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