timekeeping.c 24 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948
  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. #include <linux/module.h>
  11. #include <linux/interrupt.h>
  12. #include <linux/percpu.h>
  13. #include <linux/init.h>
  14. #include <linux/mm.h>
  15. #include <linux/sched.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. #include <linux/stop_machine.h>
  22. /* Structure holding internal timekeeping values. */
  23. struct timekeeper {
  24. /* Current clocksource used for timekeeping. */
  25. struct clocksource *clock;
  26. /* The shift value of the current clocksource. */
  27. int shift;
  28. /* Number of clock cycles in one NTP interval. */
  29. cycle_t cycle_interval;
  30. /* Number of clock shifted nano seconds in one NTP interval. */
  31. u64 xtime_interval;
  32. /* shifted nano seconds left over when rounding cycle_interval */
  33. s64 xtime_remainder;
  34. /* Raw nano seconds accumulated per NTP interval. */
  35. u32 raw_interval;
  36. /* Clock shifted nano seconds remainder not stored in xtime.tv_nsec. */
  37. u64 xtime_nsec;
  38. /* Difference between accumulated time and NTP time in ntp
  39. * shifted nano seconds. */
  40. s64 ntp_error;
  41. /* Shift conversion between clock shifted nano seconds and
  42. * ntp shifted nano seconds. */
  43. int ntp_error_shift;
  44. /* NTP adjusted clock multiplier */
  45. u32 mult;
  46. };
  47. static struct timekeeper timekeeper;
  48. /**
  49. * timekeeper_setup_internals - Set up internals to use clocksource clock.
  50. *
  51. * @clock: Pointer to clocksource.
  52. *
  53. * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
  54. * pair and interval request.
  55. *
  56. * Unless you're the timekeeping code, you should not be using this!
  57. */
  58. static void timekeeper_setup_internals(struct clocksource *clock)
  59. {
  60. cycle_t interval;
  61. u64 tmp, ntpinterval;
  62. timekeeper.clock = clock;
  63. clock->cycle_last = clock->read(clock);
  64. /* Do the ns -> cycle conversion first, using original mult */
  65. tmp = NTP_INTERVAL_LENGTH;
  66. tmp <<= clock->shift;
  67. ntpinterval = tmp;
  68. tmp += clock->mult/2;
  69. do_div(tmp, clock->mult);
  70. if (tmp == 0)
  71. tmp = 1;
  72. interval = (cycle_t) tmp;
  73. timekeeper.cycle_interval = interval;
  74. /* Go back from cycles -> shifted ns */
  75. timekeeper.xtime_interval = (u64) interval * clock->mult;
  76. timekeeper.xtime_remainder = ntpinterval - timekeeper.xtime_interval;
  77. timekeeper.raw_interval =
  78. ((u64) interval * clock->mult) >> clock->shift;
  79. timekeeper.xtime_nsec = 0;
  80. timekeeper.shift = clock->shift;
  81. timekeeper.ntp_error = 0;
  82. timekeeper.ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
  83. /*
  84. * The timekeeper keeps its own mult values for the currently
  85. * active clocksource. These value will be adjusted via NTP
  86. * to counteract clock drifting.
  87. */
  88. timekeeper.mult = clock->mult;
  89. }
  90. /* Timekeeper helper functions. */
  91. static inline s64 timekeeping_get_ns(void)
  92. {
  93. cycle_t cycle_now, cycle_delta;
  94. struct clocksource *clock;
  95. /* read clocksource: */
  96. clock = timekeeper.clock;
  97. cycle_now = clock->read(clock);
  98. /* calculate the delta since the last update_wall_time: */
  99. cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
  100. /* return delta convert to nanoseconds using ntp adjusted mult. */
  101. return clocksource_cyc2ns(cycle_delta, timekeeper.mult,
  102. timekeeper.shift);
  103. }
  104. static inline s64 timekeeping_get_ns_raw(void)
  105. {
  106. cycle_t cycle_now, cycle_delta;
  107. struct clocksource *clock;
  108. /* read clocksource: */
  109. clock = timekeeper.clock;
  110. cycle_now = clock->read(clock);
  111. /* calculate the delta since the last update_wall_time: */
  112. cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
  113. /* return delta convert to nanoseconds using ntp adjusted mult. */
  114. return clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
  115. }
  116. /*
  117. * This read-write spinlock protects us from races in SMP while
  118. * playing with xtime.
