LCOV - code coverage report
Current view: top level - kernel/time - timekeeping.c (source / functions) Hit Total Coverage
Test: landlock.info Lines: 394 873 45.1 %
Date: 2021-04-22 12:43:58 Functions: 33 63 52.4 %

          Line data    Source code
       1             : // SPDX-License-Identifier: GPL-2.0
       2             : /*
       3             :  *  Kernel timekeeping code and accessor functions. Based on code from
       4             :  *  timer.c, moved in commit 8524070b7982.
       5             :  */
       6             : #include <linux/timekeeper_internal.h>
       7             : #include <linux/module.h>
       8             : #include <linux/interrupt.h>
       9             : #include <linux/percpu.h>
      10             : #include <linux/init.h>
      11             : #include <linux/mm.h>
      12             : #include <linux/nmi.h>
      13             : #include <linux/sched.h>
      14             : #include <linux/sched/loadavg.h>
      15             : #include <linux/sched/clock.h>
      16             : #include <linux/syscore_ops.h>
      17             : #include <linux/clocksource.h>
      18             : #include <linux/jiffies.h>
      19             : #include <linux/time.h>
      20             : #include <linux/tick.h>
      21             : #include <linux/stop_machine.h>
      22             : #include <linux/pvclock_gtod.h>
      23             : #include <linux/compiler.h>
      24             : #include <linux/audit.h>
      25             : 
      26             : #include "tick-internal.h"
      27             : #include "ntp_internal.h"
      28             : #include "timekeeping_internal.h"
      29             : 
      30             : #define TK_CLEAR_NTP            (1 << 0)
      31             : #define TK_MIRROR               (1 << 1)
      32             : #define TK_CLOCK_WAS_SET        (1 << 2)
      33             : 
      34             : enum timekeeping_adv_mode {
      35             :         /* Update timekeeper when a tick has passed */
      36             :         TK_ADV_TICK,
      37             : 
      38             :         /* Update timekeeper on a direct frequency change */
      39             :         TK_ADV_FREQ
      40             : };
      41             : 
      42             : DEFINE_RAW_SPINLOCK(timekeeper_lock);
      43             : 
      44             : /*
      45             :  * The most important data for readout fits into a single 64 byte
      46             :  * cache line.
      47             :  */
      48             : static struct {
      49             :         seqcount_raw_spinlock_t seq;
      50             :         struct timekeeper       timekeeper;
      51             : } tk_core ____cacheline_aligned = {
      52             :         .seq = SEQCNT_RAW_SPINLOCK_ZERO(tk_core.seq, &timekeeper_lock),
      53             : };
      54             : 
      55             : static struct timekeeper shadow_timekeeper;
      56             : 
      57             : /* flag for if timekeeping is suspended */
      58             : int __read_mostly timekeeping_suspended;
      59             : 
      60             : /**
      61             :  * struct tk_fast - NMI safe timekeeper
      62             :  * @seq:        Sequence counter for protecting updates. The lowest bit
      63             :  *              is the index for the tk_read_base array
      64             :  * @base:       tk_read_base array. Access is indexed by the lowest bit of
      65             :  *              @seq.
      66             :  *
      67             :  * See @update_fast_timekeeper() below.
      68             :  */
      69             : struct tk_fast {
      70             :         seqcount_latch_t        seq;
      71             :         struct tk_read_base     base[2];
      72             : };
      73             : 
      74             : /* Suspend-time cycles value for halted fast timekeeper. */
      75             : static u64 cycles_at_suspend;
      76             : 
      77           0 : static u64 dummy_clock_read(struct clocksource *cs)
      78             : {
      79           0 :         if (timekeeping_suspended)
      80           0 :                 return cycles_at_suspend;
      81           0 :         return local_clock();
      82             : }
      83             : 
      84             : static struct clocksource dummy_clock = {
      85             :         .read = dummy_clock_read,
      86             : };
      87             : 
      88             : /*
      89             :  * Boot time initialization which allows local_clock() to be utilized
      90             :  * during early boot when clocksources are not available. local_clock()
      91             :  * returns nanoseconds already so no conversion is required, hence mult=1
      92             :  * and shift=0. When the first proper clocksource is installed then
      93             :  * the fast time keepers are updated with the correct values.
      94             :  */
      95             : #define FAST_TK_INIT                                            \
      96             :         {                                                       \
      97             :                 .clock          = &dummy_clock,                     \
      98             :                 .mask           = CLOCKSOURCE_MASK(64),         \
      99             :                 .mult           = 1,                            \
     100             :                 .shift          = 0,                            \
     101             :         }
     102             : 
     103             : static struct tk_fast tk_fast_mono ____cacheline_aligned = {
     104             :         .seq     = SEQCNT_LATCH_ZERO(tk_fast_mono.seq),
     105             :         .base[0] = FAST_TK_INIT,
     106             :         .base[1] = FAST_TK_INIT,
     107             : };
     108             : 
     109             : static struct tk_fast tk_fast_raw  ____cacheline_aligned = {
     110             :         .seq     = SEQCNT_LATCH_ZERO(tk_fast_raw.seq),
     111             :         .base[0] = FAST_TK_INIT,
     112             :         .base[1] = FAST_TK_INIT,
     113             : };
     114             : 
     115           1 : static inline void tk_normalize_xtime(struct timekeeper *tk)
     116             : {
     117           1 :         while (tk->tkr_mono.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_mono.shift)) {
     118           0 :                 tk->tkr_mono.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
     119           0 :                 tk->xtime_sec++;
     120             :         }
     121           1 :         while (tk->tkr_raw.xtime_nsec >= ((u64)NSEC_PER_SEC << tk->tkr_raw.shift)) {
     122           0 :                 tk->tkr_raw.xtime_nsec -= (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
     123           0 :                 tk->raw_sec++;
     124             :         }
     125           1 : }
     126             : 
     127       79951 : static inline struct timespec64 tk_xtime(const struct timekeeper *tk)
     128             : {
     129       79951 :         struct timespec64 ts;
     130             : 
     131       79951 :         ts.tv_sec = tk->xtime_sec;
     132       79951 :         ts.tv_nsec = (long)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
     133       79951 :         return ts;
     134             : }
     135             : 
     136           1 : static void tk_set_xtime(struct timekeeper *tk, const struct timespec64 *ts)
     137             : {
     138           1 :         tk->xtime_sec = ts->tv_sec;
     139           1 :         tk->tkr_mono.xtime_nsec = (u64)ts->tv_nsec << tk->tkr_mono.shift;
     140           0 : }
     141             : 
     142           0 : static void tk_xtime_add(struct timekeeper *tk, const struct timespec64 *ts)
     143             : {
     144           0 :         tk->xtime_sec += ts->tv_sec;
     145           0 :         tk->tkr_mono.xtime_nsec += (u64)ts->tv_nsec << tk->tkr_mono.shift;
     146           0 :         tk_normalize_xtime(tk);
     147           0 : }
     148             : 
     149           1 : static void tk_set_wall_to_mono(struct timekeeper *tk, struct timespec64 wtm)
     150             : {
     151           1 :         struct timespec64 tmp;
     152             : 
     153             :         /*
     154             :          * Verify consistency of: offset_real = -wall_to_monotonic
     155             :          * before modifying anything
     156             :          */
     157           1 :         set_normalized_timespec64(&tmp, -tk->wall_to_monotonic.tv_sec,
     158           1 :                                         -tk->wall_to_monotonic.tv_nsec);
     159           2 :         WARN_ON_ONCE(tk->offs_real != timespec64_to_ktime(tmp));
     160           1 :         tk->wall_to_monotonic = wtm;
     161           1 :         set_normalized_timespec64(&tmp, -wtm.tv_sec, -wtm.tv_nsec);
     162           1 :         tk->offs_real = timespec64_to_ktime(tmp);
     163           1 :         tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tk->tai_offset, 0));
     164           1 : }
     165             : 
     166           0 : static inline void tk_update_sleep_time(struct timekeeper *tk, ktime_t delta)
     167             : {
     168           0 :         tk->offs_boot = ktime_add(tk->offs_boot, delta);
     169             :         /*
     170             :          * Timespec representation for VDSO update to avoid 64bit division
     171             :          * on every update.
     172             :          */
     173           0 :         tk->monotonic_to_boot = ktime_to_timespec64(tk->offs_boot);
     174             : }
     175             : 
     176             : /*
     177             :  * tk_clock_read - atomic clocksource read() helper
     178             :  *
     179             :  * This helper is necessary to use in the read paths because, while the
     180             :  * seqcount ensures we don't return a bad value while structures are updated,
     181             :  * it doesn't protect from potential crashes. There is the possibility that
     182             :  * the tkr's clocksource may change between the read reference, and the
     183             :  * clock reference passed to the read function.  This can cause crashes if
     184             :  * the wrong clocksource is passed to the wrong read function.
     185             :  * This isn't necessary to use when holding the timekeeper_lock or doing
     186             :  * a read of the fast-timekeeper tkrs (which is protected by its own locking
     187             :  * and update logic).
     188             :  */
     189      148444 : static inline u64 tk_clock_read(const struct tk_read_base *tkr)
     190             : {
     191      148444 :         struct clocksource *clock = READ_ONCE(tkr->clock);
     192             : 
     193      148444 :         return clock->read(clock);
     194             : }
     195             : 
     196             : #ifdef CONFIG_DEBUG_TIMEKEEPING
     197             : #define WARNING_FREQ (HZ*300) /* 5 minute rate-limiting */
     198             : 
     199             : static void timekeeping_check_update(struct timekeeper *tk, u64 offset)
     200             : {
     201             : 
     202             :         u64 max_cycles = tk->tkr_mono.clock->max_cycles;
     203             :         const char *name = tk->tkr_mono.clock->name;
     204             : 
     205             :         if (offset > max_cycles) {
     206             :                 printk_deferred("WARNING: timekeeping: Cycle offset (%lld) is larger than allowed by the '%s' clock's max_cycles value (%lld): time overflow danger\n",
     207             :                                 offset, name, max_cycles);
     208             :                 printk_deferred("         timekeeping: Your kernel is sick, but tries to cope by capping time updates\n");
     209             :         } else {
     210             :                 if (offset > (max_cycles >> 1)) {
     211             :                         printk_deferred("INFO: timekeeping: Cycle offset (%lld) is larger than the '%s' clock's 50%% safety margin (%lld)\n",
     212             :                                         offset, name, max_cycles >> 1);
     213             :                         printk_deferred("      timekeeping: Your kernel is still fine, but is feeling a bit nervous\n");
     214             :                 }
     215             :         }
     216             : 
     217             :         if (tk->underflow_seen) {
     218             :                 if (jiffies - tk->last_warning > WARNING_FREQ) {
     219             :                         printk_deferred("WARNING: Underflow in clocksource '%s' observed, time update ignored.\n", name);
     220             :                         printk_deferred("         Please report this, consider using a different clocksource, if possible.\n");
     221             :                         printk_deferred("         Your kernel is probably still fine.\n");
     222             :                         tk->last_warning = jiffies;
     223             :                 }
     224             :                 tk->underflow_seen = 0;
     225             :         }
     226             : 
     227             :         if (tk->overflow_seen) {
     228             :                 if (jiffies - tk->last_warning > WARNING_FREQ) {
     229             :                         printk_deferred("WARNING: Overflow in clocksource '%s' observed, time update capped.\n", name);
     230             :                         printk_deferred("         Please report this, consider using a different clocksource, if possible.\n");
     231             :                         printk_deferred("         Your kernel is probably still fine.\n");
     232             :                         tk->last_warning = jiffies;
     233             :                 }
     234             :                 tk->overflow_seen = 0;
     235             :         }
     236             : }
     237             : 
     238             : static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
     239             : {
     240             :         struct timekeeper *tk = &tk_core.timekeeper;
     241             :         u64 now, last, mask, max, delta;
     242             :         unsigned int seq;
     243             : 
     244             :         /*
     245             :          * Since we're called holding a seqcount, the data may shift
     246             :          * under us while we're doing the calculation. This can cause
     247             :          * false positives, since we'd note a problem but throw the
     248             :          * results away. So nest another seqcount here to atomically
     249             :          * grab the points we are checking with.
     250             :          */
     251             :         do {
     252             :                 seq = read_seqcount_begin(&tk_core.seq);
     253             :                 now = tk_clock_read(tkr);
     254             :                 last = tkr->cycle_last;
     255             :                 mask = tkr->mask;
     256             :                 max = tkr->clock->max_cycles;
     257             :         } while (read_seqcount_retry(&tk_core.seq, seq));
     258             : 
     259             :         delta = clocksource_delta(now, last, mask);
     260             : 
     261             :         /*
     262             :          * Try to catch underflows by checking if we are seeing small
     263             :          * mask-relative negative values.
     264             :          */
     265             :         if (unlikely((~delta & mask) < (mask >> 3))) {
     266             :                 tk->underflow_seen = 1;
     267             :                 delta = 0;
     268             :         }
     269             : 
     270             :         /* Cap delta value to the max_cycles values to avoid mult overflows */
     271             :         if (unlikely(delta > max)) {
     272             :                 tk->overflow_seen = 1;
     273             :                 delta = tkr->clock->max_cycles;
     274             :         }
     275             : 
     276             :         return delta;
     277             : }
     278             : #else
     279        7943 : static inline void timekeeping_check_update(struct timekeeper *tk, u64 offset)
     280             : {
     281        7943 : }
     282      140374 : static inline u64 timekeeping_get_delta(const struct tk_read_base *tkr)
     283             : {
     284      140374 :         u64 cycle_now, delta;
     285             : 
     286             :         /* read clocksource */
     287      140374 :         cycle_now = tk_clock_read(tkr);
     288             : 
     289             :         /* calculate the delta since the last update_wall_time */
     290      141365 :         delta = clocksource_delta(cycle_now, tkr->cycle_last, tkr->mask);
     291             : 
     292      141365 :         return delta;
     293             : }
     294             : #endif
     295             : 
     296             : /**
     297             :  * tk_setup_internals - Set up internals to use clocksource clock.
     298             :  *
     299             :  * @tk:         The target timekeeper to setup.
     300             :  * @clock:              Pointer to clocksource.
     301             :  *
     302             :  * Calculates a fixed cycle/nsec interval for a given clocksource/adjustment
     303             :  * pair and interval request.
     304             :  *
     305             :  * Unless you're the timekeeping code, you should not be using this!
