
864 /**
865  * timekeeping_inject_sleeptime - Adds suspend interval to timeekeeping values
866  * @delta: pointer to a timespec delta value
867  *
868  * This hook is for architectures that cannot support read_persistent_clock
869  * because their RTC/persistent clock is only accessible when irqs are enabled.
870  *
871  * This function should only be called by rtc_resume(), and allows
872  * a suspend offset to be injected into the timekeeping values.
873  */
874 void timekeeping_inject_sleeptime(struct timespec *delta)
875 {
876         struct timekeeper *tk = &timekeeper;
877         unsigned long flags;
878 
879         /*
880          * Make sure we don't set the clock twice, as timekeeping_resume()
881          * already did it
882          */
883         if (has_persistent_clock())
884                 return;
885 
886         raw_spin_lock_irqsave(&timekeeper_lock, flags);
887         write_seqcount_begin(&timekeeper_seq);
888 
889         timekeeping_forward_now(tk);
890 
891         __timekeeping_inject_sleeptime(tk, delta);
892 
893         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
894 
895         write_seqcount_end(&timekeeper_seq);
896         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
897 
898         /* signal hrtimers about time change */
899         clock_was_set();
900 }
901 
902 /**
903  * timekeeping_resume - Resumes the generic timekeeping subsystem.
904  *
905  * This is for the generic clocksource timekeeping.
906  * xtime/wall_to_monotonic/jiffies/etc are
907  * still managed by arch specific suspend/resume code.
908  */
909 static void timekeeping_resume(void)
910 {
911         struct timekeeper *tk = &timekeeper;
912         struct clocksource *clock = tk->clock;
913         unsigned long flags;
914         struct timespec ts_new, ts_delta;
915         cycle_t cycle_now, cycle_delta;
916         bool suspendtime_found = false;
917 
918         read_persistent_clock(&ts_new);
919 
920         clockevents_resume();
921         clocksource_resume();
922 
923         raw_spin_lock_irqsave(&timekeeper_lock, flags);
924         write_seqcount_begin(&timekeeper_seq);
925 
926         /*
927          * After system resumes, we need to calculate the suspended time and
928          * compensate it for the OS time. There are 3 sources that could be
929          * used: Nonstop clocksource during suspend, persistent clock and rtc
930          * device.
931          *
932          * One specific platform may have 1 or 2 or all of them, and the
933          * preference will be:
934          *      suspend-nonstop clocksource -> persistent clock -> rtc
935          * The less preferred source will only be tried if there is no better
936          * usable source. The rtc part is handled separately in rtc core code.
937          */
938         cycle_now = clock->read(clock);
939         if ((clock->flags & CLOCK_SOURCE_SUSPEND_NONSTOP) &&
940                 cycle_now > clock->cycle_last) {
941                 u64 num, max = ULLONG_MAX;
942                 u32 mult = clock->mult;
943                 u32 shift = clock->shift;
944                 s64 nsec = 0;
945 
946                 cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
947 
948                 /*
949                  * "cycle_delta * mutl" may cause 64 bits overflow, if the
950                  * suspended time is too long. In that case we need do the
951                  * 64 bits math carefully
952                  */
953                 do_div(max, mult);
954                 if (cycle_delta > max) {
955                         num = div64_u64(cycle_delta, max);
956                         nsec = (((u64) max * mult) >> shift) * num;
957                         cycle_delta -= num * max;
958                 }
959                 nsec += ((u64) cycle_delta * mult) >> shift;
960 
961                 ts_delta = ns_to_timespec(nsec);
962                 suspendtime_found = true;
963         } else if (timespec_compare(&ts_new, &timekeeping_suspend_time) > 0) {
964                 ts_delta = timespec_sub(ts_new, timekeeping_suspend_time);
965                 suspendtime_found = true;
966         }
967 
968         if (suspendtime_found)
969                 __timekeeping_inject_sleeptime(tk, &ts_delta);
970 
971         /* Re-base the last cycle value */
972         tk->cycle_last = clock->cycle_last = cycle_now;
973         tk->ntp_error = 0;
974         timekeeping_suspended = 0;
975         timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
976         write_seqcount_end(&timekeeper_seq);
977         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
978 
979         touch_softlockup_watchdog();
980 
981         clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
982 
983         /* Resume hrtimers */
984         hrtimers_resume();
985 }
986 
987 static int timekeeping_suspend(void)
988 {
989         struct timekeeper *tk = &timekeeper;
990         unsigned long flags;
991         struct timespec         delta, delta_delta;
992         static struct timespec  old_delta;
993 
994         read_persistent_clock(&timekeeping_suspend_time);
995 
996         /*
997          * On some systems the persistent_clock can not be detected at
998          * timekeeping_init by its return value, so if we see a valid
999          * value returned, update the persistent_clock_exists flag.