  119. */
  120. __cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
  121. /*
  122. * The current time
  123. * wall_to_monotonic is what we need to add to xtime (or xtime corrected
  124. * for sub jiffie times) to get to monotonic time. Monotonic is pegged
  125. * at zero at system boot time, so wall_to_monotonic will be negative,
  126. * however, we will ALWAYS keep the tv_nsec part positive so we can use
  127. * the usual normalization.
  128. *
  129. * wall_to_monotonic is moved after resume from suspend for the monotonic
  130. * time not to jump. We need to add total_sleep_time to wall_to_monotonic
  131. * to get the real boot based time offset.
  132. *
  133. * - wall_to_monotonic is no longer the boot time, getboottime must be
  134. * used instead.
  135. */
  136. static struct timespec xtime __attribute__ ((aligned (16)));
  137. static struct timespec wall_to_monotonic __attribute__ ((aligned (16)));
  138. static struct timespec total_sleep_time;
  139. /*
  140. * The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock.
  141. */
  142. static struct timespec raw_time;
  143. /* flag for if timekeeping is suspended */
  144. int __read_mostly timekeeping_suspended;
  145. /* must hold xtime_lock */
  146. void timekeeping_leap_insert(int leapsecond)
  147. {
  148. xtime.tv_sec += leapsecond;
  149. wall_to_monotonic.tv_sec -= leapsecond;
  150. update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
  151. timekeeper.mult);
  152. }
  153. /**
  154. * timekeeping_forward_now - update clock to the current time
  155. *
  156. * Forward the current clock to update its state since the last call to
  157. * update_wall_time(). This is useful before significant clock changes,
  158. * as it avoids having to deal with this time offset explicitly.
  159. */
  160. static void timekeeping_forward_now(void)
  161. {
  162. cycle_t cycle_now, cycle_delta;
  163. struct clocksource *clock;
  164. s64 nsec;
  165. clock = timekeeper.clock;
  166. cycle_now = clock->read(clock);
  167. cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
  168. clock->cycle_last = cycle_now;
  169. nsec = clocksource_cyc2ns(cycle_delta, timekeeper.mult,
  170. timekeeper.shift);
  171. /* If arch requires, add in gettimeoffset() */
  172. nsec += arch_gettimeoffset();
  173. timespec_add_ns(&xtime, nsec);
  174. nsec = clocksource_cyc2ns(cycle_delta, clock->mult, clock->shift);
  175. timespec_add_ns(&raw_time, nsec);
  176. }
  177. /**
  178. * getnstimeofday - Returns the time of day in a timespec
  179. * @ts: pointer to the timespec to be set
  180. *
  181. * Returns the time of day in a timespec.
  182. */
  183. void getnstimeofday(struct timespec *ts)
  184. {
  185. unsigned long seq;
  186. s64 nsecs;
  187. WARN_ON(timekeeping_suspended);
  188. do {
  189. seq = read_seqbegin(&xtime_lock);
  190. *ts = xtime;
  191. nsecs = timekeeping_get_ns();
  192. /* If arch requires, add in gettimeoffset() */
  193. nsecs += arch_gettimeoffset();
  194. } while (read_seqretry(&xtime_lock, seq));
  195. timespec_add_ns(ts, nsecs);
  196. }
  197. EXPORT_SYMBOL(getnstimeofday);
  198. ktime_t ktime_get(void)
  199. {
  200. unsigned int seq;
  201. s64 secs, nsecs;
  202. WARN_ON(timekeeping_suspended);
  203. do {
  204. seq = read_seqbegin(&xtime_lock);
  205. secs = xtime.tv_sec + wall_to_monotonic.tv_sec;
  206. nsecs = xtime.tv_nsec + wall_to_monotonic.tv_nsec;
  207. nsecs += timekeeping_get_ns();
  208. } while (read_seqretry(&xtime_lock, seq));
  209. /*
  210. * Use ktime_set/ktime_add_ns to create a proper ktime on
  211. * 32-bit architectures without CONFIG_KTIME_SCALAR.
