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