1000          */
1001         if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
1002                 persistent_clock_exist = true;
1003 
1004         raw_spin_lock_irqsave(&timekeeper_lock, flags);
1005         write_seqcount_begin(&timekeeper_seq);
1006         timekeeping_forward_now(tk);
1007         timekeeping_suspended = 1;
1008 
1009         /*
1010          * To avoid drift caused by repeated suspend/resumes,
1011          * which each can add ~1 second drift error,
1012          * try to compensate so the difference in system time
1013          * and persistent_clock time stays close to constant.
1014          */
1015         delta = timespec_sub(tk_xtime(tk), timekeeping_suspend_time);
1016         delta_delta = timespec_sub(delta, old_delta);
1017         if (abs(delta_delta.tv_sec)  >= 2) {
1018                 /*
1019                  * if delta_delta is too large, assume time correction
1020                  * has occured and set old_delta to the current delta.
1021                  */
1022                 old_delta = delta;
1023         } else {
1024                 /* Otherwise try to adjust old_system to compensate */
1025                 timekeeping_suspend_time =
1026                         timespec_add(timekeeping_suspend_time, delta_delta);
1027         }
1028 
1029         timekeeping_update(tk, TK_MIRROR);
1030         write_seqcount_end(&timekeeper_seq);
1031         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1032 
1033         clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
1034         clocksource_suspend();
1035         clockevents_suspend();
1036 
1037         return 0;
1038 }
1039 
1040 /* sysfs resume/suspend bits for timekeeping */
1041 static struct syscore_ops timekeeping_syscore_ops = {
1042         .resume         = timekeeping_resume,
1043         .suspend        = timekeeping_suspend,
1044 };
1045 
1046 static int __init timekeeping_init_ops(void)
1047 {
1048         register_syscore_ops(&timekeeping_syscore_ops);
1049         return 0;
1050 }
1051 
1052 device_initcall(timekeeping_init_ops);
1053 
1054 /*
1055  * If the error is already larger, we look ahead even further
1056  * to compensate for late or lost adjustments.
1057  */
1058 static __always_inline int timekeeping_bigadjust(struct timekeeper *tk,
1059                                                  s64 error, s64 *interval,
1060                                                  s64 *offset)
1061 {
1062         s64 tick_error, i;
1063         u32 look_ahead, adj;
1064         s32 error2, mult;
1065 
1066         /*
1067          * Use the current error value to determine how much to look ahead.
1068          * The larger the error the slower we adjust for it to avoid problems
1069          * with losing too many ticks, otherwise we would overadjust and
1070          * produce an even larger error.  The smaller the adjustment the
1071          * faster we try to adjust for it, as lost ticks can do less harm
1072          * here.  This is tuned so that an error of about 1 msec is adjusted
1073          * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
1074          */
1075         error2 = tk->ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
1076         error2 = abs(error2);
1077         for (look_ahead = 0; error2 > 0; look_ahead++)
1078                 error2 >>= 2;
1079 
1080         /*
1081          * Now calculate the error in (1 << look_ahead) ticks, but first
1082          * remove the single look ahead already included in the error.