  212. */
  213. return ktime_add_ns(ktime_set(secs, 0), nsecs);
  214. }
  215. EXPORT_SYMBOL_GPL(ktime_get);
  216. /**
  217. * ktime_get_ts - get the monotonic clock in timespec format
  218. * @ts: pointer to timespec variable
  219. *
  220. * The function calculates the monotonic clock from the realtime
  221. * clock and the wall_to_monotonic offset and stores the result
  222. * in normalized timespec format in the variable pointed to by @ts.
  223. */
  224. void ktime_get_ts(struct timespec *ts)
  225. {
  226. struct timespec tomono;
  227. unsigned int seq;
  228. s64 nsecs;
  229. WARN_ON(timekeeping_suspended);
  230. do {
  231. seq = read_seqbegin(&xtime_lock);
  232. *ts = xtime;
  233. tomono = wall_to_monotonic;
  234. nsecs = timekeeping_get_ns();
  235. } while (read_seqretry(&xtime_lock, seq));
  236. set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
  237. ts->tv_nsec + tomono.tv_nsec + nsecs);
  238. }
  239. EXPORT_SYMBOL_GPL(ktime_get_ts);
  240. #ifdef CONFIG_NTP_PPS
  241. /**
  242. * getnstime_raw_and_real - get day and raw monotonic time in timespec format
  243. * @ts_raw: pointer to the timespec to be set to raw monotonic time
  244. * @ts_real: pointer to the timespec to be set to the time of day
  245. *
  246. * This function reads both the time of day and raw monotonic time at the
  247. * same time atomically and stores the resulting timestamps in timespec
  248. * format.
  249. */
  250. void getnstime_raw_and_real(struct timespec *ts_raw, struct timespec *ts_real)
  251. {
  252. unsigned long seq;
  253. s64 nsecs_raw, nsecs_real;
  254. WARN_ON_ONCE(timekeeping_suspended);
  255. do {
  256. u32 arch_offset;
  257. seq = read_seqbegin(&xtime_lock);
  258. *ts_raw = raw_time;
  259. *ts_real = xtime;
  260. nsecs_raw = timekeeping_get_ns_raw();
  261. nsecs_real = timekeeping_get_ns();
  262. /* If arch requires, add in gettimeoffset() */
  263. arch_offset = arch_gettimeoffset();
  264. nsecs_raw += arch_offset;
  265. nsecs_real += arch_offset;
  266. } while (read_seqretry(&xtime_lock, seq));
  267. timespec_add_ns(ts_raw, nsecs_raw);
  268. timespec_add_ns(ts_real, nsecs_real);
  269. }
  270. EXPORT_SYMBOL(getnstime_raw_and_real);
  271. #endif /* CONFIG_NTP_PPS */
  272. /**
  273. * do_gettimeofday - Returns the time of day in a timeval
  274. * @tv: pointer to the timeval to be set
  275. *
  276. * NOTE: Users should be converted to using getnstimeofday()
  277. */
  278. void do_gettimeofday(struct timeval *tv)
  279. {
  280. struct timespec now;
  281. getnstimeofday(&now);
  282. tv->tv_sec = now.tv_sec;
  283. tv->tv_usec = now.tv_nsec/1000;
  284. }
  285. EXPORT_SYMBOL(do_gettimeofday);
  286. /**
  287. * do_settimeofday - Sets the time of day
  288. * @tv: pointer to the timespec variable containing the new time
  289. *
  290. * Sets the time of day to the new time and update NTP and notify hrtimers
  291. */
  292. int do_settimeofday(struct timespec *tv)
  293. {
  294. struct timespec ts_delta;
  295. unsigned long flags;
  296. if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
  297. return -EINVAL;
  298. write_seqlock_irqsave(&xtime_lock, flags);
  299. timekeeping_forward_now();
  300. ts_delta.tv_sec = tv->tv_sec - xtime.tv_sec;
  301. ts_delta.tv_nsec = tv->tv_nsec - xtime.tv_nsec;
  302. wall_to_monotonic = timespec_sub(wall_to_monotonic, ts_delta);
  303. xtime = *tv;