     306             :  */
     307           2 : static void tk_setup_internals(struct timekeeper *tk, struct clocksource *clock)
     308             : {
     309           2 :         u64 interval;
     310           2 :         u64 tmp, ntpinterval;
     311           2 :         struct clocksource *old_clock;
     312             : 
     313           2 :         ++tk->cs_was_changed_seq;
     314           2 :         old_clock = tk->tkr_mono.clock;
     315           2 :         tk->tkr_mono.clock = clock;
     316           2 :         tk->tkr_mono.mask = clock->mask;
     317           2 :         tk->tkr_mono.cycle_last = tk_clock_read(&tk->tkr_mono);
     318             : 
     319           2 :         tk->tkr_raw.clock = clock;
     320           2 :         tk->tkr_raw.mask = clock->mask;
     321           2 :         tk->tkr_raw.cycle_last = tk->tkr_mono.cycle_last;
     322             : 
     323             :         /* Do the ns -> cycle conversion first, using original mult */
     324           2 :         tmp = NTP_INTERVAL_LENGTH;
     325           2 :         tmp <<= clock->shift;
     326           2 :         ntpinterval = tmp;
     327           2 :         tmp += clock->mult/2;
     328           2 :         do_div(tmp, clock->mult);
     329           2 :         if (tmp == 0)
     330           0 :                 tmp = 1;
     331             : 
     332           2 :         interval = (u64) tmp;
     333           2 :         tk->cycle_interval = interval;
     334             : 
     335             :         /* Go back from cycles -> shifted ns */
     336           2 :         tk->xtime_interval = interval * clock->mult;
     337           2 :         tk->xtime_remainder = ntpinterval - tk->xtime_interval;
     338           2 :         tk->raw_interval = interval * clock->mult;
     339             : 
     340             :          /* if changing clocks, convert xtime_nsec shift units */
     341           2 :         if (old_clock) {
     342           1 :                 int shift_change = clock->shift - old_clock->shift;
     343           1 :                 if (shift_change < 0) {
     344           0 :                         tk->tkr_mono.xtime_nsec >>= -shift_change;
     345           0 :                         tk->tkr_raw.xtime_nsec >>= -shift_change;
     346             :                 } else {
     347           1 :                         tk->tkr_mono.xtime_nsec <<= shift_change;
     348           1 :                         tk->tkr_raw.xtime_nsec <<= shift_change;
     349             :                 }
     350             :         }
     351             : 
     352           2 :         tk->tkr_mono.shift = clock->shift;
     353           2 :         tk->tkr_raw.shift = clock->shift;
     354             : 
     355           2 :         tk->ntp_error = 0;
     356           2 :         tk->ntp_error_shift = NTP_SCALE_SHIFT - clock->shift;
     357           2 :         tk->ntp_tick = ntpinterval << tk->ntp_error_shift;
     358             : 
     359             :         /*
     360             :          * The timekeeper keeps its own mult values for the currently
     361             :          * active clocksource. These value will be adjusted via NTP
     362             :          * to counteract clock drifting.
     363             :          */
     364           2 :         tk->tkr_mono.mult = clock->mult;
     365           2 :         tk->tkr_raw.mult = clock->mult;
     366           2 :         tk->ntp_err_mult = 0;
     367           2 :         tk->skip_second_overflow = 0;
     368           2 : }
     369             : 
     370             : /* Timekeeper helper functions. */
     371             : 
     372      141504 : static inline u64 timekeeping_delta_to_ns(const struct tk_read_base *tkr, u64 delta)
     373             : {
     374      141504 :         u64 nsec;
     375             : 
     376      141504 :         nsec = delta * tkr->mult + tkr->xtime_nsec;
     377      141504 :         nsec >>= tkr->shift;
     378             : 
     379      141504 :         return nsec;
     380             : }
     381             : 
     382      140362 : static inline u64 timekeeping_get_ns(const struct tk_read_base *tkr)
     383             : {
     384      140362 :         u64 delta;
     385             : 
     386      140362 :         delta = timekeeping_get_delta(tkr);
     387      141382 :         return timekeeping_delta_to_ns(tkr, delta);
     388             : }
     389             : 
     390           0 : static inline u64 timekeeping_cycles_to_ns(const struct tk_read_base *tkr, u64 cycles)
     391             : {
     392           0 :         u64 delta;
     393             : 
     394             :         /* calculate the delta since the last update_wall_time */
     395           0 :         delta = clocksource_delta(cycles, tkr->cycle_last, tkr->mask);
     396           0 :         return timekeeping_delta_to_ns(tkr, delta);
     397             : }
     398             : 
     399             : /**
     400             :  * update_fast_timekeeper - Update the fast and NMI safe monotonic timekeeper.
     401             :  * @tkr: Timekeeping readout base from which we take the update
     402             :  * @tkf: Pointer to NMI safe timekeeper
     403             :  *
     404             :  * We want to use this from any context including NMI and tracing /
     405             :  * instrumenting the timekeeping code itself.
     406             :  *
     407             :  * Employ the latch technique; see @raw_write_seqcount_latch.
     408             :  *
     409             :  * So if a NMI hits the update of base[0] then it will use base[1]
     410             :  * which is still consistent. In the worst case this can result is a
     411             :  * slightly wrong timestamp (a few nanoseconds). See
     412             :  * @ktime_get_mono_fast_ns.
     413             :  */
     414       15894 : static void update_fast_timekeeper(const struct tk_read_base *tkr,
     415             :                                    struct tk_fast *tkf)
     416             : {
     417       15894 :         struct tk_read_base *base = tkf->base;
     418             : 
     419             :         /* Force readers off to base[1] */
     420       15894 :         raw_write_seqcount_latch(&tkf->seq);
     421             : 
     422             :         /* Update base[0] */
     423       15894 :         memcpy(base, tkr, sizeof(*base));
     424             : 
     425             :         /* Force readers back to base[0] */
     426       15894 :         raw_write_seqcount_latch(&tkf->seq);
     427             : 
     428             :         /* Update base[1] */
     429       15894 :         memcpy(base + 1, base, sizeof(*base));
     430       15894 : }
     431             : 
     432         122 : static __always_inline u64 __ktime_get_fast_ns(struct tk_fast *tkf)
     433             : {
     434         122 :         struct tk_read_base *tkr;
     435         122 :         unsigned int seq;
     436         122 :         u64 now;
     437             : 
     438         122 :         do {
     439         122 :                 seq = raw_read_seqcount_latch(&tkf->seq);
     440         122 :                 tkr = tkf->base + (seq & 0x01);
     441         122 :                 now = ktime_to_ns(tkr->base);
     442             : 
     443         122 :                 now += timekeeping_delta_to_ns(tkr,
     444             :                                 clocksource_delta(
     445             :                                         tk_clock_read(tkr),
     446             :                                         tkr->cycle_last,
     447             :                                         tkr->mask));
     448         122 :         } while (read_seqcount_latch_retry(&tkf->seq, seq));
     449             : 
     450         122 :         return now;
     451             : }
     452             : 
     453             : /**
     454             :  * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
     455             :  *
     456             :  * This timestamp is not guaranteed to be monotonic across an update.
     457             :  * The timestamp is calculated by:
     458             :  *
     459             :  *      now = base_mono + clock_delta * slope
     460             :  *
     461             :  * So if the update lowers the slope, readers who are forced to the
     462             :  * not yet updated second array are still using the old steeper slope.
     463             :  *
     464             :  * tmono
     465             :  * ^
     466             :  * |    o  n
     467             :  * |   o n
     468             :  * |  u
     469             :  * | o
     470             :  * |o
     471             :  * |12345678---> reader order
     472             :  *
     473             :  * o = old slope
     474             :  * u = update
     475             :  * n = new slope
     476             :  *
     477             :  * So reader 6 will observe time going backwards versus reader 5.
     478             :  *
     479             :  * While other CPUs are likely to be able to observe that, the only way
     480             :  * for a CPU local observation is when an NMI hits in the middle of
     481             :  * the update. Timestamps taken from that NMI context might be ahead
     482             :  * of the following timestamps. Callers need to be aware of that and
     483             :  * deal with it.
     484             :  */
     485         122 : u64 ktime_get_mono_fast_ns(void)
     486             : {
     487         122 :         return __ktime_get_fast_ns(&tk_fast_mono);
     488             : }
     489             : EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
     490             : 
     491             : /**
     492             :  * ktime_get_raw_fast_ns - Fast NMI safe access to clock monotonic raw
     493             :  *
     494             :  * Contrary to ktime_get_mono_fast_ns() this is always correct because the
     495             :  * conversion factor is not affected by NTP/PTP correction.
     496             :  */
     497           0 : u64 ktime_get_raw_fast_ns(void)
     498             : {
     499           0 :         return __ktime_get_fast_ns(&tk_fast_raw);
     500             : }
     501             : EXPORT_SYMBOL_GPL(ktime_get_raw_fast_ns);
     502             : 
     503             : /**
     504             :  * ktime_get_boot_fast_ns - NMI safe and fast access to boot clock.
     505             :  *
     506             :  * To keep it NMI safe since we're accessing from tracing, we're not using a
     507             :  * separate timekeeper with updates to monotonic clock and boot offset
     508             :  * protected with seqcounts. This has the following minor side effects:
     509             :  *
     510             :  * (1) Its possible that a timestamp be taken after the boot offset is updated
     511             :  * but before the timekeeper is updated. If this happens, the new boot offset
     512             :  * is added to the old timekeeping making the clock appear to update slightly
     513             :  * earlier:
     514             :  *    CPU 0                                        CPU 1
     515             :  *    timekeeping_inject_sleeptime64()
     516             :  *    __timekeeping_inject_sleeptime(tk, delta);
     517             :  *                                                 timestamp();
     518             :  *    timekeeping_update(tk, TK_CLEAR_NTP...);
     519             :  *
     520             :  * (2) On 32-bit systems, the 64-bit boot offset (tk->offs_boot) may be
     521             :  * partially updated.  Since the tk->offs_boot update is a rare event, this
     522             :  * should be a rare occurrence which postprocessing should be able to handle.
     523             :  *
     524             :  * The caveats vs. timestamp ordering as documented for ktime_get_fast_ns()
     525             :  * apply as well.
     526             :  */
     527           0 : u64 notrace ktime_get_boot_fast_ns(void)
     528             : {
     529           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     530             : 
     531           0 :         return (ktime_get_mono_fast_ns() + ktime_to_ns(tk->offs_boot));
     532             : }
     533             : EXPORT_SYMBOL_GPL(ktime_get_boot_fast_ns);
     534             : 
     535           0 : static __always_inline u64 __ktime_get_real_fast(struct tk_fast *tkf, u64 *mono)
     536             : {
     537           0 :         struct tk_read_base *tkr;
     538           0 :         u64 basem, baser, delta;
     539           0 :         unsigned int seq;
     540             : 
     541           0 :         do {
     542           0 :                 seq = raw_read_seqcount_latch(&tkf->seq);
     543           0 :                 tkr = tkf->base + (seq & 0x01);
     544           0 :                 basem = ktime_to_ns(tkr->base);
     545           0 :                 baser = ktime_to_ns(tkr->base_real);
     546             : 
     547           0 :                 delta = timekeeping_delta_to_ns(tkr,
     548             :                                 clocksource_delta(tk_clock_read(tkr),
     549             :                                 tkr->cycle_last, tkr->mask));
     550           0 :         } while (read_seqcount_latch_retry(&tkf->seq, seq));
     551             : 
     552           0 :         if (mono)
     553           0 :                 *mono = basem + delta;
     554           0 :         return baser + delta;
     555             : }
     556             : 
     557             : /**
     558             :  * ktime_get_real_fast_ns: - NMI safe and fast access to clock realtime.
     559             :  *
     560             :  * See ktime_get_fast_ns() for documentation of the time stamp ordering.
     561             :  */
     562           0 : u64 ktime_get_real_fast_ns(void)
     563             : {
     564           0 :         return __ktime_get_real_fast(&tk_fast_mono, NULL);
     565             : }
     566             : EXPORT_SYMBOL_GPL(ktime_get_real_fast_ns);
     567             : 
     568             : /**
     569             :  * ktime_get_fast_timestamps: - NMI safe timestamps
     570             :  * @snapshot:   Pointer to timestamp storage
     571             :  *
     572             :  * Stores clock monotonic, boottime and realtime timestamps.
     573             :  *
     574             :  * Boot time is a racy access on 32bit systems if the sleep time injection
     575             :  * happens late during resume and not in timekeeping_resume(). That could
     576             :  * be avoided by expanding struct tk_read_base with boot offset for 32bit
     577             :  * and adding more overhead to the update. As this is a hard to observe
     578             :  * once per resume event which can be filtered with reasonable effort using
     579             :  * the accurate mono/real timestamps, it's probably not worth the trouble.
     580             :  *
     581             :  * Aside of that it might be possible on 32 and 64 bit to observe the
     582             :  * following when the sleep time injection happens late:
     583             :  *
     584             :  * CPU 0                                CPU 1
     585             :  * timekeeping_resume()
     586             :  * ktime_get_fast_timestamps()
     587             :  *      mono, real = __ktime_get_real_fast()
     588             :  *                                      inject_sleep_time()
     589             :  *                                         update boot offset
     590             :  *      boot = mono + bootoffset;
     591             :  *
     592             :  * That means that boot time already has the sleep time adjustment, but
     593             :  * real time does not. On the next readout both are in sync again.
     594             :  *
     595             :  * Preventing this for 64bit is not really feasible without destroying the
     596             :  * careful cache layout of the timekeeper because the sequence count and
     597             :  * struct tk_read_base would then need two cache lines instead of one.
     598             :  *
     599             :  * Access to the time keeper clock source is disabled accross the innermost
     600             :  * steps of suspend/resume. The accessors still work, but the timestamps
     601             :  * are frozen until time keeping is resumed which happens very early.
     602             :  *
     603             :  * For regular suspend/resume there is no observable difference vs. sched
     604             :  * clock, but it might affect some of the nasty low level debug printks.
     605             :  *
     606             :  * OTOH, access to sched clock is not guaranteed accross suspend/resume on
     607             :  * all systems either so it depends on the hardware in use.
     608             :  *
     609             :  * If that turns out to be a real problem then this could be mitigated by
     610             :  * using sched clock in a similar way as during early boot. But it's not as
     611             :  * trivial as on early boot because it needs some careful protection
     612             :  * against the clock monotonic timestamp jumping backwards on resume.