1083          */
1084         tick_error = ntp_tick_length() >> (tk->ntp_error_shift + 1);
1085         tick_error -= tk->xtime_interval >> 1;
1086         error = ((error - tick_error) >> look_ahead) + tick_error;
1087 
1088         /* Finally calculate the adjustment shift value.  */
1089         i = *interval;
1090         mult = 1;
1091         if (error < 0) {
1092                 error = -error;
1093                 *interval = -*interval;
1094                 *offset = -*offset;
1095                 mult = -1;
1096         }
1097         for (adj = 0; error > i; adj++)
1098                 error >>= 1;
1099 
1100         *interval <<= adj;
1101         *offset <<= adj;
1102         return mult << adj;
1103 }
1104 
1105 /*
1106  * Adjust the multiplier to reduce the error value,
1107  * this is optimized for the most common adjustments of -1,0,1,
1108  * for other values we can do a bit more work.
1109  */
1110 static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
1111 {
1112         s64 error, interval = tk->cycle_interval;
1113         int adj;
1114 
1115         /*
1116          * The point of this is to check if the error is greater than half
1117          * an interval.
1118          *
1119          * First we shift it down from NTP_SHIFT to clocksource->shifted nsecs.
1120          *
1121          * Note we subtract one in the shift, so that error is really error*2.
1122          * This "saves" dividing(shifting) interval twice, but keeps the
1123          * (error > interval) comparison as still measuring if error is
1124          * larger than half an interval.
1125          *
1126          * Note: It does not "save" on aggravation when reading the code.
1127          */
1128         error = tk->ntp_error >> (tk->ntp_error_shift - 1);
1129         if (error > interval) {
1130                 /*
1131                  * We now divide error by 4(via shift), which checks if
1132                  * the error is greater than twice the interval.
1133                  * If it is greater, we need a bigadjust, if its smaller,
1134                  * we can adjust by 1.
1135                  */
1136                 error >>= 2;
1137                 if (likely(error <= interval))
1138                         adj = 1;
1139                 else
1140                         adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1141         } else {
1142                 if (error < -interval) {
1143                         /* See comment above, this is just switched for the negative */
1144                         error >>= 2;
1145                         if (likely(error >= -interval)) {
1146                                 adj = -1;
1147                                 interval = -interval;
1148                                 offset = -offset;
1149                         } else {
1150                                 adj = timekeeping_bigadjust(tk, error, &interval, &offset);
1151                         }
1152                 } else {
1153                         goto out_adjust;
1154                 }
1155         }
1156 
1157         if (unlikely(tk->clock->maxadj &&
1158                 (tk->mult + adj > tk->clock->mult + tk->clock->maxadj))) {
1159                 printk_once(KERN_WARNING
1160                         "Adjusting %s more than 11%% (%ld vs %ld)\n",
1161                         tk->clock->name, (long)tk->mult + adj,
1162                         (long)tk->clock->mult + tk->clock->maxadj);
1163         }
1164         /*
1165          * So the following can be confusing.
1166          *
1167          * To keep things simple, lets assume adj == 1 for now.
1168          *
1169          * When adj != 1, remember that the interval and offset values
1170          * have been appropriately scaled so the math is the same.
1171          *
1172          * The basic idea here is that we're increasing the multiplier
1173          * by one, this causes the xtime_interval to be incremented by
1174          * one cycle_interval. This is because:
1175          *      xtime_interval = cycle_interval * mult
1176          * So if mult is being incremented by one:
1177          *      xtime_interval = cycle_interval * (mult + 1)
1178          * Its the same as:
1179          *      xtime_interval = (cycle_interval * mult) + cycle_interval
1180          * Which can be shortened to:
1181          *      xtime_interval += cycle_interval
1182          *
1183          * So offset stores the non-accumulated cycles. Thus the current
1184          * time (in shifted nanoseconds) is:
1185          *      now = (offset * adj) + xtime_nsec
1186          * Now, even though we're adjusting the clock frequency, we have
1187          * to keep time consistent. In other words, we can't jump back
1188          * in time, and we also want to avoid jumping forward in time.
1189          *
1190          * So given the same offset value, we need the time to be the same
1191          * both before and after the freq adjustment.