  304. timekeeper.ntp_error = 0;
  305. ntp_clear();
  306. update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
  307. timekeeper.mult);
  308. write_sequnlock_irqrestore(&xtime_lock, flags);
  309. /* signal hrtimers about time change */
  310. clock_was_set();
  311. return 0;
  312. }
  313. EXPORT_SYMBOL(do_settimeofday);
  314. /**
  315. * change_clocksource - Swaps clocksources if a new one is available
  316. *
  317. * Accumulates current time interval and initializes new clocksource
  318. */
  319. static int change_clocksource(void *data)
  320. {
  321. struct clocksource *new, *old;
  322. new = (struct clocksource *) data;
  323. timekeeping_forward_now();
  324. if (!new->enable || new->enable(new) == 0) {
  325. old = timekeeper.clock;
  326. timekeeper_setup_internals(new);
  327. if (old->disable)
  328. old->disable(old);
  329. }
  330. return 0;
  331. }
  332. /**
  333. * timekeeping_notify - Install a new clock source
  334. * @clock: pointer to the clock source
  335. *
  336. * This function is called from clocksource.c after a new, better clock
  337. * source has been registered. The caller holds the clocksource_mutex.
  338. */
  339. void timekeeping_notify(struct clocksource *clock)
  340. {
  341. if (timekeeper.clock == clock)
  342. return;
  343. stop_machine(change_clocksource, clock, NULL);
  344. tick_clock_notify();
  345. }
  346. /**
  347. * ktime_get_real - get the real (wall-) time in ktime_t format
  348. *
  349. * returns the time in ktime_t format
  350. */
  351. ktime_t ktime_get_real(void)
  352. {
  353. struct timespec now;
  354. getnstimeofday(&now);
  355. return timespec_to_ktime(now);
  356. }
  357. EXPORT_SYMBOL_GPL(ktime_get_real);
  358. /**
  359. * getrawmonotonic - Returns the raw monotonic time in a timespec
  360. * @ts: pointer to the timespec to be set
  361. *
  362. * Returns the raw monotonic time (completely un-modified by ntp)
  363. */
  364. void getrawmonotonic(struct timespec *ts)
  365. {
  366. unsigned long seq;
  367. s64 nsecs;
  368. do {
  369. seq = read_seqbegin(&xtime_lock);
  370. nsecs = timekeeping_get_ns_raw();
  371. *ts = raw_time;
  372. } while (read_seqretry(&xtime_lock, seq));
  373. timespec_add_ns(ts, nsecs);
  374. }
  375. EXPORT_SYMBOL(getrawmonotonic);
  376. /**
  377. * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
  378. */
  379. int timekeeping_valid_for_hres(void)
  380. {
  381. unsigned long seq;
  382. int ret;
  383. do {
  384. seq = read_seqbegin(&xtime_lock);
  385. ret = timekeeper.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
  386. } while (read_seqretry(&xtime_lock, seq));
  387. return ret;
  388. }
  389. /**
  390. * timekeeping_max_deferment - Returns max time the clocksource can be deferred
  391. *
  392. * Caller must observe xtime_lock via read_seqbegin/read_seqretry to
  393. * ensure that the clocksource does not change!
  394. */
  395. u64 timekeeping_max_deferment(void)
  396. {
  397. return timekeeper.clock->max_idle_ns;
  398. }
  399. /**
  400. * read_persistent_clock - Return time from the persistent clock.
  401. *
  402. * Weak dummy function for arches that do not yet support it.
  403. * Reads the time from the battery backed persistent clock.
  404. * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
  405. *
  406. * XXX - Do be sure to remove it once all arches implement it.