     613             :  */
     614           0 : void ktime_get_fast_timestamps(struct ktime_timestamps *snapshot)
     615             : {
     616           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     617             : 
     618           0 :         snapshot->real = __ktime_get_real_fast(&tk_fast_mono, &snapshot->mono);
     619           0 :         snapshot->boot = snapshot->mono + ktime_to_ns(data_race(tk->offs_boot));
     620           0 : }
     621             : 
     622             : /**
     623             :  * halt_fast_timekeeper - Prevent fast timekeeper from accessing clocksource.
     624             :  * @tk: Timekeeper to snapshot.
     625             :  *
     626             :  * It generally is unsafe to access the clocksource after timekeeping has been
     627             :  * suspended, so take a snapshot of the readout base of @tk and use it as the
     628             :  * fast timekeeper's readout base while suspended.  It will return the same
     629             :  * number of cycles every time until timekeeping is resumed at which time the
     630             :  * proper readout base for the fast timekeeper will be restored automatically.
     631             :  */
     632           0 : static void halt_fast_timekeeper(const struct timekeeper *tk)
     633             : {
     634           0 :         static struct tk_read_base tkr_dummy;
     635           0 :         const struct tk_read_base *tkr = &tk->tkr_mono;
     636             : 
     637           0 :         memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
     638           0 :         cycles_at_suspend = tk_clock_read(tkr);
     639           0 :         tkr_dummy.clock = &dummy_clock;
     640           0 :         tkr_dummy.base_real = tkr->base + tk->offs_real;
     641           0 :         update_fast_timekeeper(&tkr_dummy, &tk_fast_mono);
     642             : 
     643           0 :         tkr = &tk->tkr_raw;
     644           0 :         memcpy(&tkr_dummy, tkr, sizeof(tkr_dummy));
     645           0 :         tkr_dummy.clock = &dummy_clock;
     646           0 :         update_fast_timekeeper(&tkr_dummy, &tk_fast_raw);
     647           0 : }
     648             : 
     649             : static RAW_NOTIFIER_HEAD(pvclock_gtod_chain);
     650             : 
     651        7946 : static void update_pvclock_gtod(struct timekeeper *tk, bool was_set)
     652             : {
     653        7946 :         raw_notifier_call_chain(&pvclock_gtod_chain, was_set, tk);
     654             : }
     655             : 
     656             : /**
     657             :  * pvclock_gtod_register_notifier - register a pvclock timedata update listener
     658             :  * @nb: Pointer to the notifier block to register
     659             :  */
     660           0 : int pvclock_gtod_register_notifier(struct notifier_block *nb)
     661             : {
     662           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     663           0 :         unsigned long flags;
     664           0 :         int ret;
     665             : 
     666           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
     667           0 :         ret = raw_notifier_chain_register(&pvclock_gtod_chain, nb);
     668           0 :         update_pvclock_gtod(tk, true);
     669           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
     670             : 
     671           0 :         return ret;
     672             : }
     673             : EXPORT_SYMBOL_GPL(pvclock_gtod_register_notifier);
     674             : 
     675             : /**
     676             :  * pvclock_gtod_unregister_notifier - unregister a pvclock
     677             :  * timedata update listener
     678             :  * @nb: Pointer to the notifier block to unregister
     679             :  */
     680           0 : int pvclock_gtod_unregister_notifier(struct notifier_block *nb)
     681             : {
     682           0 :         unsigned long flags;
     683           0 :         int ret;
     684             : 
     685           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
     686           0 :         ret = raw_notifier_chain_unregister(&pvclock_gtod_chain, nb);
     687           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
     688             : 
     689           0 :         return ret;
     690             : }
     691             : EXPORT_SYMBOL_GPL(pvclock_gtod_unregister_notifier);
     692             : 
     693             : /*
     694             :  * tk_update_leap_state - helper to update the next_leap_ktime
     695             :  */
     696        7946 : static inline void tk_update_leap_state(struct timekeeper *tk)
     697             : {
     698        7946 :         tk->next_leap_ktime = ntp_get_next_leap();
     699        7946 :         if (tk->next_leap_ktime != KTIME_MAX)
     700             :                 /* Convert to monotonic time */
     701           0 :                 tk->next_leap_ktime = ktime_sub(tk->next_leap_ktime, tk->offs_real);
     702        7946 : }
     703             : 
     704             : /*
     705             :  * Update the ktime_t based scalar nsec members of the timekeeper
     706             :  */
     707        7946 : static inline void tk_update_ktime_data(struct timekeeper *tk)
     708             : {
     709        7946 :         u64 seconds;
     710        7946 :         u32 nsec;
     711             : 
     712             :         /*
     713             :          * The xtime based monotonic readout is:
     714             :          *      nsec = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec + now();
     715             :          * The ktime based monotonic readout is:
     716             :          *      nsec = base_mono + now();
     717             :          * ==> base_mono = (xtime_sec + wtm_sec) * 1e9 + wtm_nsec
     718             :          */
     719        7946 :         seconds = (u64)(tk->xtime_sec + tk->wall_to_monotonic.tv_sec);
     720        7946 :         nsec = (u32) tk->wall_to_monotonic.tv_nsec;
     721        7946 :         tk->tkr_mono.base = ns_to_ktime(seconds * NSEC_PER_SEC + nsec);
     722             : 
     723             :         /*
     724             :          * The sum of the nanoseconds portions of xtime and
     725             :          * wall_to_monotonic can be greater/equal one second. Take
     726             :          * this into account before updating tk->ktime_sec.
     727             :          */
     728        7946 :         nsec += (u32)(tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift);
     729        7946 :         if (nsec >= NSEC_PER_SEC)
     730        2431 :                 seconds++;
     731        7946 :         tk->ktime_sec = seconds;
     732             : 
     733             :         /* Update the monotonic raw base */
     734        7946 :         tk->tkr_raw.base = ns_to_ktime(tk->raw_sec * NSEC_PER_SEC);
     735        7946 : }
     736             : 
     737             : /* must hold timekeeper_lock */
     738        7946 : static void timekeeping_update(struct timekeeper *tk, unsigned int action)
     739             : {
     740        7946 :         if (action & TK_CLEAR_NTP) {
     741           1 :                 tk->ntp_error = 0;
     742           1 :                 ntp_clear();
     743             :         }
     744             : 
     745        7946 :         tk_update_leap_state(tk);
     746        7946 :         tk_update_ktime_data(tk);
     747             : 
     748        7946 :         update_vsyscall(tk);
     749        7946 :         update_pvclock_gtod(tk, action & TK_CLOCK_WAS_SET);
     750             : 
     751        7946 :         tk->tkr_mono.base_real = tk->tkr_mono.base + tk->offs_real;
     752        7946 :         update_fast_timekeeper(&tk->tkr_mono, &tk_fast_mono);
     753        7946 :         update_fast_timekeeper(&tk->tkr_raw,  &tk_fast_raw);
     754             : 
     755        7946 :         if (action & TK_CLOCK_WAS_SET)
     756           2 :                 tk->clock_was_set_seq++;
     757             :         /*
     758             :          * The mirroring of the data to the shadow-timekeeper needs
     759             :          * to happen last here to ensure we don't over-write the
     760             :          * timekeeper structure on the next update with stale data
     761             :          */
     762        7946 :         if (action & TK_MIRROR)
     763           2 :                 memcpy(&shadow_timekeeper, &tk_core.timekeeper,
     764             :                        sizeof(tk_core.timekeeper));
     765        7946 : }
     766             : 
     767             : /**
     768             :  * timekeeping_forward_now - update clock to the current time
     769             :  * @tk:         Pointer to the timekeeper to update
     770             :  *
     771             :  * Forward the current clock to update its state since the last call to
     772             :  * update_wall_time(). This is useful before significant clock changes,
     773             :  * as it avoids having to deal with this time offset explicitly.
     774             :  */
     775           1 : static void timekeeping_forward_now(struct timekeeper *tk)
     776             : {
     777           1 :         u64 cycle_now, delta;
     778             : 
     779           1 :         cycle_now = tk_clock_read(&tk->tkr_mono);
     780           1 :         delta = clocksource_delta(cycle_now, tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
     781           1 :         tk->tkr_mono.cycle_last = cycle_now;
     782           1 :         tk->tkr_raw.cycle_last  = cycle_now;
     783             : 
     784           1 :         tk->tkr_mono.xtime_nsec += delta * tk->tkr_mono.mult;
     785           1 :         tk->tkr_raw.xtime_nsec += delta * tk->tkr_raw.mult;
     786             : 
     787           1 :         tk_normalize_xtime(tk);
     788           1 : }
     789             : 
     790             : /**
     791             :  * ktime_get_real_ts64 - Returns the time of day in a timespec64.
     792             :  * @ts:         pointer to the timespec to be set
     793             :  *
     794             :  * Returns the time of day in a timespec64 (WARN if suspended).
     795             :  */
     796         342 : void ktime_get_real_ts64(struct timespec64 *ts)
     797             : {
     798         342 :         struct timekeeper *tk = &tk_core.timekeeper;
     799         342 :         unsigned int seq;
     800         342 :         u64 nsecs;
     801             : 
     802         342 :         WARN_ON(timekeeping_suspended);
     803             : 
     804         342 :         do {
     805         342 :                 seq = read_seqcount_begin(&tk_core.seq);
     806             : 
     807         342 :                 ts->tv_sec = tk->xtime_sec;
     808         342 :                 nsecs = timekeeping_get_ns(&tk->tkr_mono);
     809             : 
     810         342 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     811             : 
     812         342 :         ts->tv_nsec = 0;
     813         342 :         timespec64_add_ns(ts, nsecs);
     814         342 : }
     815             : EXPORT_SYMBOL(ktime_get_real_ts64);
     816             : 
     817      107799 : ktime_t ktime_get(void)
     818             : {
     819      107799 :         struct timekeeper *tk = &tk_core.timekeeper;
     820      107799 :         unsigned int seq;
     821      107799 :         ktime_t base;
     822      107799 :         u64 nsecs;
     823             : 
     824      107799 :         WARN_ON(timekeeping_suspended);
     825             : 
     826      107936 :         do {
     827      268689 :                 seq = read_seqcount_begin(&tk_core.seq);
     828      109932 :                 base = tk->tkr_mono.base;
     829      109932 :                 nsecs = timekeeping_get_ns(&tk->tkr_mono);
     830             : 
     831      110373 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     832             : 
     833      110263 :         return ktime_add_ns(base, nsecs);
     834             : }
     835             : EXPORT_SYMBOL_GPL(ktime_get);
     836             : 
     837           0 : u32 ktime_get_resolution_ns(void)
     838             : {
     839           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     840           0 :         unsigned int seq;
     841           0 :         u32 nsecs;
     842             : 
     843           0 :         WARN_ON(timekeeping_suspended);
     844             : 
     845           0 :         do {
     846           0 :                 seq = read_seqcount_begin(&tk_core.seq);
     847           0 :                 nsecs = tk->tkr_mono.mult >> tk->tkr_mono.shift;
     848           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     849             : 
     850           0 :         return nsecs;
     851             : }
     852             : EXPORT_SYMBOL_GPL(ktime_get_resolution_ns);
     853             : 
     854             : static ktime_t *offsets[TK_OFFS_MAX] = {
     855             :         [TK_OFFS_REAL]  = &tk_core.timekeeper.offs_real,
     856             :         [TK_OFFS_BOOT]  = &tk_core.timekeeper.offs_boot,
     857             :         [TK_OFFS_TAI]   = &tk_core.timekeeper.offs_tai,
     858             : };
     859             : 
     860        1635 : ktime_t ktime_get_with_offset(enum tk_offsets offs)
     861             : {
     862        1635 :         struct timekeeper *tk = &tk_core.timekeeper;
     863        1635 :         unsigned int seq;
     864        1635 :         ktime_t base, *offset = offsets[offs];
     865        1635 :         u64 nsecs;
     866             : 
     867        1635 :         WARN_ON(timekeeping_suspended);
     868             : 
     869        1635 :         do {
     870        1635 :                 seq = read_seqcount_begin(&tk_core.seq);
     871        1635 :                 base = ktime_add(tk->tkr_mono.base, *offset);
     872        1635 :                 nsecs = timekeeping_get_ns(&tk->tkr_mono);
     873             : 
     874        1635 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     875             : 
     876        1635 :         return ktime_add_ns(base, nsecs);
     877             : 
     878             : }
     879             : EXPORT_SYMBOL_GPL(ktime_get_with_offset);
     880             : 
     881           0 : ktime_t ktime_get_coarse_with_offset(enum tk_offsets offs)
     882             : {
     883           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     884           0 :         unsigned int seq;
     885           0 :         ktime_t base, *offset = offsets[offs];
     886           0 :         u64 nsecs;
     887             : 
     888           0 :         WARN_ON(timekeeping_suspended);
     889             : 
     890           0 :         do {
     891           0 :                 seq = read_seqcount_begin(&tk_core.seq);
     892           0 :                 base = ktime_add(tk->tkr_mono.base, *offset);
     893           0 :                 nsecs = tk->tkr_mono.xtime_nsec >> tk->tkr_mono.shift;
     894             : 
     895           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     896             : 
     897           0 :         return ktime_add_ns(base, nsecs);
     898             : }
     899             : EXPORT_SYMBOL_GPL(ktime_get_coarse_with_offset);
     900             : 
     901             : /**
     902             :  * ktime_mono_to_any() - convert mononotic time to any other time
     903             :  * @tmono:      time to convert.