1192          *      now = (offset * adj_1) + xtime_nsec_1
1193          *      now = (offset * adj_2) + xtime_nsec_2
1194          * So:
1195          *      (offset * adj_1) + xtime_nsec_1 =
1196          *              (offset * adj_2) + xtime_nsec_2
1197          * And we know:
1198          *      adj_2 = adj_1 + 1
1199          * So:
1200          *      (offset * adj_1) + xtime_nsec_1 =
1201          *              (offset * (adj_1+1)) + xtime_nsec_2
1202          *      (offset * adj_1) + xtime_nsec_1 =
1203          *              (offset * adj_1) + offset + xtime_nsec_2
1204          * Canceling the sides:
1205          *      xtime_nsec_1 = offset + xtime_nsec_2
1206          * Which gives us:
1207          *      xtime_nsec_2 = xtime_nsec_1 - offset
1208          * Which simplfies to:
1209          *      xtime_nsec -= offset
1210          *
1211          * XXX - TODO: Doc ntp_error calculation.
1212          */
1213         tk->mult += adj;
1214         tk->xtime_interval += interval;
1215         tk->xtime_nsec -= offset;
1216         tk->ntp_error -= (interval - offset) << tk->ntp_error_shift;
1217 
1218 out_adjust:
1219         /*
1220          * It may be possible that when we entered this function, xtime_nsec
1221          * was very small.  Further, if we're slightly speeding the clocksource
1222          * in the code above, its possible the required corrective factor to
1223          * xtime_nsec could cause it to underflow.
1224          *
1225          * Now, since we already accumulated the second, cannot simply roll
1226          * the accumulated second back, since the NTP subsystem has been
1227          * notified via second_overflow. So instead we push xtime_nsec forward
1228          * by the amount we underflowed, and add that amount into the error.
1229          *
1230          * We'll correct this error next time through this function, when
1231          * xtime_nsec is not as small.
1232          */
1233         if (unlikely((s64)tk->xtime_nsec < 0)) {
1234                 s64 neg = -(s64)tk->xtime_nsec;
1235                 tk->xtime_nsec = 0;
1236                 tk->ntp_error += neg << tk->ntp_error_shift;
1237         }
1238 
1239 }
1240 
1241 /**
1242  * accumulate_nsecs_to_secs - Accumulates nsecs into secs
1243  *
1244  * Helper function that accumulates a the nsecs greater then a second
1245  * from the xtime_nsec field to the xtime_secs field.
1246  * It also calls into the NTP code to handle leapsecond processing.
1247  *
1248  */
1249 static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
1250 {
1251         u64 nsecps = (u64)NSEC_PER_SEC << tk->shift;
1252         unsigned int clock_set = 0;
1253 
1254         while (tk->xtime_nsec >= nsecps) {
1255                 int leap;
1256 
1257                 tk->xtime_nsec -= nsecps;
1258                 tk->xtime_sec++;
1259 
1260                 /* Figure out if its a leap sec and apply if needed */
1261                 leap = second_overflow(tk->xtime_sec);
1262                 if (unlikely(leap)) {
1263                         struct timespec ts;
1264 
1265                         tk->xtime_sec += leap;
1266 
1267                         ts.tv_sec = leap;
1268                         ts.tv_nsec = 0;
1269                         tk_set_wall_to_mono(tk,
1270                                 timespec_sub(tk->wall_to_monotonic, ts));
1271 
1272                         __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
1273 
1274                         clock_set = TK_CLOCK_WAS_SET;
1275                 }
1276         }
1277         return clock_set;
1278 }
1279 
1280 /**
1281  * logarithmic_accumulation - shifted accumulation of cycles
1282  *
1283  * This functions accumulates a shifted interval of cycles into
1284  * into a shifted interval nanoseconds. Allows for O(log) accumulation
1285  * loop.
1286  *
1287  * Returns the unconsumed cycles.