  407. */
  408. void __attribute__((weak)) read_persistent_clock(struct timespec *ts)
  409. {
  410. ts->tv_sec = 0;
  411. ts->tv_nsec = 0;
  412. }
  413. /**
  414. * read_boot_clock - Return time of the system start.
  415. *
  416. * Weak dummy function for arches that do not yet support it.
  417. * Function to read the exact time the system has been started.
  418. * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
  419. *
  420. * XXX - Do be sure to remove it once all arches implement it.
  421. */
  422. void __attribute__((weak)) read_boot_clock(struct timespec *ts)
  423. {
  424. ts->tv_sec = 0;
  425. ts->tv_nsec = 0;
  426. }
  427. /*
  428. * timekeeping_init - Initializes the clocksource and common timekeeping values
  429. */
  430. void __init timekeeping_init(void)
  431. {
  432. struct clocksource *clock;
  433. unsigned long flags;
  434. struct timespec now, boot;
  435. read_persistent_clock(&now);
  436. read_boot_clock(&boot);
  437. write_seqlock_irqsave(&xtime_lock, flags);
  438. ntp_init();
  439. clock = clocksource_default_clock();
  440. if (clock->enable)
  441. clock->enable(clock);
  442. timekeeper_setup_internals(clock);
  443. xtime.tv_sec = now.tv_sec;
  444. xtime.tv_nsec = now.tv_nsec;
  445. raw_time.tv_sec = 0;
  446. raw_time.tv_nsec = 0;
  447. if (boot.tv_sec == 0 && boot.tv_nsec == 0) {
  448. boot.tv_sec = xtime.tv_sec;
  449. boot.tv_nsec = xtime.tv_nsec;
  450. }
  451. set_normalized_timespec(&wall_to_monotonic,
  452. -boot.tv_sec, -boot.tv_nsec);
  453. total_sleep_time.tv_sec = 0;
  454. total_sleep_time.tv_nsec = 0;
  455. write_sequnlock_irqrestore(&xtime_lock, flags);
  456. }
  457. /* time in seconds when suspend began */
  458. static struct timespec timekeeping_suspend_time;
  459. /**
  460. * timekeeping_resume - Resumes the generic timekeeping subsystem.
  461. * @dev: unused
  462. *
  463. * This is for the generic clocksource timekeeping.
  464. * xtime/wall_to_monotonic/jiffies/etc are
  465. * still managed by arch specific suspend/resume code.
  466. */
  467. static int timekeeping_resume(struct sys_device *dev)
  468. {
  469. unsigned long flags;
  470. struct timespec ts;
  471. read_persistent_clock(&ts);
  472. clocksource_resume();
  473. write_seqlock_irqsave(&xtime_lock, flags);
  474. if (timespec_compare(&ts, &timekeeping_suspend_time) > 0) {
  475. ts = timespec_sub(ts, timekeeping_suspend_time);
  476. xtime = timespec_add(xtime, ts);
  477. wall_to_monotonic = timespec_sub(wall_to_monotonic, ts);
  478. total_sleep_time = timespec_add(total_sleep_time, ts);
  479. }
  480. /* re-base the last cycle value */
  481. timekeeper.clock->cycle_last = timekeeper.clock->read(timekeeper.clock);
  482. timekeeper.ntp_error = 0;
  483. timekeeping_suspended = 0;
  484. write_sequnlock_irqrestore(&xtime_lock, flags);
  485. touch_softlockup_watchdog();
  486. clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
  487. /* Resume hrtimers */
  488. hres_timers_resume();
  489. return 0;
  490. }
  491. static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
  492. {
  493. unsigned long flags;
  494. read_persistent_clock(&timekeeping_suspend_time);
  495. write_seqlock_irqsave(&xtime_lock, flags);
  496. timekeeping_forward_now();
  497. timekeeping_suspended = 1;
  498. write_sequnlock_irqrestore(&xtime_lock, flags);
  499. clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
  500. clocksource_suspend();
  501. return 0;
  502. }
  503. /* sysfs resume/suspend bits for timekeeping */
  504. static struct sysdev_class timekeeping_sysclass = {
  505. .name = "timekeeping",
  506. .resume = timekeeping_resume,
  507. .suspend = timekeeping_suspend,
  508. };
  509. static struct sys_device device_timer = {
  510. .id = 0,
  511. .cls = &timekeeping_sysclass,
  512. };
  513. static int __init timekeeping_init_device(void)
  514. {
  515. int error = sysdev_class_register(&timekeeping_sysclass);
  516. if (!error)
  517. error = sysdev_register(&device_timer);
  518. return error;
  519. }
  520. device_initcall(timekeeping_init_device);
  521. /*
  522. * If the error is already larger, we look ahead even further
  523. * to compensate for late or lost adjustments.