     904             :  * @offs:       which offset to use
     905             :  */
     906          24 : ktime_t ktime_mono_to_any(ktime_t tmono, enum tk_offsets offs)
     907             : {
     908          24 :         ktime_t *offset = offsets[offs];
     909          24 :         unsigned int seq;
     910          24 :         ktime_t tconv;
     911             : 
     912          24 :         do {
     913          24 :                 seq = read_seqcount_begin(&tk_core.seq);
     914          24 :                 tconv = ktime_add(tmono, *offset);
     915          24 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     916             : 
     917          24 :         return tconv;
     918             : }
     919             : EXPORT_SYMBOL_GPL(ktime_mono_to_any);
     920             : 
     921             : /**
     922             :  * ktime_get_raw - Returns the raw monotonic time in ktime_t format
     923             :  */
     924           0 : ktime_t ktime_get_raw(void)
     925             : {
     926           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     927           0 :         unsigned int seq;
     928           0 :         ktime_t base;
     929           0 :         u64 nsecs;
     930             : 
     931           0 :         do {
     932           0 :                 seq = read_seqcount_begin(&tk_core.seq);
     933           0 :                 base = tk->tkr_raw.base;
     934           0 :                 nsecs = timekeeping_get_ns(&tk->tkr_raw);
     935             : 
     936           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     937             : 
     938           0 :         return ktime_add_ns(base, nsecs);
     939             : }
     940             : EXPORT_SYMBOL_GPL(ktime_get_raw);
     941             : 
     942             : /**
     943             :  * ktime_get_ts64 - get the monotonic clock in timespec64 format
     944             :  * @ts:         pointer to timespec variable
     945             :  *
     946             :  * The function calculates the monotonic clock from the realtime
     947             :  * clock and the wall_to_monotonic offset and stores the result
     948             :  * in normalized timespec64 format in the variable pointed to by @ts.
     949             :  */
     950         802 : void ktime_get_ts64(struct timespec64 *ts)
     951             : {
     952         802 :         struct timekeeper *tk = &tk_core.timekeeper;
     953         802 :         struct timespec64 tomono;
     954         802 :         unsigned int seq;
     955         802 :         u64 nsec;
     956             : 
     957         802 :         WARN_ON(timekeeping_suspended);
     958             : 
     959         802 :         do {
     960         802 :                 seq = read_seqcount_begin(&tk_core.seq);
     961         802 :                 ts->tv_sec = tk->xtime_sec;
     962         802 :                 nsec = timekeeping_get_ns(&tk->tkr_mono);
     963         802 :                 tomono = tk->wall_to_monotonic;
     964             : 
     965         802 :         } while (read_seqcount_retry(&tk_core.seq, seq));
     966             : 
     967         802 :         ts->tv_sec += tomono.tv_sec;
     968         802 :         ts->tv_nsec = 0;
     969         802 :         timespec64_add_ns(ts, nsec + tomono.tv_nsec);
     970         802 : }
     971             : EXPORT_SYMBOL_GPL(ktime_get_ts64);
     972             : 
     973             : /**
     974             :  * ktime_get_seconds - Get the seconds portion of CLOCK_MONOTONIC
     975             :  *
     976             :  * Returns the seconds portion of CLOCK_MONOTONIC with a single non
     977             :  * serialized read. tk->ktime_sec is of type 'unsigned long' so this
     978             :  * works on both 32 and 64 bit systems. On 32 bit systems the readout
     979             :  * covers ~136 years of uptime which should be enough to prevent
     980             :  * premature wrap arounds.
     981             :  */
     982           0 : time64_t ktime_get_seconds(void)
     983             : {
     984           0 :         struct timekeeper *tk = &tk_core.timekeeper;
     985             : 
     986           0 :         WARN_ON(timekeeping_suspended);
     987           0 :         return tk->ktime_sec;
     988             : }
     989             : EXPORT_SYMBOL_GPL(ktime_get_seconds);
     990             : 
     991             : /**
     992             :  * ktime_get_real_seconds - Get the seconds portion of CLOCK_REALTIME
     993             :  *
     994             :  * Returns the wall clock seconds since 1970.
     995             :  *
     996             :  * For 64bit systems the fast access to tk->xtime_sec is preserved. On
     997             :  * 32bit systems the access must be protected with the sequence
     998             :  * counter to provide "atomic" access to the 64bit tk->xtime_sec
     999             :  * value.
    1000             :  */
    1001         432 : time64_t ktime_get_real_seconds(void)
    1002             : {
    1003         432 :         struct timekeeper *tk = &tk_core.timekeeper;
    1004         432 :         time64_t seconds;
    1005         432 :         unsigned int seq;
    1006             : 
    1007         432 :         if (IS_ENABLED(CONFIG_64BIT))
    1008         432 :                 return tk->xtime_sec;
    1009             : 
    1010             :         do {
    1011             :                 seq = read_seqcount_begin(&tk_core.seq);
    1012             :                 seconds = tk->xtime_sec;
    1013             : 
    1014             :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1015             : 
    1016             :         return seconds;
    1017             : }
    1018             : EXPORT_SYMBOL_GPL(ktime_get_real_seconds);
    1019             : 
    1020             : /**
    1021             :  * __ktime_get_real_seconds - The same as ktime_get_real_seconds
    1022             :  * but without the sequence counter protect. This internal function
    1023             :  * is called just when timekeeping lock is already held.
    1024             :  */
    1025           0 : noinstr time64_t __ktime_get_real_seconds(void)
    1026             : {
    1027           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1028             : 
    1029           0 :         return tk->xtime_sec;
    1030             : }
    1031             : 
    1032             : /**
    1033             :  * ktime_get_snapshot - snapshots the realtime/monotonic raw clocks with counter
    1034             :  * @systime_snapshot:   pointer to struct receiving the system time snapshot
    1035             :  */
    1036           0 : void ktime_get_snapshot(struct system_time_snapshot *systime_snapshot)
    1037             : {
    1038           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1039           0 :         unsigned int seq;
    1040           0 :         ktime_t base_raw;
    1041           0 :         ktime_t base_real;
    1042           0 :         u64 nsec_raw;
    1043           0 :         u64 nsec_real;
    1044           0 :         u64 now;
    1045             : 
    1046           0 :         WARN_ON_ONCE(timekeeping_suspended);
    1047             : 
    1048           0 :         do {
    1049           0 :                 seq = read_seqcount_begin(&tk_core.seq);
    1050           0 :                 now = tk_clock_read(&tk->tkr_mono);
    1051           0 :                 systime_snapshot->cs_was_changed_seq = tk->cs_was_changed_seq;
    1052           0 :                 systime_snapshot->clock_was_set_seq = tk->clock_was_set_seq;
    1053           0 :                 base_real = ktime_add(tk->tkr_mono.base,
    1054             :                                       tk_core.timekeeper.offs_real);
    1055           0 :                 base_raw = tk->tkr_raw.base;
    1056           0 :                 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono, now);
    1057           0 :                 nsec_raw  = timekeeping_cycles_to_ns(&tk->tkr_raw, now);
    1058           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1059             : 
    1060           0 :         systime_snapshot->cycles = now;
    1061           0 :         systime_snapshot->real = ktime_add_ns(base_real, nsec_real);
    1062           0 :         systime_snapshot->raw = ktime_add_ns(base_raw, nsec_raw);
    1063           0 : }
    1064             : EXPORT_SYMBOL_GPL(ktime_get_snapshot);
    1065             : 
    1066             : /* Scale base by mult/div checking for overflow */
    1067           0 : static int scale64_check_overflow(u64 mult, u64 div, u64 *base)
    1068             : {
    1069           0 :         u64 tmp, rem;
    1070             : 
    1071           0 :         tmp = div64_u64_rem(*base, div, &rem);
    1072             : 
    1073           0 :         if (((int)sizeof(u64)*8 - fls64(mult) < fls64(tmp)) ||
    1074           0 :             ((int)sizeof(u64)*8 - fls64(mult) < fls64(rem)))
    1075             :                 return -EOVERFLOW;
    1076           0 :         tmp *= mult;
    1077             : 
    1078           0 :         rem = div64_u64(rem * mult, div);
    1079           0 :         *base = tmp + rem;
    1080           0 :         return 0;
    1081             : }
    1082             : 
    1083             : /**
    1084             :  * adjust_historical_crosststamp - adjust crosstimestamp previous to current interval
    1085             :  * @history:                    Snapshot representing start of history
    1086             :  * @partial_history_cycles:     Cycle offset into history (fractional part)
    1087             :  * @total_history_cycles:       Total history length in cycles
    1088             :  * @discontinuity:              True indicates clock was set on history period
    1089             :  * @ts:                         Cross timestamp that should be adjusted using
    1090             :  *      partial/total ratio
    1091             :  *
    1092             :  * Helper function used by get_device_system_crosststamp() to correct the
    1093             :  * crosstimestamp corresponding to the start of the current interval to the
    1094             :  * system counter value (timestamp point) provided by the driver. The
    1095             :  * total_history_* quantities are the total history starting at the provided
    1096             :  * reference point and ending at the start of the current interval. The cycle
    1097             :  * count between the driver timestamp point and the start of the current
    1098             :  * interval is partial_history_cycles.
    1099             :  */
    1100           0 : static int adjust_historical_crosststamp(struct system_time_snapshot *history,
    1101             :                                          u64 partial_history_cycles,
    1102             :                                          u64 total_history_cycles,
    1103             :                                          bool discontinuity,
    1104             :                                          struct system_device_crosststamp *ts)
    1105             : {
    1106           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1107           0 :         u64 corr_raw, corr_real;
    1108           0 :         bool interp_forward;
    1109           0 :         int ret;
    1110             : 
    1111           0 :         if (total_history_cycles == 0 || partial_history_cycles == 0)
    1112             :                 return 0;
    1113             : 
    1114             :         /* Interpolate shortest distance from beginning or end of history */
    1115           0 :         interp_forward = partial_history_cycles > total_history_cycles / 2;
    1116           0 :         partial_history_cycles = interp_forward ?
    1117           0 :                 total_history_cycles - partial_history_cycles :
    1118             :                 partial_history_cycles;
    1119             : 
    1120             :         /*
    1121             :          * Scale the monotonic raw time delta by:
    1122             :          *      partial_history_cycles / total_history_cycles
    1123             :          */
    1124           0 :         corr_raw = (u64)ktime_to_ns(
    1125           0 :                 ktime_sub(ts->sys_monoraw, history->raw));
    1126           0 :         ret = scale64_check_overflow(partial_history_cycles,
    1127             :                                      total_history_cycles, &corr_raw);
    1128           0 :         if (ret)
    1129             :                 return ret;
    1130             : 
    1131             :         /*
    1132             :          * If there is a discontinuity in the history, scale monotonic raw
    1133             :          *      correction by:
    1134             :          *      mult(real)/mult(raw) yielding the realtime correction
    1135             :          * Otherwise, calculate the realtime correction similar to monotonic
    1136             :          *      raw calculation
    1137             :          */
    1138           0 :         if (discontinuity) {
    1139           0 :                 corr_real = mul_u64_u32_div
    1140             :                         (corr_raw, tk->tkr_mono.mult, tk->tkr_raw.mult);
    1141             :         } else {
    1142           0 :                 corr_real = (u64)ktime_to_ns(
    1143           0 :                         ktime_sub(ts->sys_realtime, history->real));
    1144           0 :                 ret = scale64_check_overflow(partial_history_cycles,
    1145             :                                              total_history_cycles, &corr_real);
    1146           0 :                 if (ret)
    1147             :                         return ret;
    1148             :         }
    1149             : 
    1150             :         /* Fixup monotonic raw and real time time values */
    1151           0 :         if (interp_forward) {
    1152           0 :                 ts->sys_monoraw = ktime_add_ns(history->raw, corr_raw);
    1153           0 :                 ts->sys_realtime = ktime_add_ns(history->real, corr_real);
    1154             :         } else {
    1155           0 :                 ts->sys_monoraw = ktime_sub_ns(ts->sys_monoraw, corr_raw);
    1156           0 :                 ts->sys_realtime = ktime_sub_ns(ts->sys_realtime, corr_real);
    1157             :         }
    1158             : 
    1159             :         return 0;
    1160             : }
    1161             : 
    1162             : /*
    1163             :  * cycle_between - true if test occurs chronologically between before and after
    1164             :  */
    1165           0 : static bool cycle_between(u64 before, u64 test, u64 after)
    1166             : {
    1167           0 :         if (test > before && test < after)
    1168             :                 return true;
    1169           0 :         if (test < before && before > after)
    1170             :                 return true;
    1171             :         return false;
    1172             : }
    1173             : 
    1174             : /**
    1175             :  * get_device_system_crosststamp - Synchronously capture system/device timestamp
    1176             :  * @get_time_fn:        Callback to get simultaneous device time and
    1177             :  *      system counter from the device driver
    1178             :  * @ctx:                Context passed to get_time_fn()
    1179             :  * @history_begin:      Historical reference point used to interpolate system
    1180             :  *      time when counter provided by the driver is before the current interval
    1181             :  * @xtstamp:            Receives simultaneously captured system and device time
    1182             :  *
    1183             :  * Reads a timestamp from a device and correlates it to system time
    1184             :  */
    1185           0 : int get_device_system_crosststamp(int (*get_time_fn)
    1186             :                                   (ktime_t *device_time,
    1187             :                                    struct system_counterval_t *sys_counterval,
    1188             :                                    void *ctx),
    1189             :                                   void *ctx,
    1190             :                                   struct system_time_snapshot *history_begin,
    1191             :                                   struct system_device_crosststamp *xtstamp)
    1192             : {
    1193           0 :         struct system_counterval_t system_counterval;
    1194           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1195           0 :         u64 cycles, now, interval_start;
    1196           0 :         unsigned int clock_was_set_seq = 0;
    1197           0 :         ktime_t base_real, base_raw;
    1198           0 :         u64 nsec_real, nsec_raw;
    1199           0 :         u8 cs_was_changed_seq;
    1200           0 :         unsigned int seq;
    1201           0 :         bool do_interp;
    1202           0 :         int ret;
    1203             : 
    1204           0 :         do {
    1205           0 :                 seq = read_seqcount_begin(&tk_core.seq);
    1206             :                 /*
    1207             :                  * Try to synchronously capture device time and a system
    1208             :                  * counter value calling back into the device driver
    1209             :                  */
    1210           0 :                 ret = get_time_fn(&xtstamp->device, &system_counterval, ctx);
    1211           0 :                 if (ret)
    1212           0 :                         return ret;
    1213             : 
    1214             :                 /*
    1215             :                  * Verify that the clocksource associated with the captured
    1216             :                  * system counter value is the same as the currently installed
    1217             :                  * timekeeper clocksource
    1218             :                  */
    1219           0 :                 if (tk->tkr_mono.clock != system_counterval.cs)
    1220             :                         return -ENODEV;
    1221           0 :                 cycles = system_counterval.cycles;
    1222             : 
    1223             :                 /*
    1224             :                  * Check whether the system counter value provided by the
    1225             :                  * device driver is on the current timekeeping interval.