1288  */
1289 static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1290                                                 u32 shift,
1291                                                 unsigned int *clock_set)
1292 {
1293         cycle_t interval = tk->cycle_interval << shift;
1294         u64 raw_nsecs;
1295 
1296         /* If the offset is smaller then a shifted interval, do nothing */
1297         if (offset < interval)
1298                 return offset;
1299 
1300         /* Accumulate one shifted interval */
1301         offset -= interval;
1302         tk->cycle_last += interval;
1303 
1304         tk->xtime_nsec += tk->xtime_interval << shift;
1305         *clock_set |= accumulate_nsecs_to_secs(tk);
1306 
1307         /* Accumulate raw time */
1308         raw_nsecs = (u64)tk->raw_interval << shift;
1309         raw_nsecs += tk->raw_time.tv_nsec;
1310         if (raw_nsecs >= NSEC_PER_SEC) {
1311                 u64 raw_secs = raw_nsecs;
1312                 raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
1313                 tk->raw_time.tv_sec += raw_secs;
1314         }
1315         tk->raw_time.tv_nsec = raw_nsecs;
1316 
1317         /* Accumulate error between NTP and clock interval */
1318         tk->ntp_error += ntp_tick_length() << shift;
1319         tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
1320                                                 (tk->ntp_error_shift + shift);
1321 
1322         return offset;
1323 }
1324 
1325 #ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1326 static inline void old_vsyscall_fixup(struct timekeeper *tk)
1327 {
1328         s64 remainder;
1329 
1330         /*
1331         * Store only full nanoseconds into xtime_nsec after rounding
1332         * it up and add the remainder to the error difference.
1333         * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1334         * by truncating the remainder in vsyscalls. However, it causes
1335         * additional work to be done in timekeeping_adjust(). Once
1336         * the vsyscall implementations are converted to use xtime_nsec
1337         * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1338         * users are removed, this can be killed.
1339         */
1340         remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1341         tk->xtime_nsec -= remainder;
1342         tk->xtime_nsec += 1ULL << tk->shift;
1343         tk->ntp_error += remainder << tk->ntp_error_shift;
1344         tk->ntp_error -= (1ULL << tk->shift) << tk->ntp_error_shift;
1345 }
1346 #else
1347 #define old_vsyscall_fixup(tk)
1348 #endif
1349 
1350 
1351 
1352 /**
1353  * update_wall_time - Uses the current clocksource to increment the wall time
1354  *
1355  */
1356 void update_wall_time(void)
1357 {
1358         struct clocksource *clock;
1359         struct timekeeper *real_tk = &timekeeper;
1360         struct timekeeper *tk = &shadow_timekeeper;
1361         cycle_t offset;
1362         int shift = 0, maxshift;
1363         unsigned int clock_set = 0;
1364         unsigned long flags;
1365 
1366         raw_spin_lock_irqsave(&timekeeper_lock, flags);
1367 
1368         /* Make sure we're fully resumed: */
1369         if (unlikely(timekeeping_suspended))
1370                 goto out;
1371 
1372         clock = real_tk->clock;
1373 
1374 #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
1375         offset = real_tk->cycle_interval;
1376 #else
1377         offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
1378 #endif
1379 
1380         /* Check if there's really nothing to do */
1381         if (offset < real_tk->cycle_interval)
1382                 goto out;
1383 
1384         /*
1385          * With NO_HZ we may have to accumulate many cycle_intervals
1386          * (think "ticks") worth of time at once. To do this efficiently,
1387          * we calculate the largest doubling multiple of cycle_intervals
1388          * that is smaller than the offset.  We then accumulate that
1389          * chunk in one go, and then try to consume the next smaller
1390          * doubled multiple.
1391          */
1392         shift = ilog2(offset) - ilog2(tk->cycle_interval);
1393         shift = max(0, shift);
1394         /* Bound shift to one less than what overflows tick_length */
1395         maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
1396         shift = min(shift, maxshift);
1397         while (offset >= tk->cycle_interval) {
1398                 offset = logarithmic_accumulation(tk, offset, shift,
1399                                                         &clock_set);
1400                 if (offset < tk->cycle_interval<<shift)
1401                         shift--;
1402         }
1403 
1404         /* correct the clock when NTP error is too big */
1405         timekeeping_adjust(tk, offset);
1406 
1407         /*
1408          * XXX This can be killed once everyone converts
1409          * to the new update_vsyscall.