  524. */
  525. static __always_inline int timekeeping_bigadjust(s64 error, s64 *interval,
  526. s64 *offset)
  527. {
  528. s64 tick_error, i;
  529. u32 look_ahead, adj;
  530. s32 error2, mult;
  531. /*
  532. * Use the current error value to determine how much to look ahead.
  533. * The larger the error the slower we adjust for it to avoid problems
  534. * with losing too many ticks, otherwise we would overadjust and
  535. * produce an even larger error. The smaller the adjustment the
  536. * faster we try to adjust for it, as lost ticks can do less harm
  537. * here. This is tuned so that an error of about 1 msec is adjusted
  538. * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
  539. */
  540. error2 = timekeeper.ntp_error >> (NTP_SCALE_SHIFT + 22 - 2 * SHIFT_HZ);
  541. error2 = abs(error2);
  542. for (look_ahead = 0; error2 > 0; look_ahead++)
  543. error2 >>= 2;
  544. /*
  545. * Now calculate the error in (1 << look_ahead) ticks, but first
  546. * remove the single look ahead already included in the error.
  547. */
  548. tick_error = tick_length >> (timekeeper.ntp_error_shift + 1);
  549. tick_error -= timekeeper.xtime_interval >> 1;
  550. error = ((error - tick_error) >> look_ahead) + tick_error;
  551. /* Finally calculate the adjustment shift value. */
  552. i = *interval;
  553. mult = 1;
  554. if (error < 0) {
  555. error = -error;
  556. *interval = -*interval;
  557. *offset = -*offset;
  558. mult = -1;
  559. }
  560. for (adj = 0; error > i; adj++)
  561. error >>= 1;
  562. *interval <<= adj;
  563. *offset <<= adj;
  564. return mult << adj;
  565. }
  566. /*
  567. * Adjust the multiplier to reduce the error value,
  568. * this is optimized for the most common adjustments of -1,0,1,
  569. * for other values we can do a bit more work.
  570. */
  571. static void timekeeping_adjust(s64 offset)
  572. {
  573. s64 error, interval = timekeeper.cycle_interval;
  574. int adj;
  575. error = timekeeper.ntp_error >> (timekeeper.ntp_error_shift - 1);
  576. if (error > interval) {
  577. error >>= 2;
  578. if (likely(error <= interval))
  579. adj = 1;
  580. else
  581. adj = timekeeping_bigadjust(error, &interval, &offset);
  582. } else if (error < -interval) {
  583. error >>= 2;
  584. if (likely(error >= -interval)) {
  585. adj = -1;
  586. interval = -interval;
  587. offset = -offset;
  588. } else
  589. adj = timekeeping_bigadjust(error, &interval, &offset);
  590. } else
  591. return;
  592. timekeeper.mult += adj;
  593. timekeeper.xtime_interval += interval;
  594. timekeeper.xtime_nsec -= offset;
  595. timekeeper.ntp_error -= (interval - offset) <<
  596. timekeeper.ntp_error_shift;
  597. }
  598. /**
  599. * logarithmic_accumulation - shifted accumulation of cycles
  600. *
  601. * This functions accumulates a shifted interval of cycles into
  602. * into a shifted interval nanoseconds. Allows for O(log) accumulation
  603. * loop.
  604. *
  605. * Returns the unconsumed cycles.