    1226             :                  */
    1227           0 :                 now = tk_clock_read(&tk->tkr_mono);
    1228           0 :                 interval_start = tk->tkr_mono.cycle_last;
    1229           0 :                 if (!cycle_between(interval_start, cycles, now)) {
    1230           0 :                         clock_was_set_seq = tk->clock_was_set_seq;
    1231           0 :                         cs_was_changed_seq = tk->cs_was_changed_seq;
    1232           0 :                         cycles = interval_start;
    1233           0 :                         do_interp = true;
    1234             :                 } else {
    1235             :                         do_interp = false;
    1236             :                 }
    1237             : 
    1238           0 :                 base_real = ktime_add(tk->tkr_mono.base,
    1239             :                                       tk_core.timekeeper.offs_real);
    1240           0 :                 base_raw = tk->tkr_raw.base;
    1241             : 
    1242           0 :                 nsec_real = timekeeping_cycles_to_ns(&tk->tkr_mono,
    1243             :                                                      system_counterval.cycles);
    1244           0 :                 nsec_raw = timekeeping_cycles_to_ns(&tk->tkr_raw,
    1245             :                                                     system_counterval.cycles);
    1246           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1247             : 
    1248           0 :         xtstamp->sys_realtime = ktime_add_ns(base_real, nsec_real);
    1249           0 :         xtstamp->sys_monoraw = ktime_add_ns(base_raw, nsec_raw);
    1250             : 
    1251             :         /*
    1252             :          * Interpolate if necessary, adjusting back from the start of the
    1253             :          * current interval
    1254             :          */
    1255           0 :         if (do_interp) {
    1256           0 :                 u64 partial_history_cycles, total_history_cycles;
    1257           0 :                 bool discontinuity;
    1258             : 
    1259             :                 /*
    1260             :                  * Check that the counter value occurs after the provided
    1261             :                  * history reference and that the history doesn't cross a
    1262             :                  * clocksource change
    1263             :                  */
    1264           0 :                 if (!history_begin ||
    1265           0 :                     !cycle_between(history_begin->cycles,
    1266           0 :                                    system_counterval.cycles, cycles) ||
    1267           0 :                     history_begin->cs_was_changed_seq != cs_was_changed_seq)
    1268             :                         return -EINVAL;
    1269           0 :                 partial_history_cycles = cycles - system_counterval.cycles;
    1270           0 :                 total_history_cycles = cycles - history_begin->cycles;
    1271           0 :                 discontinuity =
    1272           0 :                         history_begin->clock_was_set_seq != clock_was_set_seq;
    1273             : 
    1274           0 :                 ret = adjust_historical_crosststamp(history_begin,
    1275             :                                                     partial_history_cycles,
    1276             :                                                     total_history_cycles,
    1277             :                                                     discontinuity, xtstamp);
    1278           0 :                 if (ret)
    1279           0 :                         return ret;
    1280             :         }
    1281             : 
    1282             :         return 0;
    1283             : }
    1284             : EXPORT_SYMBOL_GPL(get_device_system_crosststamp);
    1285             : 
    1286             : /**
    1287             :  * do_settimeofday64 - Sets the time of day.
    1288             :  * @ts:     pointer to the timespec64 variable containing the new time
    1289             :  *
    1290             :  * Sets the time of day to the new time and update NTP and notify hrtimers
    1291             :  */
    1292           0 : int do_settimeofday64(const struct timespec64 *ts)
    1293             : {
    1294           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1295           0 :         struct timespec64 ts_delta, xt;
    1296           0 :         unsigned long flags;
    1297           0 :         int ret = 0;
    1298             : 
    1299           0 :         if (!timespec64_valid_settod(ts))
    1300             :                 return -EINVAL;
    1301             : 
    1302           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1303           0 :         write_seqcount_begin(&tk_core.seq);
    1304             : 
    1305           0 :         timekeeping_forward_now(tk);
    1306             : 
    1307           0 :         xt = tk_xtime(tk);
    1308           0 :         ts_delta.tv_sec = ts->tv_sec - xt.tv_sec;
    1309           0 :         ts_delta.tv_nsec = ts->tv_nsec - xt.tv_nsec;
    1310             : 
    1311           0 :         if (timespec64_compare(&tk->wall_to_monotonic, &ts_delta) > 0) {
    1312           0 :                 ret = -EINVAL;
    1313           0 :                 goto out;
    1314             :         }
    1315             : 
    1316           0 :         tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, ts_delta));
    1317             : 
    1318           0 :         tk_set_xtime(tk, ts);
    1319           0 : out:
    1320           0 :         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
    1321             : 
    1322           0 :         write_seqcount_end(&tk_core.seq);
    1323           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1324             : 
    1325             :         /* signal hrtimers about time change */
    1326           0 :         clock_was_set();
    1327             : 
    1328           0 :         if (!ret)
    1329             :                 audit_tk_injoffset(ts_delta);
    1330             : 
    1331             :         return ret;
    1332             : }
    1333             : EXPORT_SYMBOL(do_settimeofday64);
    1334             : 
    1335             : /**
    1336             :  * timekeeping_inject_offset - Adds or subtracts from the current time.
    1337             :  * @ts:         Pointer to the timespec variable containing the offset
    1338             :  *
    1339             :  * Adds or subtracts an offset value from the current time.
    1340             :  */
    1341           0 : static int timekeeping_inject_offset(const struct timespec64 *ts)
    1342             : {
    1343           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1344           0 :         unsigned long flags;
    1345           0 :         struct timespec64 tmp;
    1346           0 :         int ret = 0;
    1347             : 
    1348           0 :         if (ts->tv_nsec < 0 || ts->tv_nsec >= NSEC_PER_SEC)
    1349             :                 return -EINVAL;
    1350             : 
    1351           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1352           0 :         write_seqcount_begin(&tk_core.seq);
    1353             : 
    1354           0 :         timekeeping_forward_now(tk);
    1355             : 
    1356             :         /* Make sure the proposed value is valid */
    1357           0 :         tmp = timespec64_add(tk_xtime(tk), *ts);
    1358           0 :         if (timespec64_compare(&tk->wall_to_monotonic, ts) > 0 ||
    1359           0 :             !timespec64_valid_settod(&tmp)) {
    1360           0 :                 ret = -EINVAL;
    1361           0 :                 goto error;
    1362             :         }
    1363             : 
    1364           0 :         tk_xtime_add(tk, ts);
    1365           0 :         tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *ts));
    1366             : 
    1367           0 : error: /* even if we error out, we forwarded the time, so call update */
    1368           0 :         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
    1369             : 
    1370           0 :         write_seqcount_end(&tk_core.seq);
    1371           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1372             : 
    1373             :         /* signal hrtimers about time change */
    1374           0 :         clock_was_set();
    1375             : 
    1376           0 :         return ret;
    1377             : }
    1378             : 
    1379             : /*
    1380             :  * Indicates if there is an offset between the system clock and the hardware
    1381             :  * clock/persistent clock/rtc.
    1382             :  */
    1383             : int persistent_clock_is_local;
    1384             : 
    1385             : /*
    1386             :  * Adjust the time obtained from the CMOS to be UTC time instead of
    1387             :  * local time.
    1388             :  *
    1389             :  * This is ugly, but preferable to the alternatives.  Otherwise we
    1390             :  * would either need to write a program to do it in /etc/rc (and risk
    1391             :  * confusion if the program gets run more than once; it would also be
    1392             :  * hard to make the program warp the clock precisely n hours)  or
    1393             :  * compile in the timezone information into the kernel.  Bad, bad....
    1394             :  *
    1395             :  *                                              - TYT, 1992-01-01
    1396             :  *
    1397             :  * The best thing to do is to keep the CMOS clock in universal time (UTC)
    1398             :  * as real UNIX machines always do it. This avoids all headaches about
    1399             :  * daylight saving times and warping kernel clocks.
    1400             :  */
    1401           1 : void timekeeping_warp_clock(void)
    1402             : {
    1403           1 :         if (sys_tz.tz_minuteswest != 0) {
    1404           0 :                 struct timespec64 adjust;
    1405             : 
    1406           0 :                 persistent_clock_is_local = 1;
    1407           0 :                 adjust.tv_sec = sys_tz.tz_minuteswest * 60;
    1408           0 :                 adjust.tv_nsec = 0;
    1409           0 :                 timekeeping_inject_offset(&adjust);
    1410             :         }
    1411           1 : }
    1412             : 
    1413             : /*
    1414             :  * __timekeeping_set_tai_offset - Sets the TAI offset from UTC and monotonic
    1415             :  */
    1416           0 : static void __timekeeping_set_tai_offset(struct timekeeper *tk, s32 tai_offset)
    1417             : {
    1418           0 :         tk->tai_offset = tai_offset;
    1419           0 :         tk->offs_tai = ktime_add(tk->offs_real, ktime_set(tai_offset, 0));
    1420             : }
    1421             : 
    1422             : /*
    1423             :  * change_clocksource - Swaps clocksources if a new one is available
    1424             :  *
    1425             :  * Accumulates current time interval and initializes new clocksource
    1426             :  */
    1427           1 : static int change_clocksource(void *data)
    1428             : {
    1429           1 :         struct timekeeper *tk = &tk_core.timekeeper;
    1430           1 :         struct clocksource *new, *old;
    1431           1 :         unsigned long flags;
    1432             : 
    1433           1 :         new = (struct clocksource *) data;
    1434             : 
    1435           1 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1436           2 :         write_seqcount_begin(&tk_core.seq);
    1437             : 
    1438           1 :         timekeeping_forward_now(tk);
    1439             :         /*
    1440             :          * If the cs is in module, get a module reference. Succeeds
    1441             :          * for built-in code (owner == NULL) as well.
    1442             :          */
    1443           1 :         if (try_module_get(new->owner)) {
    1444           1 :                 if (!new->enable || new->enable(new) == 0) {
    1445           1 :                         old = tk->tkr_mono.clock;
    1446           1 :                         tk_setup_internals(tk, new);
    1447           1 :                         if (old->disable)
    1448           0 :                                 old->disable(old);
    1449           1 :                         module_put(old->owner);
    1450             :                 } else {
    1451           1 :                         module_put(new->owner);
    1452             :                 }
    1453             :         }
    1454           1 :         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
    1455             : 
    1456           1 :         write_seqcount_end(&tk_core.seq);
    1457           1 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1458             : 
    1459           1 :         return 0;
    1460             : }
    1461             : 
    1462             : /**
    1463             :  * timekeeping_notify - Install a new clock source
    1464             :  * @clock:              pointer to the clock source
    1465             :  *
    1466             :  * This function is called from clocksource.c after a new, better clock
    1467             :  * source has been registered. The caller holds the clocksource_mutex.
    1468             :  */
    1469           1 : int timekeeping_notify(struct clocksource *clock)
    1470             : {
    1471           1 :         struct timekeeper *tk = &tk_core.timekeeper;
    1472             : 
    1473           1 :         if (tk->tkr_mono.clock == clock)
    1474             :                 return 0;
    1475           1 :         stop_machine(change_clocksource, clock, NULL);
    1476           1 :         tick_clock_notify();
    1477           1 :         return tk->tkr_mono.clock == clock ? 0 : -1;
    1478             : }
    1479             : 
    1480             : /**
    1481             :  * ktime_get_raw_ts64 - Returns the raw monotonic time in a timespec
    1482             :  * @ts:         pointer to the timespec64 to be set
    1483             :  *
    1484             :  * Returns the raw monotonic time (completely un-modified by ntp)
    1485             :  */
    1486           0 : void ktime_get_raw_ts64(struct timespec64 *ts)
    1487             : {
    1488           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1489           0 :         unsigned int seq;
    1490           0 :         u64 nsecs;
    1491             : 
    1492           0 :         do {
    1493           0 :                 seq = read_seqcount_begin(&tk_core.seq);
    1494           0 :                 ts->tv_sec = tk->raw_sec;
    1495           0 :                 nsecs = timekeeping_get_ns(&tk->tkr_raw);
    1496             : 
    1497           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1498             : 
    1499           0 :         ts->tv_nsec = 0;
    1500           0 :         timespec64_add_ns(ts, nsecs);
    1501           0 : }
    1502             : EXPORT_SYMBOL(ktime_get_raw_ts64);
    1503             : 
    1504             : 
    1505             : /**
    1506             :  * timekeeping_valid_for_hres - Check if timekeeping is suitable for hres
    1507             :  */
    1508           8 : int timekeeping_valid_for_hres(void)
    1509             : {
    1510           8 :         struct timekeeper *tk = &tk_core.timekeeper;
    1511           8 :         unsigned int seq;
    1512           8 :         int ret;
    1513             : 
    1514           8 :         do {
    1515           8 :                 seq = read_seqcount_begin(&tk_core.seq);
    1516             : 
    1517           8 :                 ret = tk->tkr_mono.clock->flags & CLOCK_SOURCE_VALID_FOR_HRES;
    1518             : 
    1519           8 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1520             : 
    1521           8 :         return ret;
    1522             : }
    1523             : 
    1524             : /**
    1525             :  * timekeeping_max_deferment - Returns max time the clocksource can be deferred
    1526             :  */
    1527        1174 : u64 timekeeping_max_deferment(void)
    1528             : {
    1529        1174 :         struct timekeeper *tk = &tk_core.timekeeper;
    1530        1174 :         unsigned int seq;
    1531        1174 :         u64 ret;
    1532             : 
    1533        1174 :         do {
    1534        1174 :                 seq = read_seqcount_begin(&tk_core.seq);
    1535             : 
    1536        1182 :                 ret = tk->tkr_mono.clock->max_idle_ns;
    1537             : 
    1538        1182 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    1539             : 
    1540        1182 :         return ret;
    1541             : }
    1542             : 
    1543             : /**
    1544             :  * read_persistent_clock64 -  Return time from the persistent clock.
    1545             :  * @ts: Pointer to the storage for the readout value
    1546             :  *
    1547             :  * Weak dummy function for arches that do not yet support it.
    1548             :  * Reads the time from the battery backed persistent clock.
    1549             :  * Returns a timespec with tv_sec=0 and tv_nsec=0 if unsupported.
    1550             :  *
    1551             :  *  XXX - Do be sure to remove it once all arches implement it.
    1552             :  */
    1553           0 : void __weak read_persistent_clock64(struct timespec64 *ts)
    1554             : {
    1555           0 :         ts->tv_sec = 0;
    1556           0 :         ts->tv_nsec = 0;
    1557           0 : }
    1558             : 
    1559             : /**
    1560             :  * read_persistent_wall_and_boot_offset - Read persistent clock, and also offset
    1561             :  *                                        from the boot.