1410          */
1411         old_vsyscall_fixup(tk);
1412 
1413         /*
1414          * Finally, make sure that after the rounding
1415          * xtime_nsec isn't larger than NSEC_PER_SEC
1416          */
1417         clock_set |= accumulate_nsecs_to_secs(tk);
1418 
1419         write_seqcount_begin(&timekeeper_seq);
1420         /* Update clock->cycle_last with the new value */
1421         clock->cycle_last = tk->cycle_last;
1422         /*
1423          * Update the real timekeeper.
1424          *
1425          * We could avoid this memcpy by switching pointers, but that
1426          * requires changes to all other timekeeper usage sites as
1427          * well, i.e. move the timekeeper pointer getter into the
1428          * spinlocked/seqcount protected sections. And we trade this
1429          * memcpy under the timekeeper_seq against one before we start
1430          * updating.
1431          */
1432         memcpy(real_tk, tk, sizeof(*tk));
1433         timekeeping_update(real_tk, clock_set);
1434         write_seqcount_end(&timekeeper_seq);
1435 out:
1436         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1437         if (clock_set)
1438                 /* Have to call _delayed version, since in irq context*/
1439                 clock_was_set_delayed();
1440 }
1441 
1442 /**
1443  * getboottime - Return the real time of system boot.
1444  * @ts:         pointer to the timespec to be set
1445  *
1446  * Returns the wall-time of boot in a timespec.
1447  *
1448  * This is based on the wall_to_monotonic offset and the total suspend
1449  * time. Calls to settimeofday will affect the value returned (which
1450  * basically means that however wrong your real time clock is at boot time,
1451  * you get the right time here).
1452  */
1453 void getboottime(struct timespec *ts)
1454 {
1455         struct timekeeper *tk = &timekeeper;
1456         struct timespec boottime = {
1457                 .tv_sec = tk->wall_to_monotonic.tv_sec +
1458                                 tk->total_sleep_time.tv_sec,
1459                 .tv_nsec = tk->wall_to_monotonic.tv_nsec +
1460                                 tk->total_sleep_time.tv_nsec
1461         };
1462 
1463         set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
1464 }
1465 EXPORT_SYMBOL_GPL(getboottime);
1466 
1467 /**
1468  * get_monotonic_boottime - Returns monotonic time since boot
1469  * @ts:         pointer to the timespec to be set
1470  *
1471  * Returns the monotonic time since boot in a timespec.
1472  *
1473  * This is similar to CLOCK_MONTONIC/ktime_get_ts, but also
1474  * includes the time spent in suspend.
1475  */
1476 void get_monotonic_boottime(struct timespec *ts)
1477 {
1478         struct timekeeper *tk = &timekeeper;
1479         struct timespec tomono, sleep;
1480         s64 nsec;
1481         unsigned int seq;
1482 
1483         WARN_ON(timekeeping_suspended);
1484 
1485         do {
1486                 seq = read_seqcount_begin(&timekeeper_seq);
1487                 ts->tv_sec = tk->xtime_sec;
1488                 nsec = timekeeping_get_ns(tk);
1489                 tomono = tk->wall_to_monotonic;
1490                 sleep = tk->total_sleep_time;
1491 
1492         } while (read_seqcount_retry(&timekeeper_seq, seq));
1493 
1494         ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1495         ts->tv_nsec = 0;
1496         timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1497 }
1498 EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1499 
1500 /**
1501  * ktime_get_boottime - Returns monotonic time since boot in a ktime
1502  *
1503  * Returns the monotonic time since boot in a ktime
1504  *
1505  * This is similar to CLOCK_MONTONIC/ktime_get, but also
1506  * includes the time spent in suspend.
1507  */
1508 ktime_t ktime_get_boottime(void)
1509 {
1510         struct timespec ts;
1511 
1512         get_monotonic_boottime(&ts);
1513         return timespec_to_ktime(ts);
1514 }
1515 EXPORT_SYMBOL_GPL(ktime_get_boottime);
1516 
1517 /**
1518  * monotonic_to_bootbased - Convert the monotonic time to boot based.