  606. */
  607. static cycle_t logarithmic_accumulation(cycle_t offset, int shift)
  608. {
  609. u64 nsecps = (u64)NSEC_PER_SEC << timekeeper.shift;
  610. u64 raw_nsecs;
  611. /* If the offset is smaller then a shifted interval, do nothing */
  612. if (offset < timekeeper.cycle_interval<<shift)
  613. return offset;
  614. /* Accumulate one shifted interval */
  615. offset -= timekeeper.cycle_interval << shift;
  616. timekeeper.clock->cycle_last += timekeeper.cycle_interval << shift;
  617. timekeeper.xtime_nsec += timekeeper.xtime_interval << shift;
  618. while (timekeeper.xtime_nsec >= nsecps) {
  619. timekeeper.xtime_nsec -= nsecps;
  620. xtime.tv_sec++;
  621. second_overflow();
  622. }
  623. /* Accumulate raw time */
  624. raw_nsecs = timekeeper.raw_interval << shift;
  625. raw_nsecs += raw_time.tv_nsec;
  626. if (raw_nsecs >= NSEC_PER_SEC) {
  627. u64 raw_secs = raw_nsecs;
  628. raw_nsecs = do_div(raw_secs, NSEC_PER_SEC);
  629. raw_time.tv_sec += raw_secs;
  630. }
  631. raw_time.tv_nsec = raw_nsecs;
  632. /* Accumulate error between NTP and clock interval */
  633. timekeeper.ntp_error += tick_length << shift;
  634. timekeeper.ntp_error -=
  635. (timekeeper.xtime_interval + timekeeper.xtime_remainder) <<
  636. (timekeeper.ntp_error_shift + shift);
  637. return offset;
  638. }
  639. /**
  640. * update_wall_time - Uses the current clocksource to increment the wall time
  641. *
  642. * Called from the timer interrupt, must hold a write on xtime_lock.
  643. */
  644. void update_wall_time(void)
  645. {
  646. struct clocksource *clock;
  647. cycle_t offset;
  648. int shift = 0, maxshift;
  649. /* Make sure we're fully resumed: */
  650. if (unlikely(timekeeping_suspended))
  651. return;
  652. clock = timekeeper.clock;
  653. #ifdef CONFIG_ARCH_USES_GETTIMEOFFSET
  654. offset = timekeeper.cycle_interval;
  655. #else
  656. offset = (clock->read(clock) - clock->cycle_last) & clock->mask;
  657. #endif
  658. timekeeper.xtime_nsec = (s64)xtime.tv_nsec << timekeeper.shift;
  659. /*
  660. * With NO_HZ we may have to accumulate many cycle_intervals
  661. * (think "ticks") worth of time at once. To do this efficiently,
  662. * we calculate the largest doubling multiple of cycle_intervals
  663. * that is smaller then the offset. We then accumulate that
  664. * chunk in one go, and then try to consume the next smaller
  665. * doubled multiple.
  666. */
  667. shift = ilog2(offset) - ilog2(timekeeper.cycle_interval);
  668. shift = max(0, shift);
  669. /* Bound shift to one less then what overflows tick_length */
  670. maxshift = (8*sizeof(tick_length) - (ilog2(tick_length)+1)) - 1;
  671. shift = min(shift, maxshift);
  672. while (offset >= timekeeper.cycle_interval) {
  673. offset = logarithmic_accumulation(offset, shift);
  674. if(offset < timekeeper.cycle_interval<<shift)
  675. shift--;
  676. }
  677. /* correct the clock when NTP error is too big */
  678. timekeeping_adjust(offset);
  679. /*
  680. * Since in the loop above, we accumulate any amount of time
  681. * in xtime_nsec over a second into xtime.tv_sec, its possible for
  682. * xtime_nsec to be fairly small after the loop. Further, if we're
  683. * slightly speeding the clocksource up in timekeeping_adjust(),
  684. * its possible the required corrective factor to xtime_nsec could
  685. * cause it to underflow.
  686. *
  687. * Now, we cannot simply roll the accumulated second back, since
  688. * the NTP subsystem has been notified via second_overflow. So
  689. * instead we push xtime_nsec forward by the amount we underflowed,
  690. * and add that amount into the error.
  691. *
  692. * We'll correct this error next time through this function, when
  693. * xtime_nsec is not as small.
  694. */
  695. if (unlikely((s64)timekeeper.xtime_nsec < 0)) {
  696. s64 neg = -(s64)timekeeper.xtime_nsec;
  697. timekeeper.xtime_nsec = 0;
  698. timekeeper.ntp_error += neg << timekeeper.ntp_error_shift;
  699. }
  700. /*
  701. * Store full nanoseconds into xtime after rounding it up and
  702. * add the remainder to the error difference.
  703. */
  704. xtime.tv_nsec = ((s64) timekeeper.xtime_nsec >> timekeeper.shift) + 1;
  705. timekeeper.xtime_nsec -= (s64) xtime.tv_nsec << timekeeper.shift;
  706. timekeeper.ntp_error += timekeeper.xtime_nsec <<
  707. timekeeper.ntp_error_shift;
  708. /*
  709. * Finally, make sure that after the rounding
  710. * xtime.tv_nsec isn't larger then NSEC_PER_SEC
  711. */
  712. if (unlikely(xtime.tv_nsec >= NSEC_PER_SEC)) {
  713. xtime.tv_nsec -= NSEC_PER_SEC;
  714. xtime.tv_sec++;
  715. second_overflow();
  716. }
  717. /* check to see if there is a new clocksource to use */
  718. update_vsyscall(&xtime, &wall_to_monotonic, timekeeper.clock,
  719. timekeeper.mult);
  720. }
  721. /**
  722. * getboottime - Return the real time of system boot.
  723. * @ts: pointer to the timespec to be set
  724. *
  725. * Returns the time of day in a timespec.
  726. *
  727. * This is based on the wall_to_monotonic offset and the total suspend
  728. * time. Calls to settimeofday will affect the value returned (which
  729. * basically means that however wrong your real time clock is at boot time,
  730. * you get the right time here).
  731. */
  732. void getboottime(struct timespec *ts)
  733. {
  734. struct timespec boottime = {
  735. .tv_sec = wall_to_monotonic.tv_sec + total_sleep_time.tv_sec,
  736. .tv_nsec = wall_to_monotonic.tv_nsec + total_sleep_time.tv_nsec
  737. };
  738. set_normalized_timespec(ts, -boottime.tv_sec, -boottime.tv_nsec);
  739. }
  740. EXPORT_SYMBOL_GPL(getboottime);
  741. /**
  742. * monotonic_to_bootbased - Convert the monotonic time to boot based.
  743. * @ts: pointer to the timespec to be converted
  744. */
  745. void monotonic_to_bootbased(struct timespec *ts)
  746. {
  747. *ts = timespec_add(*ts, total_sleep_time);
  748. }
  749. EXPORT_SYMBOL_GPL(monotonic_to_bootbased);
  750. unsigned long get_seconds(void)
  751. {
  752. return xtime.tv_sec;
  753. }
  754. EXPORT_SYMBOL(get_seconds);
  755. struct timespec __current_kernel_time(void)
  756. {
  757. return xtime;
  758. }
  759. struct timespec __get_wall_to_monotonic(void)
  760. {
  761. return wall_to_monotonic;
  762. }
  763. struct timespec current_kernel_time(void)
  764. {
  765. struct timespec now;
  766. unsigned long seq;
  767. do {
  768. seq = read_seqbegin(&xtime_lock);
  769. now = xtime;
  770. } while (read_seqretry(&xtime_lock, seq));
  771. return now;
  772. }
  773. EXPORT_SYMBOL(current_kernel_time);
  774. struct timespec get_monotonic_coarse(void)
  775. {
  776. struct timespec now, mono;
  777. unsigned long seq;
  778. do {
  779. seq = read_seqbegin(&xtime_lock);
  780. now = xtime;
  781. mono = wall_to_monotonic;
  782. } while (read_seqretry(&xtime_lock, seq));
  783. set_normalized_timespec(&now, now.tv_sec + mono.tv_sec,
  784. now.tv_nsec + mono.tv_nsec);
  785. return now;
  786. }