    1562             :  *
    1563             :  * Weak dummy function for arches that do not yet support it.
    1564             :  * @wall_time:  - current time as returned by persistent clock
    1565             :  * @boot_offset: - offset that is defined as wall_time - boot_time
    1566             :  *
    1567             :  * The default function calculates offset based on the current value of
    1568             :  * local_clock(). This way architectures that support sched_clock() but don't
    1569             :  * support dedicated boot time clock will provide the best estimate of the
    1570             :  * boot time.
    1571             :  */
    1572             : void __weak __init
    1573           1 : read_persistent_wall_and_boot_offset(struct timespec64 *wall_time,
    1574             :                                      struct timespec64 *boot_offset)
    1575             : {
    1576           1 :         read_persistent_clock64(wall_time);
    1577           2 :         *boot_offset = ns_to_timespec64(local_clock());
    1578           1 : }
    1579             : 
    1580             : /*
    1581             :  * Flag reflecting whether timekeeping_resume() has injected sleeptime.
    1582             :  *
    1583             :  * The flag starts of false and is only set when a suspend reaches
    1584             :  * timekeeping_suspend(), timekeeping_resume() sets it to false when the
    1585             :  * timekeeper clocksource is not stopping across suspend and has been
    1586             :  * used to update sleep time. If the timekeeper clocksource has stopped
    1587             :  * then the flag stays true and is used by the RTC resume code to decide
    1588             :  * whether sleeptime must be injected and if so the flag gets false then.
    1589             :  *
    1590             :  * If a suspend fails before reaching timekeeping_resume() then the flag
    1591             :  * stays false and prevents erroneous sleeptime injection.
    1592             :  */
    1593             : static bool suspend_timing_needed;
    1594             : 
    1595             : /* Flag for if there is a persistent clock on this platform */
    1596             : static bool persistent_clock_exists;
    1597             : 
    1598             : /*
    1599             :  * timekeeping_init - Initializes the clocksource and common timekeeping values
    1600             :  */
    1601           1 : void __init timekeeping_init(void)
    1602             : {
    1603           1 :         struct timespec64 wall_time, boot_offset, wall_to_mono;
    1604           1 :         struct timekeeper *tk = &tk_core.timekeeper;
    1605           1 :         struct clocksource *clock;
    1606           1 :         unsigned long flags;
    1607             : 
    1608           1 :         read_persistent_wall_and_boot_offset(&wall_time, &boot_offset);
    1609           2 :         if (timespec64_valid_settod(&wall_time) &&
    1610           1 :             timespec64_to_ns(&wall_time) > 0) {
    1611           1 :                 persistent_clock_exists = true;
    1612           0 :         } else if (timespec64_to_ns(&wall_time) != 0) {
    1613           0 :                 pr_warn("Persistent clock returned invalid value");
    1614           0 :                 wall_time = (struct timespec64){0};
    1615             :         }
    1616             : 
    1617           1 :         if (timespec64_compare(&wall_time, &boot_offset) < 0)
    1618           0 :                 boot_offset = (struct timespec64){0};
    1619             : 
    1620             :         /*
    1621             :          * We want set wall_to_mono, so the following is true:
    1622             :          * wall time + wall_to_mono = boot time
    1623             :          */
    1624           1 :         wall_to_mono = timespec64_sub(boot_offset, wall_time);
    1625             : 
    1626           1 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1627           2 :         write_seqcount_begin(&tk_core.seq);
    1628           1 :         ntp_init();
    1629             : 
    1630           1 :         clock = clocksource_default_clock();
    1631           1 :         if (clock->enable)
    1632           0 :                 clock->enable(clock);
    1633           1 :         tk_setup_internals(tk, clock);
    1634             : 
    1635           1 :         tk_set_xtime(tk, &wall_time);
    1636           1 :         tk->raw_sec = 0;
    1637             : 
    1638           1 :         tk_set_wall_to_mono(tk, wall_to_mono);
    1639             : 
    1640           1 :         timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
    1641             : 
    1642           1 :         write_seqcount_end(&tk_core.seq);
    1643           1 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1644           1 : }
    1645             : 
    1646             : /* time in seconds when suspend began for persistent clock */
    1647             : static struct timespec64 timekeeping_suspend_time;
    1648             : 
    1649             : /**
    1650             :  * __timekeeping_inject_sleeptime - Internal function to add sleep interval
    1651             :  * @tk:         Pointer to the timekeeper to be updated
    1652             :  * @delta:      Pointer to the delta value in timespec64 format
    1653             :  *
    1654             :  * Takes a timespec offset measuring a suspend interval and properly
    1655             :  * adds the sleep offset to the timekeeping variables.
    1656             :  */
    1657           0 : static void __timekeeping_inject_sleeptime(struct timekeeper *tk,
    1658             :                                            const struct timespec64 *delta)
    1659             : {
    1660           0 :         if (!timespec64_valid_strict(delta)) {
    1661           0 :                 printk_deferred(KERN_WARNING
    1662             :                                 "__timekeeping_inject_sleeptime: Invalid "
    1663             :                                 "sleep delta value!\n");
    1664           0 :                 return;
    1665             :         }
    1666           0 :         tk_xtime_add(tk, delta);
    1667           0 :         tk_set_wall_to_mono(tk, timespec64_sub(tk->wall_to_monotonic, *delta));
    1668           0 :         tk_update_sleep_time(tk, timespec64_to_ktime(*delta));
    1669           0 :         tk_debug_account_sleep_time(delta);
    1670             : }
    1671             : 
    1672             : #if defined(CONFIG_PM_SLEEP) && defined(CONFIG_RTC_HCTOSYS_DEVICE)
    1673             : /**
    1674             :  * We have three kinds of time sources to use for sleep time
    1675             :  * injection, the preference order is:
    1676             :  * 1) non-stop clocksource
    1677             :  * 2) persistent clock (ie: RTC accessible when irqs are off)
    1678             :  * 3) RTC
    1679             :  *
    1680             :  * 1) and 2) are used by timekeeping, 3) by RTC subsystem.
    1681             :  * If system has neither 1) nor 2), 3) will be used finally.
    1682             :  *
    1683             :  *
    1684             :  * If timekeeping has injected sleeptime via either 1) or 2),
    1685             :  * 3) becomes needless, so in this case we don't need to call
    1686             :  * rtc_resume(), and this is what timekeeping_rtc_skipresume()
    1687             :  * means.
    1688             :  */
    1689             : bool timekeeping_rtc_skipresume(void)
    1690             : {
    1691             :         return !suspend_timing_needed;
    1692             : }
    1693             : 
    1694             : /**
    1695             :  * 1) can be determined whether to use or not only when doing
    1696             :  * timekeeping_resume() which is invoked after rtc_suspend(),
    1697             :  * so we can't skip rtc_suspend() surely if system has 1).
    1698             :  *
    1699             :  * But if system has 2), 2) will definitely be used, so in this
    1700             :  * case we don't need to call rtc_suspend(), and this is what
    1701             :  * timekeeping_rtc_skipsuspend() means.
    1702             :  */
    1703             : bool timekeeping_rtc_skipsuspend(void)
    1704             : {
    1705             :         return persistent_clock_exists;
    1706             : }
    1707             : 
    1708             : /**
    1709             :  * timekeeping_inject_sleeptime64 - Adds suspend interval to timeekeeping values
    1710             :  * @delta: pointer to a timespec64 delta value
    1711             :  *
    1712             :  * This hook is for architectures that cannot support read_persistent_clock64
    1713             :  * because their RTC/persistent clock is only accessible when irqs are enabled.
    1714             :  * and also don't have an effective nonstop clocksource.
    1715             :  *
    1716             :  * This function should only be called by rtc_resume(), and allows
    1717             :  * a suspend offset to be injected into the timekeeping values.
    1718             :  */
    1719             : void timekeeping_inject_sleeptime64(const struct timespec64 *delta)
    1720             : {
    1721             :         struct timekeeper *tk = &tk_core.timekeeper;
    1722             :         unsigned long flags;
    1723             : 
    1724             :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1725             :         write_seqcount_begin(&tk_core.seq);
    1726             : 
    1727             :         suspend_timing_needed = false;
    1728             : 
    1729             :         timekeeping_forward_now(tk);
    1730             : 
    1731             :         __timekeeping_inject_sleeptime(tk, delta);
    1732             : 
    1733             :         timekeeping_update(tk, TK_CLEAR_NTP | TK_MIRROR | TK_CLOCK_WAS_SET);
    1734             : 
    1735             :         write_seqcount_end(&tk_core.seq);
    1736             :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1737             : 
    1738             :         /* signal hrtimers about time change */
    1739             :         clock_was_set();
    1740             : }
    1741             : #endif
    1742             : 
    1743             : /**
    1744             :  * timekeeping_resume - Resumes the generic timekeeping subsystem.
    1745             :  */
    1746           0 : void timekeeping_resume(void)
    1747             : {
    1748           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1749           0 :         struct clocksource *clock = tk->tkr_mono.clock;
    1750           0 :         unsigned long flags;
    1751           0 :         struct timespec64 ts_new, ts_delta;
    1752           0 :         u64 cycle_now, nsec;
    1753           0 :         bool inject_sleeptime = false;
    1754             : 
    1755           0 :         read_persistent_clock64(&ts_new);
    1756             : 
    1757           0 :         clockevents_resume();
    1758           0 :         clocksource_resume();
    1759             : 
    1760           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1761           0 :         write_seqcount_begin(&tk_core.seq);
    1762             : 
    1763             :         /*
    1764             :          * After system resumes, we need to calculate the suspended time and
    1765             :          * compensate it for the OS time. There are 3 sources that could be
    1766             :          * used: Nonstop clocksource during suspend, persistent clock and rtc
    1767             :          * device.
    1768             :          *
    1769             :          * One specific platform may have 1 or 2 or all of them, and the
    1770             :          * preference will be:
    1771             :          *      suspend-nonstop clocksource -> persistent clock -> rtc
    1772             :          * The less preferred source will only be tried if there is no better
    1773             :          * usable source. The rtc part is handled separately in rtc core code.
    1774             :          */
    1775           0 :         cycle_now = tk_clock_read(&tk->tkr_mono);
    1776           0 :         nsec = clocksource_stop_suspend_timing(clock, cycle_now);
    1777           0 :         if (nsec > 0) {
    1778           0 :                 ts_delta = ns_to_timespec64(nsec);
    1779           0 :                 inject_sleeptime = true;
    1780           0 :         } else if (timespec64_compare(&ts_new, &timekeeping_suspend_time) > 0) {
    1781           0 :                 ts_delta = timespec64_sub(ts_new, timekeeping_suspend_time);
    1782           0 :                 inject_sleeptime = true;
    1783             :         }
    1784             : 
    1785           0 :         if (inject_sleeptime) {
    1786           0 :                 suspend_timing_needed = false;
    1787           0 :                 __timekeeping_inject_sleeptime(tk, &ts_delta);
    1788             :         }
    1789             : 
    1790             :         /* Re-base the last cycle value */
    1791           0 :         tk->tkr_mono.cycle_last = cycle_now;
    1792           0 :         tk->tkr_raw.cycle_last  = cycle_now;
    1793             : 
    1794           0 :         tk->ntp_error = 0;
    1795           0 :         timekeeping_suspended = 0;
    1796           0 :         timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
    1797           0 :         write_seqcount_end(&tk_core.seq);
    1798           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1799             : 
    1800           0 :         touch_softlockup_watchdog();
    1801             : 
    1802           0 :         tick_resume();
    1803           0 :         hrtimers_resume();
    1804           0 : }
    1805             : 
    1806           0 : int timekeeping_suspend(void)
    1807             : {
    1808           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    1809           0 :         unsigned long flags;
    1810           0 :         struct timespec64               delta, delta_delta;
    1811           0 :         static struct timespec64        old_delta;
    1812           0 :         struct clocksource *curr_clock;
    1813           0 :         u64 cycle_now;
    1814             : 
    1815           0 :         read_persistent_clock64(&timekeeping_suspend_time);
    1816             : 
    1817             :         /*
    1818             :          * On some systems the persistent_clock can not be detected at
    1819             :          * timekeeping_init by its return value, so if we see a valid
    1820             :          * value returned, update the persistent_clock_exists flag.
    1821             :          */
    1822           0 :         if (timekeeping_suspend_time.tv_sec || timekeeping_suspend_time.tv_nsec)
    1823           0 :                 persistent_clock_exists = true;
    1824             : 
    1825           0 :         suspend_timing_needed = true;
    1826             : 
    1827           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    1828           0 :         write_seqcount_begin(&tk_core.seq);
    1829           0 :         timekeeping_forward_now(tk);
    1830           0 :         timekeeping_suspended = 1;
    1831             : 
    1832             :         /*
    1833             :          * Since we've called forward_now, cycle_last stores the value
    1834             :          * just read from the current clocksource. Save this to potentially
    1835             :          * use in suspend timing.
    1836             :          */
    1837           0 :         curr_clock = tk->tkr_mono.clock;
    1838           0 :         cycle_now = tk->tkr_mono.cycle_last;
    1839           0 :         clocksource_start_suspend_timing(curr_clock, cycle_now);
    1840             : 
    1841           0 :         if (persistent_clock_exists) {
    1842             :                 /*
    1843             :                  * To avoid drift caused by repeated suspend/resumes,
    1844             :                  * which each can add ~1 second drift error,
    1845             :                  * try to compensate so the difference in system time
    1846             :                  * and persistent_clock time stays close to constant.
    1847             :                  */
    1848           0 :                 delta = timespec64_sub(tk_xtime(tk), timekeeping_suspend_time);
    1849           0 :                 delta_delta = timespec64_sub(delta, old_delta);
    1850           0 :                 if (abs(delta_delta.tv_sec) >= 2) {
    1851             :                         /*
    1852             :                          * if delta_delta is too large, assume time correction
    1853             :                          * has occurred and set old_delta to the current delta.
    1854             :                          */
    1855           0 :                         old_delta = delta;
    1856             :                 } else {
    1857             :                         /* Otherwise try to adjust old_system to compensate */
    1858           0 :                         timekeeping_suspend_time =
    1859           0 :                                 timespec64_add(timekeeping_suspend_time, delta_delta);
    1860             :                 }
    1861             :         }
    1862             : 
    1863           0 :         timekeeping_update(tk, TK_MIRROR);
    1864           0 :         halt_fast_timekeeper(tk);
    1865           0 :         write_seqcount_end(&tk_core.seq);
    1866           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    1867             : 
    1868           0 :         tick_suspend();
    1869           0 :         clocksource_suspend();
    1870           0 :         clockevents_suspend();
    1871             : 
    1872           0 :         return 0;
    1873             : }
    1874             : 
    1875             : /* sysfs resume/suspend bits for timekeeping */
    1876             : static struct syscore_ops timekeeping_syscore_ops = {
    1877             :         .resume         = timekeeping_resume,
    1878             :         .suspend        = timekeeping_suspend,
    1879             : };
    1880             : 
    1881           1 : static int __init timekeeping_init_ops(void)
    1882             : {
    1883           1 :         register_syscore_ops(&timekeeping_syscore_ops);
    1884           1 :         return 0;
    1885             : }
    1886             : device_initcall(timekeeping_init_ops);
    1887             : 
    1888             : /*
    1889             :  * Apply a multiplier adjustment to the timekeeper
    1890             :  */
    1891        7943 : static __always_inline void timekeeping_apply_adjustment(struct timekeeper *tk,
    1892             :                                                          s64 offset,
    1893             :                                                          s32 mult_adj)
    1894             : {
    1895        7943 :         s64 interval = tk->cycle_interval;
    1896             : 
    1897        7943 :         if (mult_adj == 0) {
    1898             :                 return;
    1899           0 :         } else if (mult_adj == -1) {
    1900           0 :                 interval = -interval;
    1901           0 :                 offset = -offset;
    1902           0 :         } else if (mult_adj != 1) {
    1903           0 :                 interval *= mult_adj;
    1904           0 :                 offset *= mult_adj;
    1905             :         }
    1906             : 
    1907             :         /*
    1908             :          * So the following can be confusing.
    1909             :          *
    1910             :          * To keep things simple, lets assume mult_adj == 1 for now.
    1911             :          *
    1912             :          * When mult_adj != 1, remember that the interval and offset values
    1913             :          * have been appropriately scaled so the math is the same.
    1914             :          *
    1915             :          * The basic idea here is that we're increasing the multiplier
    1916             :          * by one, this causes the xtime_interval to be incremented by
    1917             :          * one cycle_interval. This is because:
    1918             :          *      xtime_interval = cycle_interval * mult
    1919             :          * So if mult is being incremented by one:
    1920             :          *      xtime_interval = cycle_interval * (mult + 1)
    1921             :          * Its the same as:
    1922             :          *      xtime_interval = (cycle_interval * mult) + cycle_interval
    1923             :          * Which can be shortened to:
    1924             :          *      xtime_interval += cycle_interval
    1925             :          *
    1926             :          * So offset stores the non-accumulated cycles. Thus the current
    1927             :          * time (in shifted nanoseconds) is:
    1928             :          *      now = (offset * adj) + xtime_nsec
    1929             :          * Now, even though we're adjusting the clock frequency, we have
    1930             :          * to keep time consistent. In other words, we can't jump back
    1931             :          * in time, and we also want to avoid jumping forward in time.
    1932             :          *
    1933             :          * So given the same offset value, we need the time to be the same
    1934             :          * both before and after the freq adjustment.
    1935             :          *      now = (offset * adj_1) + xtime_nsec_1
    1936             :          *      now = (offset * adj_2) + xtime_nsec_2
    1937             :          * So:
    1938             :          *      (offset * adj_1) + xtime_nsec_1 =
    1939             :          *              (offset * adj_2) + xtime_nsec_2
    1940             :          * And we know:
    1941             :          *      adj_2 = adj_1 + 1
    1942             :          * So:
    1943             :          *      (offset * adj_1) + xtime_nsec_1 =
    1944             :          *              (offset * (adj_1+1)) + xtime_nsec_2
    1945             :          *      (offset * adj_1) + xtime_nsec_1 =
    1946             :          *              (offset * adj_1) + offset + xtime_nsec_2
    1947             :          * Canceling the sides:
    1948             :          *      xtime_nsec_1 = offset + xtime_nsec_2
    1949             :          * Which gives us:
    1950             :          *      xtime_nsec_2 = xtime_nsec_1 - offset
    1951             :          * Which simplfies to:
    1952             :          *      xtime_nsec -= offset
    1953             :          */
    1954           0 :         if ((mult_adj > 0) && (tk->tkr_mono.mult + mult_adj < mult_adj)) {
    1955             :                 /* NTP adjustment caused clocksource mult overflow */
    1956           0 :                 WARN_ON_ONCE(1);
    1957           0 :                 return;
    1958             :         }
    1959             : 
    1960           0 :         tk->tkr_mono.mult += mult_adj;
    1961           0 :         tk->xtime_interval += interval;
    1962           0 :         tk->tkr_mono.xtime_nsec -= offset;
    1963             : }
    1964             : 
    1965             : /*
    1966             :  * Adjust the timekeeper's multiplier to the correct frequency
    1967             :  * and also to reduce the accumulated error value.
    1968             :  */
    1969        7943 : static void timekeeping_adjust(struct timekeeper *tk, s64 offset)
    1970             : {
    1971        7943 :         u32 mult;
    1972             : 
    1973             :         /*
    1974             :          * Determine the multiplier from the current NTP tick length.
    1975             :          * Avoid expensive division when the tick length doesn't change.
    1976             :          */
    1977        7943 :         if (likely(tk->ntp_tick == ntp_tick_length())) {
    1978        7943 :                 mult = tk->tkr_mono.mult - tk->ntp_err_mult;
    1979             :         } else {
    1980           0 :                 tk->ntp_tick = ntp_tick_length();
    1981           0 :                 mult = div64_u64((tk->ntp_tick >> tk->ntp_error_shift) -
    1982           0 :                                  tk->xtime_remainder, tk->cycle_interval);
    1983             :         }
    1984             : 
    1985             :         /*
    1986             :          * If the clock is behind the NTP time, increase the multiplier by 1
    1987             :          * to catch up with it. If it's ahead and there was a remainder in the
    1988             :          * tick division, the clock will slow down. Otherwise it will stay
    1989             :          * ahead until the tick length changes to a non-divisible value.
    1990             :          */
    1991        7943 :         tk->ntp_err_mult = tk->ntp_error > 0 ? 1 : 0;
    1992        7943 :         mult += tk->ntp_err_mult;
    1993             : 
    1994        7943 :         timekeeping_apply_adjustment(tk, offset, mult - tk->tkr_mono.mult);
    1995             : 
    1996        7943 :         if (unlikely(tk->tkr_mono.clock->maxadj &&
    1997             :                 (abs(tk->tkr_mono.mult - tk->tkr_mono.clock->mult)
    1998             :                         > tk->tkr_mono.clock->maxadj))) {
    1999           0 :                 printk_once(KERN_WARNING
    2000             :                         "Adjusting %s more than 11%% (%ld vs %ld)\n",
    2001             :                         tk->tkr_mono.clock->name, (long)tk->tkr_mono.mult,
    2002             :                         (long)tk->tkr_mono.clock->mult + tk->tkr_mono.clock->maxadj);
    2003             :         }
    2004             : 
    2005             :         /*
    2006             :          * It may be possible that when we entered this function, xtime_nsec
    2007             :          * was very small.  Further, if we're slightly speeding the clocksource
    2008             :          * in the code above, its possible the required corrective factor to
    2009             :          * xtime_nsec could cause it to underflow.
    2010             :          *
    2011             :          * Now, since we have already accumulated the second and the NTP
    2012             :          * subsystem has been notified via second_overflow(), we need to skip
    2013             :          * the next update.
    2014             :          */
    2015        7943 :         if (unlikely((s64)tk->tkr_mono.xtime_nsec < 0)) {
    2016           0 :                 tk->tkr_mono.xtime_nsec += (u64)NSEC_PER_SEC <<
    2017           0 :                                                         tk->tkr_mono.shift;
    2018           0 :                 tk->xtime_sec--;
    2019           0 :                 tk->skip_second_overflow = 1;
    2020             :         }
    2021        7943 : }
    2022             : 
    2023             : /*
    2024             :  * accumulate_nsecs_to_secs - Accumulates nsecs into secs
    2025             :  *
    2026             :  * Helper function that accumulates the nsecs greater than a second
    2027             :  * from the xtime_nsec field to the xtime_secs field.
    2028             :  * It also calls into the NTP code to handle leapsecond processing.
    2029             :  */
    2030       15915 : static inline unsigned int accumulate_nsecs_to_secs(struct timekeeper *tk)
    2031             : {
    2032       15915 :         u64 nsecps = (u64)NSEC_PER_SEC << tk->tkr_mono.shift;
    2033       15915 :         unsigned int clock_set = 0;
    2034             : 
    2035       15950 :         while (tk->tkr_mono.xtime_nsec >= nsecps) {
    2036          35 :                 int leap;
    2037             : 
    2038          35 :                 tk->tkr_mono.xtime_nsec -= nsecps;
    2039          35 :                 tk->xtime_sec++;
    2040             : 
    2041             :                 /*
    2042             :                  * Skip NTP update if this second was accumulated before,
    2043             :                  * i.e. xtime_nsec underflowed in timekeeping_adjust()
    2044             :                  */
    2045          35 :                 if (unlikely(tk->skip_second_overflow)) {
    2046           0 :                         tk->skip_second_overflow = 0;
    2047           0 :                         continue;
    2048             :                 }
    2049             : 
    2050             :                 /* Figure out if its a leap sec and apply if needed */
    2051          35 :                 leap = second_overflow(tk->xtime_sec);
    2052          35 :                 if (unlikely(leap)) {
    2053           0 :                         struct timespec64 ts;
    2054             : 
    2055           0 :                         tk->xtime_sec += leap;
    2056             : 
    2057           0 :                         ts.tv_sec = leap;
    2058           0 :                         ts.tv_nsec = 0;
    2059           0 :                         tk_set_wall_to_mono(tk,
    2060             :                                 timespec64_sub(tk->wall_to_monotonic, ts));
    2061             : 
    2062           0 :                         __timekeeping_set_tai_offset(tk, tk->tai_offset - leap);
    2063             : 
    2064           0 :                         clock_set = TK_CLOCK_WAS_SET;
    2065             :                 }
    2066             :         }
    2067       15915 :         return clock_set;
    2068             : }
    2069             : 
    2070             : /*
    2071             :  * logarithmic_accumulation - shifted accumulation of cycles
    2072             :  *
    2073             :  * This functions accumulates a shifted interval of cycles into
    2074             :  * a shifted interval nanoseconds. Allows for O(log) accumulation
    2075             :  * loop.
    2076             :  *
    2077             :  * Returns the unconsumed cycles.
    2078             :  */
    2079        8252 : static u64 logarithmic_accumulation(struct timekeeper *tk, u64 offset,
    2080             :                                     u32 shift, unsigned int *clock_set)
    2081             : {
    2082        8252 :         u64 interval = tk->cycle_interval << shift;
    2083        8252 :         u64 snsec_per_sec;
    2084             : 
    2085             :         /* If the offset is smaller than a shifted interval, do nothing */
    2086        8252 :         if (offset < interval)
    2087             :                 return offset;
    2088             : 
    2089             :         /* Accumulate one shifted interval */
    2090        7972 :         offset -= interval;
    2091        7972 :         tk->tkr_mono.cycle_last += interval;
    2092        7972 :         tk->tkr_raw.cycle_last  += interval;
    2093             : 
    2094        7972 :         tk->tkr_mono.xtime_nsec += tk->xtime_interval << shift;
    2095        7972 :         *clock_set |= accumulate_nsecs_to_secs(tk);
    2096             : 
    2097             :         /* Accumulate raw time */
    2098        7972 :         tk->tkr_raw.xtime_nsec += tk->raw_interval << shift;
    2099        7972 :         snsec_per_sec = (u64)NSEC_PER_SEC << tk->tkr_raw.shift;
    2100        8006 :         while (tk->tkr_raw.xtime_nsec >= snsec_per_sec) {
    2101          34 :                 tk->tkr_raw.xtime_nsec -= snsec_per_sec;
    2102          34 :                 tk->raw_sec++;
    2103             :         }
    2104             : 
    2105             :         /* Accumulate error between NTP and clock interval */
    2106        7972 :         tk->ntp_error += tk->ntp_tick << shift;
    2107        7972 :         tk->ntp_error -= (tk->xtime_interval + tk->xtime_remainder) <<
    2108        7972 :                                                 (tk->ntp_error_shift + shift);
    2109             : 
    2110        7972 :         return offset;
    2111             : }
    2112             : 
    2113             : /*
    2114             :  * timekeeping_advance - Updates the timekeeper to the current time and
    2115             :  * current NTP tick length
    2116             :  */
    2117        7945 : static void timekeeping_advance(enum timekeeping_adv_mode mode)
    2118             : {
    2119        7945 :         struct timekeeper *real_tk = &tk_core.timekeeper;
    2120        7945 :         struct timekeeper *tk = &shadow_timekeeper;
    2121        7945 :         u64 offset;
    2122        7945 :         int shift = 0, maxshift;
    2123        7945 :         unsigned int clock_set = 0;
    2124        7945 :         unsigned long flags;
    2125             : 
    2126        7945 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    2127             : 
    2128             :         /* Make sure we're fully resumed: */
    2129        7945 :         if (unlikely(timekeeping_suspended))
    2130           0 :                 goto out;
    2131             : 
    2132        7945 :         offset = clocksource_delta(tk_clock_read(&tk->tkr_mono),
    2133             :                                    tk->tkr_mono.cycle_last, tk->tkr_mono.mask);
    2134             : 
    2135             :         /* Check if there's really nothing to do */
    2136        7945 :         if (offset < real_tk->cycle_interval && mode == TK_ADV_TICK)
    2137           2 :                 goto out;
    2138             : 
    2139             :         /* Do some additional sanity checking */
    2140        7943 :         timekeeping_check_update(tk, offset);
    2141             : 
    2142             :         /*
    2143             :          * With NO_HZ we may have to accumulate many cycle_intervals
    2144             :          * (think "ticks") worth of time at once. To do this efficiently,
    2145             :          * we calculate the largest doubling multiple of cycle_intervals
    2146             :          * that is smaller than the offset.  We then accumulate that
    2147             :          * chunk in one go, and then try to consume the next smaller
    2148             :          * doubled multiple.
    2149             :          */
    2150        7943 :         shift = ilog2(offset) - ilog2(tk->cycle_interval);
    2151        7943 :         shift = max(0, shift);
    2152             :         /* Bound shift to one less than what overflows tick_length */
    2153        7943 :         maxshift = (64 - (ilog2(ntp_tick_length())+1)) - 1;
    2154        7943 :         shift = min(shift, maxshift);
    2155       16195 :         while (offset >= tk->cycle_interval) {
    2156        8252 :                 offset = logarithmic_accumulation(tk, offset, shift,
    2157             :                                                         &clock_set);
    2158        8252 :                 if (offset < tk->cycle_interval<<shift)
    2159        8252 :                         shift--;
    2160             :         }
    2161             : 
    2162             :         /* Adjust the multiplier to correct NTP error */
    2163        7943 :         timekeeping_adjust(tk, offset);
    2164             : 
    2165             :         /*
    2166             :          * Finally, make sure that after the rounding
    2167             :          * xtime_nsec isn't larger than NSEC_PER_SEC
    2168             :          */
    2169        7943 :         clock_set |= accumulate_nsecs_to_secs(tk);
    2170             : 
    2171       15886 :         write_seqcount_begin(&tk_core.seq);
    2172             :         /*
    2173             :          * Update the real timekeeper.
    2174             :          *
    2175             :          * We could avoid this memcpy by switching pointers, but that
    2176             :          * requires changes to all other timekeeper usage sites as
    2177             :          * well, i.e. move the timekeeper pointer getter into the
    2178             :          * spinlocked/seqcount protected sections. And we trade this
    2179             :          * memcpy under the tk_core.seq against one before we start
    2180             :          * updating.
    2181             :          */
    2182        7943 :         timekeeping_update(tk, clock_set);
    2183        7943 :         memcpy(real_tk, tk, sizeof(*tk));
    2184             :         /* The memcpy must come last. Do not put anything here! */
    2185        7943 :         write_seqcount_end(&tk_core.seq);
    2186        7943 : out:
    2187        7945 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    2188        7945 :         if (clock_set)
    2189             :                 /* Have to call _delayed version, since in irq context*/
    2190             :                 clock_was_set_delayed();
    2191        7945 : }
    2192             : 
    2193             : /**
    2194             :  * update_wall_time - Uses the current clocksource to increment the wall time
    2195             :  *
    2196             :  */
    2197        7945 : void update_wall_time(void)
    2198             : {
    2199        7945 :         timekeeping_advance(TK_ADV_TICK);
    2200        7945 : }
    2201             : 
    2202             : /**
    2203             :  * getboottime64 - Return the real time of system boot.
    2204             :  * @ts:         pointer to the timespec64 to be set
    2205             :  *
    2206             :  * Returns the wall-time of boot in a timespec64.
    2207             :  *
    2208             :  * This is based on the wall_to_monotonic offset and the total suspend
    2209             :  * time. Calls to settimeofday will affect the value returned (which
    2210             :  * basically means that however wrong your real time clock is at boot time,
    2211             :  * you get the right time here).
    2212             :  */
    2213           0 : void getboottime64(struct timespec64 *ts)
    2214             : {
    2215           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    2216           0 :         ktime_t t = ktime_sub(tk->offs_real, tk->offs_boot);
    2217             : 
    2218           0 :         *ts = ktime_to_timespec64(t);
    2219           0 : }
    2220             : EXPORT_SYMBOL_GPL(getboottime64);
    2221             : 
    2222       79951 : void ktime_get_coarse_real_ts64(struct timespec64 *ts)
    2223             : {
    2224       79951 :         struct timekeeper *tk = &tk_core.timekeeper;
    2225       79952 :         unsigned int seq;
    2226             : 
    2227       79952 :         do {
    2228       80745 :                 seq = read_seqcount_begin(&tk_core.seq);
    2229             : 
    2230       79951 :                 *ts = tk_xtime(tk);
    2231       79951 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    2232       79950 : }
    2233             : EXPORT_SYMBOL(ktime_get_coarse_real_ts64);
    2234             : 
    2235           0 : void ktime_get_coarse_ts64(struct timespec64 *ts)
    2236             : {
    2237           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    2238           0 :         struct timespec64 now, mono;
    2239           0 :         unsigned int seq;
    2240             : 
    2241           0 :         do {
    2242           0 :                 seq = read_seqcount_begin(&tk_core.seq);
    2243             : 
    2244           0 :                 now = tk_xtime(tk);
    2245           0 :                 mono = tk->wall_to_monotonic;
    2246           0 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    2247             : 
    2248           0 :         set_normalized_timespec64(ts, now.tv_sec + mono.tv_sec,
    2249           0 :                                 now.tv_nsec + mono.tv_nsec);
    2250           0 : }
    2251             : EXPORT_SYMBOL(ktime_get_coarse_ts64);
    2252             : 
    2253             : /*
    2254             :  * Must hold jiffies_lock
    2255             :  */
    2256          24 : void do_timer(unsigned long ticks)
    2257             : {
    2258          24 :         jiffies_64 += ticks;
    2259          24 :         calc_global_load();
    2260          24 : }
    2261             : 
    2262             : /**
    2263             :  * ktime_get_update_offsets_now - hrtimer helper
    2264             :  * @cwsseq:     pointer to check and store the clock was set sequence number
    2265             :  * @offs_real:  pointer to storage for monotonic -> realtime offset
    2266             :  * @offs_boot:  pointer to storage for monotonic -> boottime offset
    2267             :  * @offs_tai:   pointer to storage for monotonic -> clock tai offset
    2268             :  *
    2269             :  * Returns current monotonic time and updates the offsets if the
    2270             :  * sequence number in @cwsseq and timekeeper.clock_was_set_seq are
    2271             :  * different.
    2272             :  *
    2273             :  * Called from hrtimer_interrupt() or retrigger_next_event()
    2274             :  */
    2275       28345 : ktime_t ktime_get_update_offsets_now(unsigned int *cwsseq, ktime_t *offs_real,
    2276             :                                      ktime_t *offs_boot, ktime_t *offs_tai)
    2277             : {
    2278       28345 :         struct timekeeper *tk = &tk_core.timekeeper;
    2279       28566 :         unsigned int seq;
    2280       28566 :         ktime_t base;
    2281       28566 :         u64 nsecs;
    2282             : 
    2283       28566 :         do {
    2284      299150 :                 seq = read_seqcount_begin(&tk_core.seq);
    2285             : 
    2286       27943 :                 base = tk->tkr_mono.base;
    2287       27943 :                 nsecs = timekeeping_get_ns(&tk->tkr_mono);
    2288       28676 :                 base = ktime_add_ns(base, nsecs);
    2289             : 
    2290       28676 :                 if (*cwsseq != tk->clock_was_set_seq) {
    2291           8 :                         *cwsseq = tk->clock_was_set_seq;
    2292           8 :                         *offs_real = tk->offs_real;
    2293           8 :                         *offs_boot = tk->offs_boot;
    2294           8 :                         *offs_tai = tk->offs_tai;
    2295             :                 }
    2296             : 
    2297             :                 /* Handle leapsecond insertion adjustments */
    2298       28676 :                 if (unlikely(base >= tk->next_leap_ktime))
    2299           0 :                         *offs_real = ktime_sub(tk->offs_real, ktime_set(1, 0));
    2300             : 
    2301       28676 :         } while (read_seqcount_retry(&tk_core.seq, seq));
    2302             : 
    2303       28524 :         return base;
    2304             : }
    2305             : 
    2306             : /*
    2307             :  * timekeeping_validate_timex - Ensures the timex is ok for use in do_adjtimex
    2308             :  */
    2309           0 : static int timekeeping_validate_timex(const struct __kernel_timex *txc)
    2310             : {
    2311           0 :         if (txc->modes & ADJ_ADJTIME) {
    2312             :                 /* singleshot must not be used with any other mode bits */
    2313           0 :                 if (!(txc->modes & ADJ_OFFSET_SINGLESHOT))
    2314             :                         return -EINVAL;
    2315           0 :                 if (!(txc->modes & ADJ_OFFSET_READONLY) &&
    2316           0 :                     !capable(CAP_SYS_TIME))
    2317             :                         return -EPERM;
    2318             :         } else {
    2319             :                 /* In order to modify anything, you gotta be super-user! */
    2320           0 :                 if (txc->modes && !capable(CAP_SYS_TIME))
    2321             :                         return -EPERM;
    2322             :                 /*
    2323             :                  * if the quartz is off by more than 10% then
    2324             :                  * something is VERY wrong!
    2325             :                  */
    2326           0 :                 if (txc->modes & ADJ_TICK &&
    2327           0 :                     (txc->tick <  900000/USER_HZ ||
    2328             :                      txc->tick > 1100000/USER_HZ))
    2329             :                         return -EINVAL;
    2330             :         }
    2331             : 
    2332           0 :         if (txc->modes & ADJ_SETOFFSET) {
    2333             :                 /* In order to inject time, you gotta be super-user! */
    2334           0 :                 if (!capable(CAP_SYS_TIME))
    2335             :                         return -EPERM;
    2336             : 
    2337             :                 /*
    2338             :                  * Validate if a timespec/timeval used to inject a time
    2339             :                  * offset is valid.  Offsets can be postive or negative, so
    2340             :                  * we don't check tv_sec. The value of the timeval/timespec
    2341             :                  * is the sum of its fields,but *NOTE*:
    2342             :                  * The field tv_usec/tv_nsec must always be non-negative and
    2343             :                  * we can't have more nanoseconds/microseconds than a second.
    2344             :                  */
    2345           0 :                 if (txc->time.tv_usec < 0)
    2346             :                         return -EINVAL;
    2347             : 
    2348           0 :                 if (txc->modes & ADJ_NANO) {
    2349           0 :                         if (txc->time.tv_usec >= NSEC_PER_SEC)
    2350             :                                 return -EINVAL;
    2351             :                 } else {
    2352           0 :                         if (txc->time.tv_usec >= USEC_PER_SEC)
    2353             :                                 return -EINVAL;
    2354             :                 }
    2355             :         }
    2356             : 
    2357             :         /*
    2358             :          * Check for potential multiplication overflows that can
    2359             :          * only happen on 64-bit systems:
    2360             :          */
    2361           0 :         if ((txc->modes & ADJ_FREQUENCY) && (BITS_PER_LONG == 64)) {
    2362           0 :                 if (LLONG_MIN / PPM_SCALE > txc->freq)
    2363             :                         return -EINVAL;
    2364           0 :                 if (LLONG_MAX / PPM_SCALE < txc->freq)
    2365           0 :                         return -EINVAL;
    2366             :         }
    2367             : 
    2368             :         return 0;
    2369             : }
    2370             : 
    2371             : 
    2372             : /**
    2373             :  * do_adjtimex() - Accessor function to NTP __do_adjtimex function
    2374             :  */
    2375           0 : int do_adjtimex(struct __kernel_timex *txc)
    2376             : {
    2377           0 :         struct timekeeper *tk = &tk_core.timekeeper;
    2378           0 :         struct audit_ntp_data ad;
    2379           0 :         unsigned long flags;
    2380           0 :         struct timespec64 ts;
    2381           0 :         s32 orig_tai, tai;
    2382           0 :         int ret;
    2383             : 
    2384             :         /* Validate the data before disabling interrupts */
    2385           0 :         ret = timekeeping_validate_timex(txc);
    2386           0 :         if (ret)
    2387             :                 return ret;
    2388             : 
    2389           0 :         if (txc->modes & ADJ_SETOFFSET) {
    2390           0 :                 struct timespec64 delta;
    2391           0 :                 delta.tv_sec  = txc->time.tv_sec;
    2392           0 :                 delta.tv_nsec = txc->time.tv_usec;
    2393           0 :                 if (!(txc->modes & ADJ_NANO))
    2394           0 :                         delta.tv_nsec *= 1000;
    2395           0 :                 ret = timekeeping_inject_offset(&delta);
    2396           0 :                 if (ret)
    2397           0 :                         return ret;
    2398             : 
    2399           0 :                 audit_tk_injoffset(delta);
    2400             :         }
    2401             : 
    2402           0 :         audit_ntp_init(&ad);
    2403             : 
    2404           0 :         ktime_get_real_ts64(&ts);
    2405             : 
    2406           0 :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    2407           0 :         write_seqcount_begin(&tk_core.seq);
    2408             : 
    2409           0 :         orig_tai = tai = tk->tai_offset;
    2410           0 :         ret = __do_adjtimex(txc, &ts, &tai, &ad);
    2411             : 
    2412           0 :         if (tai != orig_tai) {
    2413           0 :                 __timekeeping_set_tai_offset(tk, tai);
    2414           0 :                 timekeeping_update(tk, TK_MIRROR | TK_CLOCK_WAS_SET);
    2415             :         }
    2416           0 :         tk_update_leap_state(tk);
    2417             : 
    2418           0 :         write_seqcount_end(&tk_core.seq);
    2419           0 :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    2420             : 
    2421           0 :         audit_ntp_log(&ad);
    2422             : 
    2423             :         /* Update the multiplier immediately if frequency was set directly */
    2424           0 :         if (txc->modes & (ADJ_FREQUENCY | ADJ_TICK))
    2425           0 :                 timekeeping_advance(TK_ADV_FREQ);
    2426             : 
    2427           0 :         if (tai != orig_tai)
    2428           0 :                 clock_was_set();
    2429             : 
    2430           0 :         ntp_notify_cmos_timer();
    2431             : 
    2432           0 :         return ret;
    2433             : }
    2434             : 
    2435             : #ifdef CONFIG_NTP_PPS
    2436             : /**
    2437             :  * hardpps() - Accessor function to NTP __hardpps function
    2438             :  */
    2439             : void hardpps(const struct timespec64 *phase_ts, const struct timespec64 *raw_ts)
    2440             : {
    2441             :         unsigned long flags;
    2442             : 
    2443             :         raw_spin_lock_irqsave(&timekeeper_lock, flags);
    2444             :         write_seqcount_begin(&tk_core.seq);
    2445             : 
    2446             :         __hardpps(phase_ts, raw_ts);
    2447             : 
    2448             :         write_seqcount_end(&tk_core.seq);
    2449             :         raw_spin_unlock_irqrestore(&timekeeper_lock, flags);
    2450             : }
    2451             : EXPORT_SYMBOL(hardpps);
    2452             : #endif /* CONFIG_NTP_PPS */

Generated by: LCOV version 1.14