1519  * @ts:         pointer to the timespec to be converted
1520  */
1521 void monotonic_to_bootbased(struct timespec *ts)
1522 {
1523         struct timekeeper *tk = &timekeeper;
1524 
1525         *ts = timespec_add(*ts, tk->total_sleep_time);
1526 }
1527 EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
1528 
1529 unsigned long get_seconds(void)
1530 {
1531         struct timekeeper *tk = &timekeeper;
1532 
1533         return tk->xtime_sec;
1534 }
1535 EXPORT_SYMBOL(get_seconds);
1536 
1537 struct timespec __current_kernel_time(void)
1538 {
1539         struct timekeeper *tk = &timekeeper;
1540 
1541         return tk_xtime(tk);
1542 }
1543 
1544 struct timespec current_kernel_time(void)
1545 {
1546         struct timekeeper *tk = &timekeeper;
1547         struct timespec now;
1548         unsigned long seq;
1549 
1550         do {
1551                 seq = read_seqcount_begin(&timekeeper_seq);
1552 
1553                 now = tk_xtime(tk);
1554         } while (read_seqcount_retry(&timekeeper_seq, seq));
1555 
1556         return now;
1557 }
1558 EXPORT_SYMBOL(current_kernel_time);
1559 
1560 struct timespec get_monotonic_coarse(void)
1561 {
1562         struct timekeeper *tk = &timekeeper;
1563         struct timespec now, mono;
1564         unsigned long seq;
1565 
1566         do {
1567                 seq = read_seqcount_begin(&timekeeper_seq);
1568 
1569                 now = tk_xtime(tk);
1570                 mono = tk->wall_to_monotonic;
1571         } while (read_seqcount_retry(&timekeeper_seq, seq));
1572 
1573         set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
1574                                 now.tv_nsec + mono.tv_nsec);
1575         return now;
1576 }
1577 
1578 /*
1579  * Must hold jiffies_lock
1580  */
1581 void do_timer(unsigned long ticks)
1582 {
1583         jiffies_64 += ticks;
1584         calc_global_load(ticks);
1585 }
1586 
1587 /**
1588  * get_xtime_and_monotonic_and_sleep_offset() - get xtime, wall_to_monotonic,
1589  *    and sleep offsets.
1590  * @xtim:       pointer to timespec to be set with xtime
1591  * @wtom:       pointer to timespec to be set with wall_to_monotonic
1592  * @sleep:      pointer to timespec to be set with time in suspend
1593  */
1594 void get_xtime_and_monotonic_and_sleep_offset(struct timespec *xtim,
1595                                 struct timespec *wtom, struct timespec *sleep)
1596 {
1597         struct timekeeper *tk = &timekeeper;
1598         unsigned long seq;
1599 
1600         do {
1601                 seq = read_seqcount_begin(&timekeeper_seq);
1602                 *xtim = tk_xtime(tk);
1603                 *wtom = tk->wall_to_monotonic;
1604                 *sleep = tk->total_sleep_time;
1605         } while (read_seqcount_retry(&timekeeper_seq, seq));
1606 }
1607 
1608 #ifdef CONFIG_HIGH_RES_TIMERS
1609 /**
1610  * ktime_get_update_offsets - hrtimer helper
1611  * @offs_real:  pointer to storage for monotonic -> realtime offset
1612  * @offs_boot:  pointer to storage for monotonic -> boottime offset
1613  * @offs_tai:   pointer to storage for monotonic -> clock tai offset
1614  *
1615  * Returns current monotonic time and updates the offsets
1616  * Called from hrtimer_interrupt() or retrigger_next_event()
1617  */
1618 ktime_t ktime_get_update_offsets(ktime_t *offs_real, ktime_t *offs_boot,
1619                                                         ktime_t *offs_tai)
1620 {
1621         struct timekeeper *tk = &timekeeper;
1622         ktime_t now;
1623         unsigned int seq;
1624         u64 secs, nsecs;
1625 
1626         do {
1627                 seq = read_seqcount_begin(&timekeeper_seq);
1628 
1629                 secs = tk->xtime_sec;
1630                 nsecs = timekeeping_get_ns(tk);
1631 
1632                 *offs_real = tk->offs_real;
1633                 *offs_boot = tk->offs_boot;
1634                 *offs_tai = tk->offs_tai;
1635         } while (read_seqcount_retry(&timekeeper_seq, seq));
1636 
1637         now = ktime_add_ns(ktime_set(secs, 0), nsecs);
1638         now = ktime_sub(now, *offs_real);
1639         return now;
1640 }
1641 #endif
1642 
1643 /**
1644  * ktime_get_monotonic_offset() - get wall_to_monotonic in ktime_t format
1645  */
1646 ktime_t ktime_get_monotonic_offset(void)
1647 {
1648         struct timekeeper *tk = &timekeeper;
1649         unsigned long seq;
1650         struct timespec wtom;
1651 
1652         do {
1653                 seq = read_seqcount_begin(&timekeeper_seq);
1654                 wtom = tk->wall_to_monotonic;
1655         } while (read_seqcount_retry(&timekeeper_seq, seq));
1656 
1657         return timespec_to_ktime(wtom);
1658 }
1659 EXPORT_SYMBOL_GPL(ktime_get_monotonic_offset);
1660 
1661 /**
1662  * do_adjtimex() - Accessor function to NTP __do_adjtimex function
1663  */
1664 int do_adjtimex(struct timex *txc)
1665 {
1666         struct timekeeper *tk = &timekeeper;
1667         unsigned long flags;
1668         struct timespec ts;
1669         s32 orig_tai, tai;
1670         int ret;
1671 
1672         /* Validate the data before disabling interrupts */
1673         ret = ntp_validate_timex(txc);
1674         if (ret)
1675                 return ret;
1676 
1677         if (txc->modes & ADJ_SETOFFSET) {
1678                 struct timespec delta;
1679                 delta.tv_sec  = txc->time.tv_sec;
1680                 delta.tv_nsec = txc->time.tv_usec;
1681                 if (!(txc->modes & ADJ_NANO))
1682                         delta.tv_nsec *= 1000;
1683                 ret = timekeeping_inject_offset(&delta);
1684                 if (ret)
1685                         return ret;
1686         }
1687 
1688         getnstimeofday(&ts);
1689 
1690         raw_spin_lock_irqsave(&timekeeper_lock, flags);
1691         write_seqcount_begin(&timekeeper_seq);
1692 
1693         orig_tai = tai = tk->tai_offset;
1694         ret = __do_adjtimex(txc, &ts, &tai);
1695 
1696         if (tai != orig_tai) {
1697                 __timekeeping_set_tai_offset(tk, tai);
1698                 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
1699         }
1700         write_seqcount_end(&timekeeper_seq);
1701         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1702 
1703         if (tai != orig_tai)
1704                 clock_was_set();
1705 
1706         ntp_notify_cmos_timer();
1707 
1708         return ret;
1709 }
1710 
1711 #ifdef CONFIG_NTP_PPS
1712 /**
1713  * hardpps() - Accessor function to NTP __hardpps function
1714  */
1715 void hardpps(const struct timespec *phase_ts, const struct timespec *raw_ts)
1716 {
1717         unsigned long flags;
1718 
1719         raw_spin_lock_irqsave(&timekeeper_lock, flags);
1720         write_seqcount_begin(&timekeeper_seq);
1721 
1722         __hardpps(phase_ts, raw_ts);
1723 
1724         write_seqcount_end(&timekeeper_seq);
1725         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
1726 }
1727 EXPORT_SYMBOL(hardpps);
1728 #endif
1729 
1730 /**
1731  * xtime_update() - advances the timekeeping infrastructure
1732  * @ticks:      number of ticks, that have elapsed since the last call.
1733  *
1734  * Must be called with interrupts disabled.
1735  */
1736 void xtime_update(unsigned long ticks)
1737 {
1738         write_seqlock(&jiffies_lock);
1739         do_timer(ticks);
1740         write_sequnlock(&jiffies_lock);
1741         update_wall_time();
1742 }
1743 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp