xnu-4570.71.2.tar.gz
[apple/xnu.git] / osfmk / kern / clock.c
index 1bd578496eebf317883bf4ded4492c61e978fc90..9bd9f3b0e8249889767ff787b4c489306b4123dd 100644 (file)
  */
 /*
  */
  */
 /*
  */
+/*-
+ * Copyright (c) 1982, 1986, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)time.h      8.5 (Berkeley) 5/4/95
+ * $FreeBSD$
+ */
 
 #include <mach/mach_types.h>
 
 
 #include <mach/mach_types.h>
 
 #include <kern/thread.h>
 #include <kern/clock.h>
 #include <kern/host_notify.h>
 #include <kern/thread.h>
 #include <kern/clock.h>
 #include <kern/host_notify.h>
+#include <kern/thread_call.h>
+#include <libkern/OSAtomic.h>
 
 #include <IOKit/IOPlatformExpert.h>
 
 #include <machine/commpage.h>
 
 #include <IOKit/IOPlatformExpert.h>
 
 #include <machine/commpage.h>
+#include <machine/config.h>
+#include <machine/machine_routines.h>
 
 #include <mach/mach_traps.h>
 #include <mach/mach_time.h>
 
 
 #include <mach/mach_traps.h>
 #include <mach/mach_time.h>
 
-uint32_t       hz_tick_interval = 1;
+#include <sys/kdebug.h>
+#include <sys/timex.h>
+#include <kern/arithmetic_128.h>
+#include <os/log.h>
 
 
+uint32_t       hz_tick_interval = 1;
+static uint64_t has_monotonic_clock = 0;
 
 decl_simple_lock_data(,clock_lock)
 
 decl_simple_lock_data(,clock_lock)
+lck_grp_attr_t * settime_lock_grp_attr;
+lck_grp_t * settime_lock_grp;
+lck_attr_t * settime_lock_attr;
+lck_mtx_t settime_lock;
 
 #define clock_lock()   \
        simple_lock(&clock_lock)
 
 #define clock_lock()   \
        simple_lock(&clock_lock)
@@ -60,27 +104,195 @@ decl_simple_lock_data(,clock_lock)
 #define clock_lock_init()      \
        simple_lock_init(&clock_lock, 0)
 
 #define clock_lock_init()      \
        simple_lock_init(&clock_lock, 0)
 
+#ifdef kdp_simple_lock_is_acquired
+boolean_t kdp_clock_is_locked()
+{
+       return kdp_simple_lock_is_acquired(&clock_lock);
+}
+#endif
+
+struct bintime {
+       time_t  sec;
+       uint64_t frac;
+};
+
+static __inline void
+bintime_addx(struct bintime *_bt, uint64_t _x)
+{
+       uint64_t _u;
+
+       _u = _bt->frac;
+       _bt->frac += _x;
+       if (_u > _bt->frac)
+               _bt->sec++;
+}
+
+static __inline void
+bintime_subx(struct bintime *_bt, uint64_t _x)
+{
+       uint64_t _u;
+
+       _u = _bt->frac;
+       _bt->frac -= _x;
+       if (_u < _bt->frac)
+               _bt->sec--;
+}
+
+static __inline void
+bintime_addns(struct bintime *bt, uint64_t ns)
+{
+       bt->sec += ns/ (uint64_t)NSEC_PER_SEC;
+       ns = ns % (uint64_t)NSEC_PER_SEC;
+       if (ns) {
+               /* 18446744073 = int(2^64 / NSEC_PER_SEC) */
+               ns = ns * (uint64_t)18446744073LL;
+               bintime_addx(bt, ns);
+       }
+}
+
+static __inline void
+bintime_subns(struct bintime *bt, uint64_t ns)
+{
+       bt->sec -= ns/ (uint64_t)NSEC_PER_SEC;
+       ns = ns % (uint64_t)NSEC_PER_SEC;
+       if (ns) {
+               /* 18446744073 = int(2^64 / NSEC_PER_SEC) */
+               ns = ns * (uint64_t)18446744073LL;
+               bintime_subx(bt, ns);
+       }
+}
+
+static __inline void
+bintime_addxns(struct bintime *bt, uint64_t a, int64_t xns)
+{
+       uint64_t uxns = (xns > 0)?(uint64_t )xns:(uint64_t)-xns;
+       uint64_t ns = multi_overflow(a, uxns);
+       if (xns > 0) {
+               if (ns)
+                       bintime_addns(bt, ns);
+               ns = (a * uxns) / (uint64_t)NSEC_PER_SEC;
+               bintime_addx(bt, ns);
+       }
+       else{
+               if (ns)
+                       bintime_subns(bt, ns);
+               ns = (a * uxns) / (uint64_t)NSEC_PER_SEC;
+               bintime_subx(bt,ns);
+       }
+}
+
+
+static __inline void
+bintime_add(struct bintime *_bt, const struct bintime *_bt2)
+{
+       uint64_t _u;
+
+       _u = _bt->frac;
+       _bt->frac += _bt2->frac;
+       if (_u > _bt->frac)
+               _bt->sec++;
+       _bt->sec += _bt2->sec;
+}
+
+static __inline void
+bintime_sub(struct bintime *_bt, const struct bintime *_bt2)
+{
+       uint64_t _u;
+
+       _u = _bt->frac;
+       _bt->frac -= _bt2->frac;
+       if (_u < _bt->frac)
+               _bt->sec--;
+       _bt->sec -= _bt2->sec;
+}
+
+static __inline void
+clock2bintime(const clock_sec_t *secs, const clock_usec_t *microsecs, struct bintime *_bt)
+{
+
+       _bt->sec = *secs;
+       /* 18446744073709 = int(2^64 / 1000000) */
+       _bt->frac = *microsecs * (uint64_t)18446744073709LL;
+}
+
+static __inline void
+bintime2usclock(const struct bintime *_bt, clock_sec_t *secs, clock_usec_t *microsecs)
+{
+
+       *secs = _bt->sec;
+       *microsecs = ((uint64_t)USEC_PER_SEC * (uint32_t)(_bt->frac >> 32)) >> 32;
+}
+
+static __inline void
+bintime2nsclock(const struct bintime *_bt, clock_sec_t *secs, clock_usec_t *nanosecs)
+{
+
+       *secs = _bt->sec;
+       *nanosecs = ((uint64_t)NSEC_PER_SEC * (uint32_t)(_bt->frac >> 32)) >> 32;
+}
+
+static __inline void
+bintime2absolutetime(const struct bintime *_bt, uint64_t *abs)
+{
+       uint64_t nsec;
+       nsec = (uint64_t) _bt->sec * (uint64_t)NSEC_PER_SEC + (((uint64_t)NSEC_PER_SEC * (uint32_t)(_bt->frac >> 32)) >> 32);
+       nanoseconds_to_absolutetime(nsec, abs);
+}
+
+struct latched_time {
+        uint64_t monotonic_time_usec;
+        uint64_t mach_time;
+};
+
+extern int
+kernel_sysctlbyname(const char *name, void *oldp, size_t *oldlenp, void *newp, size_t newlen);
 
 /*
  *     Time of day (calendar) variables.
  *
  *     Algorithm:
  *
 
 /*
  *     Time of day (calendar) variables.
  *
  *     Algorithm:
  *
- *     TOD <- (seconds + epoch, fraction) <- CONV(current absolute time + offset)
+ *     TOD <- bintime + delta*scale
  *
  *
- *     where CONV converts absolute time units into seconds and a fraction.
+ *     where :
+ *     bintime is a cumulative offset that includes bootime and scaled time elapsed betweed bootime and last scale update.
+ *     delta is ticks elapsed since last scale update.
+ *     scale is computed according to an adjustment provided by ntp_kern.
  */
 static struct clock_calend {
  */
 static struct clock_calend {
-       uint64_t        epoch;
-       uint64_t        offset;
-
-       int32_t         adjdelta;       /* Nanosecond time delta for this adjustment period */
-       uint64_t        adjstart;       /* Absolute time value for start of this adjustment period */
-       uint32_t        adjoffset;      /* Absolute time offset for this adjustment period as absolute value */
+       uint64_t                s_scale_ns; /* scale to apply for each second elapsed, it converts in ns */
+       int64_t                 s_adj_nsx; /* additional adj to apply for each second elapsed, it is expressed in 64 bit frac of ns */
+       uint64_t                tick_scale_x; /* scale to apply for each tick elapsed, it converts in 64 bit frac of s */
+       uint64_t                offset_count; /* abs time from which apply current scales */
+       struct bintime          offset; /* cumulative offset expressed in (sec, 64 bits frac of a second) */
+       struct bintime          bintime; /* cumulative offset (it includes bootime) expressed in (sec, 64 bits frac of a second) */
+       struct bintime          boottime; /* boot time expressed in (sec, 64 bits frac of a second) */
+       struct bintime          basesleep;
 } clock_calend;
 
 } clock_calend;
 
+static uint64_t ticks_per_sec; /* ticks in a second (expressed in abs time) */
+
+#if DEVELOPMENT || DEBUG
+clock_sec_t last_utc_sec = 0;
+clock_usec_t last_utc_usec = 0;
+clock_sec_t max_utc_sec = 0;
+clock_sec_t last_sys_sec = 0;
+clock_usec_t last_sys_usec = 0;
+#endif
+
+#if DEVELOPMENT || DEBUG
+extern int g_should_log_clock_adjustments;
+
+static void print_all_clock_variables(const char*, clock_sec_t* pmu_secs, clock_usec_t* pmu_usec, clock_sec_t* sys_secs, clock_usec_t* sys_usec, struct clock_calend* calend_cp);
+static void print_all_clock_variables_internal(const char *, struct clock_calend* calend_cp);
+#else
+#define print_all_clock_variables(...) do { } while (0)
+#define print_all_clock_variables_internal(...) do { } while (0)
+#endif
+
 #if    CONFIG_DTRACE
 
 #if    CONFIG_DTRACE
 
+
 /*
  *     Unlocked calendar flipflop; this is used to track a clock_calend such
  *     that we can safely access a snapshot of a valid  clock_calend structure
 /*
  *     Unlocked calendar flipflop; this is used to track a clock_calend such
  *     that we can safely access a snapshot of a valid  clock_calend structure
@@ -100,35 +312,12 @@ static void clock_track_calend_nowait(void);
 
 #endif
 
 
 #endif
 
-/*
- *     Calendar adjustment variables and values.
- */
-#define calend_adjperiod       (NSEC_PER_SEC / 100)    /* adjustment period, ns */
-#define calend_adjskew         (40 * NSEC_PER_USEC)    /* "standard" skew, ns / period */
-#define        calend_adjbig           (NSEC_PER_SEC)                  /* use 10x skew above adjbig ns */
-
-static int64_t                         calend_adjtotal;                /* Nanosecond remaining total adjustment */
-static uint64_t                                calend_adjdeadline;             /* Absolute time value for next adjustment period */
-static uint32_t                                calend_adjinterval;             /* Absolute time interval of adjustment period */
-
-static timer_call_data_t       calend_adjcall;
-static uint32_t                                calend_adjactive;
-
-static uint32_t                calend_set_adjustment(
-                                               long                    *secs,
-                                               int                             *microsecs);
-
-static void                    calend_adjust_call(void);
-static uint32_t                calend_adjust(void);
+void _clock_delay_until_deadline(uint64_t interval, uint64_t deadline);
+void _clock_delay_until_deadline_with_leeway(uint64_t interval, uint64_t deadline, uint64_t leeway);
 
 
-static thread_call_data_t      calend_wakecall;
-
-extern void    IOKitResetTime(void);
-
-void _clock_delay_until_deadline(uint64_t              interval,
-                                                                uint64_t               deadline);
-
-static uint64_t                clock_boottime;                         /* Seconds boottime epoch */
+/* Boottime variables*/
+static uint64_t clock_boottime;
+static uint32_t clock_boottime_usec;
 
 #define TIME_ADD(rsecs, secs, rfrac, frac, unit)       \
 MACRO_BEGIN                                                                                    \
 
 #define TIME_ADD(rsecs, secs, rfrac, frac, unit)       \
 MACRO_BEGIN                                                                                    \
@@ -156,12 +345,19 @@ MACRO_END
 void
 clock_config(void)
 {
 void
 clock_config(void)
 {
+
        clock_lock_init();
 
        clock_lock_init();
 
-       timer_call_setup(&calend_adjcall, (timer_call_func_t)calend_adjust_call, NULL);
-       thread_call_setup(&calend_wakecall, (thread_call_func_t)IOKitResetTime, NULL);
+       settime_lock_grp_attr = lck_grp_attr_alloc_init();
+       settime_lock_grp = lck_grp_alloc_init("settime grp", settime_lock_grp_attr);
+       settime_lock_attr = lck_attr_alloc_init();
+       lck_mtx_init(&settime_lock, settime_lock_grp, settime_lock_attr);
 
        clock_oldconfig();
 
        clock_oldconfig();
+
+       ntp_init();
+
+       nanoseconds_to_absolutetime((uint64_t)NSEC_PER_SEC, &ticks_per_sec);
 }
 
 /*
 }
 
 /*
@@ -188,9 +384,6 @@ clock_timebase_init(void)
 {
        uint64_t        abstime;
 
 {
        uint64_t        abstime;
 
-       nanoseconds_to_absolutetime(calend_adjperiod, &abstime);
-       calend_adjinterval = (uint32_t)abstime;
-
        nanoseconds_to_absolutetime(NSEC_PER_SEC / 100, &abstime);
        hz_tick_interval = (uint32_t)abstime;
 
        nanoseconds_to_absolutetime(NSEC_PER_SEC / 100, &abstime);
        hz_tick_interval = (uint32_t)abstime;
 
@@ -207,7 +400,7 @@ mach_timebase_info_trap(
        struct mach_timebase_info_trap_args *args)
 {
        mach_vm_address_t                       out_info_addr = args->info;
        struct mach_timebase_info_trap_args *args)
 {
        mach_vm_address_t                       out_info_addr = args->info;
-       mach_timebase_info_data_t       info;
+       mach_timebase_info_data_t       info = {};
 
        clock_timebase_info(&info);
 
 
        clock_timebase_info(&info);
 
@@ -228,12 +421,191 @@ mach_timebase_info_trap(
  */
 void
 clock_get_calendar_microtime(
  */
 void
 clock_get_calendar_microtime(
-       clock_sec_t                     *secs,
+       clock_sec_t             *secs,
        clock_usec_t            *microsecs)
 {
        clock_get_calendar_absolute_and_microtime(secs, microsecs, NULL);
 }
 
        clock_usec_t            *microsecs)
 {
        clock_get_calendar_absolute_and_microtime(secs, microsecs, NULL);
 }
 
+/*
+ * get_scale_factors_from_adj:
+ *
+ * computes scale factors from the value given in adjustment.
+ *
+ * Part of the code has been taken from tc_windup of FreeBSD
+ * written by Poul-Henning Kamp <phk@FreeBSD.ORG>, Julien Ridoux and
+ * Konstantin Belousov.
+ * https://github.com/freebsd/freebsd/blob/master/sys/kern/kern_tc.c
+ */
+static void
+get_scale_factors_from_adj(int64_t adjustment, uint64_t* tick_scale_x, uint64_t* s_scale_ns, int64_t* s_adj_nsx)
+{
+       uint64_t scale;
+       int64_t nano, frac;
+
+       /*-
+        * Calculating the scaling factor.  We want the number of 1/2^64
+        * fractions of a second per period of the hardware counter, taking
+        * into account the th_adjustment factor which the NTP PLL/adjtime(2)
+        * processing provides us with.
+        *
+        * The th_adjustment is nanoseconds per second with 32 bit binary
+        * fraction and we want 64 bit binary fraction of second:
+        *
+        *       x = a * 2^32 / 10^9 = a * 4.294967296
+        *
+        * The range of th_adjustment is +/- 5000PPM so inside a 64bit int
+        * we can only multiply by about 850 without overflowing, that
+        * leaves no suitably precise fractions for multiply before divide.
+        *
+        * Divide before multiply with a fraction of 2199/512 results in a
+        * systematic undercompensation of 10PPM of th_adjustment.  On a
+        * 5000PPM adjustment this is a 0.05PPM error.  This is acceptable.
+        *
+        * We happily sacrifice the lowest of the 64 bits of our result
+        * to the goddess of code clarity.
+        *
+        */
+       scale = (uint64_t)1 << 63;
+       scale += (adjustment / 1024) * 2199;
+       scale /= ticks_per_sec;
+       *tick_scale_x = scale * 2;
+
+       /*
+        * hi part of adj
+        * it contains ns (without fraction) to add to the next sec.
+        * Get ns scale factor for the next sec.
+        */
+       nano = (adjustment > 0)? adjustment >> 32 : -((-adjustment) >> 32);
+       scale = (uint64_t) NSEC_PER_SEC;
+       scale += nano;
+       *s_scale_ns = scale;
+
+       /*
+        * lo part of adj
+        * it contains 32 bit frac of ns to add to the next sec.
+        * Keep it as additional adjustment for the next sec.
+        */
+       frac = (adjustment > 0)? ((uint32_t) adjustment) : -((uint32_t) (-adjustment));
+       *s_adj_nsx = (frac>0)? frac << 32 : -( (-frac) << 32);
+
+       return;
+}
+
+/*
+ * scale_delta:
+ *
+ * returns a bintime struct representing delta scaled accordingly to the
+ * scale factors provided to this function.
+ */
+static struct bintime
+scale_delta(uint64_t delta, uint64_t tick_scale_x, uint64_t s_scale_ns, int64_t s_adj_nsx)
+{
+       uint64_t sec, new_ns, over;
+       struct bintime bt;
+
+       bt.sec = 0;
+       bt.frac = 0;
+
+       /*
+        * If more than one second is elapsed,
+        * scale fully elapsed seconds using scale factors for seconds.
+        * s_scale_ns -> scales sec to ns.
+        * s_adj_nsx -> additional adj expressed in 64 bit frac of ns to apply to each sec.
+        */
+       if (delta > ticks_per_sec) {
+               sec = (delta/ticks_per_sec);
+               new_ns = sec * s_scale_ns;
+               bintime_addns(&bt, new_ns);
+               if (s_adj_nsx) {
+                       if (sec == 1) {
+                               /* shortcut, no overflow can occur */
+                               if (s_adj_nsx > 0)
+                                       bintime_addx(&bt, (uint64_t)s_adj_nsx/ (uint64_t)NSEC_PER_SEC);
+                               else
+                                       bintime_subx(&bt, (uint64_t)-s_adj_nsx/ (uint64_t)NSEC_PER_SEC);
+                       }
+                       else{
+                               /*
+                                * s_adj_nsx is 64 bit frac of ns.
+                                * sec*s_adj_nsx might overflow in int64_t.
+                                * use bintime_addxns to not lose overflowed ns.
+                                */
+                               bintime_addxns(&bt, sec, s_adj_nsx);
+                       }
+               }
+               delta = (delta % ticks_per_sec);
+        }
+
+       over = multi_overflow(tick_scale_x, delta);
+       if(over){
+               bt.sec += over;
+       }
+
+       /*
+        * scale elapsed ticks using the scale factor for ticks.
+        */
+       bintime_addx(&bt, delta * tick_scale_x);
+
+       return bt;
+}
+
+/*
+ * get_scaled_time:
+ *
+ * returns the scaled time of the time elapsed from the last time
+ * scale factors were updated to now.
+ */
+static struct bintime
+get_scaled_time(uint64_t now)
+{
+       uint64_t delta;
+
+       /*
+        * Compute ticks elapsed since last scale update.
+        * This time will be scaled according to the value given by ntp kern.
+        */
+       delta = now - clock_calend.offset_count;
+
+       return scale_delta(delta, clock_calend.tick_scale_x, clock_calend.s_scale_ns, clock_calend.s_adj_nsx);
+}
+
+static void
+clock_get_calendar_absolute_and_microtime_locked(
+       clock_sec_t             *secs,
+       clock_usec_t            *microsecs,
+       uint64_t                *abstime)
+{
+       uint64_t now;
+       struct bintime bt;
+
+       now  = mach_absolute_time();
+       if (abstime)
+               *abstime = now;
+
+       bt = get_scaled_time(now);
+       bintime_add(&bt, &clock_calend.bintime);
+       bintime2usclock(&bt, secs, microsecs);
+}
+
+static void
+clock_get_calendar_absolute_and_nanotime_locked(
+       clock_sec_t             *secs,
+       clock_usec_t            *nanosecs,
+       uint64_t                *abstime)
+{
+       uint64_t now;
+       struct bintime bt;
+
+       now  = mach_absolute_time();
+       if (abstime)
+               *abstime = now;
+
+       bt = get_scaled_time(now);
+       bintime_add(&bt, &clock_calend.bintime);
+       bintime2nsclock(&bt, secs, nanosecs);
+}
+
 /*
  *     clock_get_calendar_absolute_and_microtime:
  *
 /*
  *     clock_get_calendar_absolute_and_microtime:
  *
@@ -244,47 +616,16 @@ clock_get_calendar_microtime(
  */
 void
 clock_get_calendar_absolute_and_microtime(
  */
 void
 clock_get_calendar_absolute_and_microtime(
-       clock_sec_t                     *secs,
+       clock_sec_t             *secs,
        clock_usec_t            *microsecs,
        uint64_t                *abstime)
 {
        clock_usec_t            *microsecs,
        uint64_t                *abstime)
 {
-       uint64_t                now;
        spl_t                   s;
 
        s = splclock();
        clock_lock();
 
        spl_t                   s;
 
        s = splclock();
        clock_lock();
 
-       now = mach_absolute_time();
-       if (abstime)
-               *abstime = now;
-
-       if (clock_calend.adjdelta < 0) {
-               uint32_t        t32;
-
-               /* 
-                * Since offset is decremented during a negative adjustment,
-                * ensure that time increases monotonically without going
-                * temporarily backwards.
-                * If the delta has not yet passed, now is set to the start
-                * of the current adjustment period; otherwise, we're between
-                * the expiry of the delta and the next call to calend_adjust(),
-                * and we offset accordingly.
-                */
-               if (now > clock_calend.adjstart) {
-                       t32 = (uint32_t)(now - clock_calend.adjstart);
-
-                       if (t32 > clock_calend.adjoffset)
-                               now -= clock_calend.adjoffset;
-                       else
-                               now = clock_calend.adjstart;
-               }
-       }
-
-       now += clock_calend.offset;
-
-       absolutetime_to_microtime(now, secs, microsecs);
-
-       *secs += (clock_sec_t)clock_calend.epoch;
+       clock_get_calendar_absolute_and_microtime_locked(secs, microsecs, abstime);
 
        clock_unlock();
        splx(s);
 
        clock_unlock();
        splx(s);
@@ -302,37 +643,15 @@ clock_get_calendar_absolute_and_microtime(
  */
 void
 clock_get_calendar_nanotime(
  */
 void
 clock_get_calendar_nanotime(
-       clock_sec_t                     *secs,
+       clock_sec_t             *secs,
        clock_nsec_t            *nanosecs)
 {
        clock_nsec_t            *nanosecs)
 {
-       uint64_t                now;
        spl_t                   s;
 
        s = splclock();
        clock_lock();
 
        spl_t                   s;
 
        s = splclock();
        clock_lock();
 
-       now = mach_absolute_time();
-
-       if (clock_calend.adjdelta < 0) {
-               uint32_t        t32;
-
-               if (now > clock_calend.adjstart) {
-                       t32 = (uint32_t)(now - clock_calend.adjstart);
-
-                       if (t32 > clock_calend.adjoffset)
-                               now -= clock_calend.adjoffset;
-                       else
-                               now = clock_calend.adjstart;
-               }
-       }
-
-       now += clock_calend.offset;
-
-       absolutetime_to_microtime(now, secs, nanosecs);
-
-       *nanosecs *= NSEC_PER_USEC;
-
-       *secs += (clock_sec_t)clock_calend.epoch;
+       clock_get_calendar_absolute_and_nanotime_locked(secs, nanosecs, NULL);
 
        clock_unlock();
        splx(s);
 
        clock_unlock();
        splx(s);
@@ -351,41 +670,56 @@ clock_get_calendar_nanotime(
  */
 void
 clock_gettimeofday(
  */
 void
 clock_gettimeofday(
-       clock_sec_t             *secs,
+       clock_sec_t     *secs,
        clock_usec_t    *microsecs)
        clock_usec_t    *microsecs)
+{
+       clock_gettimeofday_and_absolute_time(secs, microsecs, NULL);
+}
+
+void
+clock_gettimeofday_and_absolute_time(
+       clock_sec_t     *secs,
+       clock_usec_t    *microsecs,
+       uint64_t        *mach_time)
 {
        uint64_t                now;
        spl_t                   s;
 {
        uint64_t                now;
        spl_t                   s;
+       struct bintime  bt;
 
        s = splclock();
        clock_lock();
 
        now = mach_absolute_time();
 
        s = splclock();
        clock_lock();
 
        now = mach_absolute_time();
+       bt = get_scaled_time(now);
+       bintime_add(&bt, &clock_calend.bintime);
+       bintime2usclock(&bt, secs, microsecs);
 
 
-       if (clock_calend.adjdelta >= 0) {
-               clock_gettimeofday_set_commpage(now, clock_calend.epoch, clock_calend.offset, secs, microsecs);
-       }
-       else {
-               uint32_t        t32;
-
-               if (now > clock_calend.adjstart) {
-                       t32 = (uint32_t)(now - clock_calend.adjstart);
-
-                       if (t32 > clock_calend.adjoffset)
-                               now -= clock_calend.adjoffset;
-                       else
-                               now = clock_calend.adjstart;
-               }
+       clock_gettimeofday_set_commpage(now, bt.sec, bt.frac, clock_calend.tick_scale_x, ticks_per_sec);
 
 
-               now += clock_calend.offset;
-
-               absolutetime_to_microtime(now, secs, microsecs);
+       clock_unlock();
+       splx(s);
 
 
-               *secs += (clock_sec_t)clock_calend.epoch;
+       if (mach_time) {
+               *mach_time = now;
        }
        }
+}
 
 
-       clock_unlock();
-       splx(s);
+static void
+update_basesleep(struct bintime delta, bool forward)
+{
+       /*
+        * Update basesleep only if the platform does not have monotonic clock.
+        * In that case the sleep time computation will use the PMU time
+        * which offset gets modified by settimeofday.
+        * We don't need this for mononic clock because in that case the sleep
+        * time computation is independent from the offset value of the PMU.
+        */
+       if (!has_monotonic_clock) {
+               if (forward)
+                       bintime_add(&clock_calend.basesleep, &delta);
+               else
+                       bintime_sub(&clock_calend.basesleep, &delta);
+       }
 }
 
 /*
 }
 
 /*
@@ -402,390 +736,681 @@ clock_gettimeofday(
  */
 void
 clock_set_calendar_microtime(
  */
 void
 clock_set_calendar_microtime(
-       clock_sec_t                     secs,
+       clock_sec_t             secs,
        clock_usec_t            microsecs)
 {
        clock_usec_t            microsecs)
 {
-       clock_sec_t                     sys;
-       clock_usec_t            microsys;
-       clock_sec_t                     newsecs;
-    clock_usec_t        newmicrosecs;
-       spl_t                           s;
+       uint64_t                absolutesys;
+       clock_sec_t             newsecs;
+       clock_sec_t             oldsecs;
+       clock_usec_t            newmicrosecs;
+       clock_usec_t            oldmicrosecs;
+       uint64_t                commpage_value;
+       spl_t                   s;
+       struct bintime          bt;
+       clock_sec_t             deltasecs;
+       clock_usec_t            deltamicrosecs;
 
 
-    newsecs = secs;
-    newmicrosecs = microsecs;
+       newsecs = secs;
+       newmicrosecs = microsecs;
+
+       /*
+        * settime_lock mtx is used to avoid that racing settimeofdays update the wall clock and
+        * the platform clock concurrently.
+        *
+        * clock_lock cannot be used for this race because it is acquired from interrupt context
+        * and it needs interrupts disabled while instead updating the platform clock needs to be
+        * called with interrupts enabled.
+        */
+       lck_mtx_lock(&settime_lock);
 
        s = splclock();
        clock_lock();
 
 
        s = splclock();
        clock_lock();
 
+#if DEVELOPMENT || DEBUG
+       struct clock_calend clock_calend_cp = clock_calend;
+#endif
        commpage_disable_timestamp();
 
        commpage_disable_timestamp();
 
-       /*
-        *      Calculate the new calendar epoch based on
-        *      the new value and the system clock.
-        */
-       clock_get_system_microtime(&sys, &microsys);
-       TIME_SUB(secs, sys, microsecs, microsys, USEC_PER_SEC);
-
        /*
         *      Adjust the boottime based on the delta.
         */
        /*
         *      Adjust the boottime based on the delta.
         */
-       clock_boottime += secs - clock_calend.epoch;
+       clock_get_calendar_absolute_and_microtime_locked(&oldsecs, &oldmicrosecs, &absolutesys);
+
+#if DEVELOPMENT || DEBUG
+       if (g_should_log_clock_adjustments) {
+               os_log(OS_LOG_DEFAULT, "%s wall %lu s %d u computed with %llu abs\n",
+                      __func__, (unsigned long)oldsecs, oldmicrosecs, absolutesys);
+               os_log(OS_LOG_DEFAULT, "%s requested %lu s %d u\n",
+                      __func__,  (unsigned long)secs, microsecs );
+       }
+#endif
 
 
-       /*
-        *      Set the new calendar epoch.
-        */
-       clock_calend.epoch = secs;
+       if (oldsecs < secs || (oldsecs == secs && oldmicrosecs < microsecs)) {
+               // moving forwards
+               deltasecs = secs;
+               deltamicrosecs = microsecs;
 
 
-       nanoseconds_to_absolutetime((uint64_t)microsecs * NSEC_PER_USEC, &clock_calend.offset);
+               TIME_SUB(deltasecs, oldsecs, deltamicrosecs, oldmicrosecs, USEC_PER_SEC);
 
 
-       /*
-        *      Cancel any adjustment in progress.
-        */
-       calend_adjtotal = clock_calend.adjdelta = 0;
+#if DEVELOPMENT || DEBUG
+               if (g_should_log_clock_adjustments) {
+                       os_log(OS_LOG_DEFAULT, "%s delta requested %lu s %d u\n",
+                              __func__, (unsigned long)deltasecs, deltamicrosecs);
+               }
+#endif
+
+               TIME_ADD(clock_boottime, deltasecs, clock_boottime_usec, deltamicrosecs, USEC_PER_SEC);
+               clock2bintime(&deltasecs, &deltamicrosecs, &bt);
+               bintime_add(&clock_calend.boottime, &bt);
+               update_basesleep(bt, TRUE);
+       } else {
+               // moving backwards
+               deltasecs = oldsecs;
+               deltamicrosecs = oldmicrosecs;
+
+               TIME_SUB(deltasecs, secs, deltamicrosecs, microsecs, USEC_PER_SEC);
+#if DEVELOPMENT || DEBUG
+               if (g_should_log_clock_adjustments) {
+                       os_log(OS_LOG_DEFAULT, "%s negative delta requested %lu s %d u\n",
+                              __func__, (unsigned long)deltasecs, deltamicrosecs);
+               }
+#endif
+
+               TIME_SUB(clock_boottime, deltasecs, clock_boottime_usec, deltamicrosecs, USEC_PER_SEC);
+               clock2bintime(&deltasecs, &deltamicrosecs, &bt);
+               bintime_sub(&clock_calend.boottime, &bt);
+               update_basesleep(bt, FALSE);
+       }
+
+       clock_calend.bintime = clock_calend.boottime;
+       bintime_add(&clock_calend.bintime, &clock_calend.offset);
+
+       clock2bintime((clock_sec_t *) &secs, (clock_usec_t *) &microsecs, &bt);
+
+       clock_gettimeofday_set_commpage(absolutesys, bt.sec, bt.frac, clock_calend.tick_scale_x, ticks_per_sec);
+
+#if DEVELOPMENT || DEBUG
+       struct clock_calend clock_calend_cp1 = clock_calend;
+#endif
+
+       commpage_value = clock_boottime * USEC_PER_SEC + clock_boottime_usec;
 
        clock_unlock();
 
        clock_unlock();
+       splx(s);
 
        /*
         *      Set the new value for the platform clock.
 
        /*
         *      Set the new value for the platform clock.
+        *      This call might block, so interrupts must be enabled.
         */
         */
+#if DEVELOPMENT || DEBUG
+       uint64_t now_b = mach_absolute_time();
+#endif
+
        PESetUTCTimeOfDay(newsecs, newmicrosecs);
 
        PESetUTCTimeOfDay(newsecs, newmicrosecs);
 
-       splx(s);
+#if DEVELOPMENT || DEBUG
+       uint64_t now_a = mach_absolute_time();
+       if (g_should_log_clock_adjustments) {
+               os_log(OS_LOG_DEFAULT, "%s mach bef PESet %llu mach aft %llu \n", __func__, now_b, now_a);
+       }
+#endif
+
+       print_all_clock_variables_internal(__func__, &clock_calend_cp);
+       print_all_clock_variables_internal(__func__, &clock_calend_cp1);
+
+       commpage_update_boottime(commpage_value);
 
        /*
         *      Send host notifications.
         */
        host_notify_calendar_change();
 
        /*
         *      Send host notifications.
         */
        host_notify_calendar_change();
-       
+       host_notify_calendar_set();
+
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
+
+       lck_mtx_unlock(&settime_lock);
 }
 
 }
 
+uint64_t mach_absolutetime_asleep = 0;
+uint64_t mach_absolutetime_last_sleep = 0;
+
+void
+clock_get_calendar_uptime(clock_sec_t *secs)
+{
+       uint64_t now;
+       spl_t s;
+       struct bintime bt;
+
+       s = splclock();
+       clock_lock();
+
+       now = mach_absolute_time();
+
+       bt = get_scaled_time(now);
+       bintime_add(&bt, &clock_calend.offset);
+
+       *secs = bt.sec;
+
+       clock_unlock();
+       splx(s);
+}
+
+
 /*
 /*
- *     clock_initialize_calendar:
- *
- *     Set the calendar and related clocks
- *     from the platform clock at boot or
- *     wake event.
+ * clock_update_calendar:
  *
  *
- *     Also sends host notifications.
+ * called by ntp timer to update scale factors.
  */
 void
  */
 void
-clock_initialize_calendar(void)
+clock_update_calendar(void)
 {
 {
-       clock_sec_t                     sys, secs;
-       clock_usec_t            microsys, microsecs;
-       spl_t                           s;
 
 
-    PEGetUTCTimeOfDay(&secs, &microsecs);
+       uint64_t now, delta;
+       struct bintime bt;
+       spl_t s;
+       int64_t adjustment;
 
        s = splclock();
        clock_lock();
 
 
        s = splclock();
        clock_lock();
 
-       commpage_disable_timestamp();
+       now  = mach_absolute_time();
 
 
-       if ((long)secs >= (long)clock_boottime) {
-               /*
-                *      Initialize the boot time based on the platform clock.
-                */
-               if (clock_boottime == 0)
-                       clock_boottime = secs;
+       /*
+        * scale the time elapsed since the last update and
+        * add it to offset.
+        */
+       bt = get_scaled_time(now);
+       bintime_add(&clock_calend.offset, &bt);
 
 
-               /*
-                *      Calculate the new calendar epoch based on
-                *      the platform clock and the system clock.
-                */
-               clock_get_system_microtime(&sys, &microsys);
-               TIME_SUB(secs, sys, microsecs, microsys, USEC_PER_SEC);
+       /*
+        * update the base from which apply next scale factors.
+        */
+       delta = now - clock_calend.offset_count;
+       clock_calend.offset_count += delta;
 
 
-               /*
-                *      Set the new calendar epoch.
-                */
-               clock_calend.epoch = secs;
+       clock_calend.bintime = clock_calend.offset;
+       bintime_add(&clock_calend.bintime, &clock_calend.boottime);
 
 
-               nanoseconds_to_absolutetime((uint64_t)microsecs * NSEC_PER_USEC, &clock_calend.offset);
+       /*
+        * recompute next adjustment.
+        */
+       ntp_update_second(&adjustment, clock_calend.bintime.sec);
 
 
-               /*
-                *       Cancel any adjustment in progress.
-                */
-               calend_adjtotal = clock_calend.adjdelta = 0;
+#if DEVELOPMENT || DEBUG
+       if (g_should_log_clock_adjustments) {
+               os_log(OS_LOG_DEFAULT, "%s adjustment %lld\n", __func__, adjustment);
        }
        }
+#endif
+       
+       /*
+        * recomputing scale factors.
+        */
+       get_scale_factors_from_adj(adjustment, &clock_calend.tick_scale_x, &clock_calend.s_scale_ns, &clock_calend.s_adj_nsx);
+
+       clock_gettimeofday_set_commpage(now, clock_calend.bintime.sec, clock_calend.bintime.frac, clock_calend.tick_scale_x, ticks_per_sec);
+
+#if DEVELOPMENT || DEBUG
+       struct clock_calend calend_cp = clock_calend;
+#endif
 
        clock_unlock();
        splx(s);
 
 
        clock_unlock();
        splx(s);
 
-       /*
-        *      Send host notifications.
-        */
-       host_notify_calendar_change();
+       print_all_clock_variables(__func__, NULL,NULL,NULL,NULL, &calend_cp);
+}
+
+
+#if DEVELOPMENT || DEBUG
+
+void print_all_clock_variables_internal(const char* func, struct clock_calend* clock_calend_cp)
+{
+       clock_sec_t     offset_secs;
+       clock_usec_t    offset_microsecs;
+       clock_sec_t     bintime_secs;
+       clock_usec_t    bintime_microsecs;
+       clock_sec_t     bootime_secs;
+       clock_usec_t    bootime_microsecs;
        
        
-#if CONFIG_DTRACE
-       clock_track_calend_nowait();
-#endif
+       if (!g_should_log_clock_adjustments)
+                return;
+
+       bintime2usclock(&clock_calend_cp->offset, &offset_secs, &offset_microsecs);
+       bintime2usclock(&clock_calend_cp->bintime, &bintime_secs, &bintime_microsecs);
+       bintime2usclock(&clock_calend_cp->boottime, &bootime_secs, &bootime_microsecs);
+
+       os_log(OS_LOG_DEFAULT, "%s s_scale_ns %llu s_adj_nsx %lld tick_scale_x %llu offset_count %llu\n",
+              func , clock_calend_cp->s_scale_ns, clock_calend_cp->s_adj_nsx,
+              clock_calend_cp->tick_scale_x, clock_calend_cp->offset_count);
+       os_log(OS_LOG_DEFAULT, "%s offset.sec %ld offset.frac %llu offset_secs %lu offset_microsecs %d\n",
+              func, clock_calend_cp->offset.sec, clock_calend_cp->offset.frac,
+              (unsigned long)offset_secs, offset_microsecs);
+       os_log(OS_LOG_DEFAULT, "%s bintime.sec %ld bintime.frac %llu bintime_secs %lu bintime_microsecs %d\n",
+              func, clock_calend_cp->bintime.sec, clock_calend_cp->bintime.frac,
+              (unsigned long)bintime_secs, bintime_microsecs);
+       os_log(OS_LOG_DEFAULT, "%s bootime.sec %ld bootime.frac %llu bootime_secs %lu bootime_microsecs %d\n",
+              func, clock_calend_cp->boottime.sec, clock_calend_cp->boottime.frac,
+              (unsigned long)bootime_secs, bootime_microsecs);
+
+       clock_sec_t     basesleep_secs;
+        clock_usec_t    basesleep_microsecs;
+       
+       bintime2usclock(&clock_calend_cp->basesleep, &basesleep_secs, &basesleep_microsecs);
+       os_log(OS_LOG_DEFAULT, "%s basesleep.sec %ld basesleep.frac %llu basesleep_secs %lu basesleep_microsecs %d\n",
+              func, clock_calend_cp->basesleep.sec, clock_calend_cp->basesleep.frac,
+              (unsigned long)basesleep_secs, basesleep_microsecs);
+
 }
 
 }
 
-/*
- *     clock_get_boottime_nanotime:
- *
- *     Return the boottime, used by sysctl.
- */
-void
-clock_get_boottime_nanotime(
-       clock_sec_t                     *secs,
-       clock_nsec_t            *nanosecs)
+
+void print_all_clock_variables(const char* func, clock_sec_t* pmu_secs, clock_usec_t* pmu_usec, clock_sec_t* sys_secs, clock_usec_t* sys_usec, struct clock_calend* clock_calend_cp)
 {
 {
-       spl_t   s;
+       if (!g_should_log_clock_adjustments)
+               return;
 
 
-       s = splclock();
-       clock_lock();
+       struct bintime  bt;
+       clock_sec_t     wall_secs;
+       clock_usec_t    wall_microsecs;
+       uint64_t now;
+       uint64_t delta;
 
 
-       *secs = (clock_sec_t)clock_boottime;
-       *nanosecs = 0;
+       if (pmu_secs) {
+               os_log(OS_LOG_DEFAULT, "%s PMU %lu s %d u \n", func, (unsigned long)*pmu_secs, *pmu_usec); 
+       }
+       if (sys_secs) {
+               os_log(OS_LOG_DEFAULT, "%s sys %lu s %d u \n", func, (unsigned long)*sys_secs, *sys_usec);
+       }
 
 
-       clock_unlock();
-       splx(s);
+       print_all_clock_variables_internal(func, clock_calend_cp);
+
+       now = mach_absolute_time();
+        delta = now - clock_calend_cp->offset_count;
+
+        bt = scale_delta(delta, clock_calend_cp->tick_scale_x, clock_calend_cp->s_scale_ns, clock_calend_cp->s_adj_nsx);
+       bintime_add(&bt, &clock_calend_cp->bintime);
+       bintime2usclock(&bt, &wall_secs, &wall_microsecs);
+
+       os_log(OS_LOG_DEFAULT, "%s wall %lu s %d u computed with %llu abs\n",
+              func, (unsigned long)wall_secs, wall_microsecs, now);
 }
 
 }
 
+
+#endif /* DEVELOPMENT || DEBUG */
+
+
 /*
 /*
- *     clock_adjtime:
+ *     clock_initialize_calendar:
  *
  *
- *     Interface to adjtime() syscall.
+ *     Set the calendar and related clocks
+ *     from the platform clock at boot.
  *
  *
- *     Calculates adjustment variables and
- *     initiates adjustment.
+ *     Also sends host notifications.
  */
 void
  */
 void
-clock_adjtime(
-       long            *secs,
-       int                     *microsecs)
+clock_initialize_calendar(void)
 {
 {
-       uint32_t        interval;
-       spl_t           s;
+       clock_sec_t             sys;  // sleepless time since boot in seconds
+       clock_sec_t             secs; // Current UTC time
+       clock_sec_t             utc_offset_secs; // Difference in current UTC time and sleepless time since boot
+       clock_usec_t            microsys;  
+       clock_usec_t            microsecs; 
+       clock_usec_t            utc_offset_microsecs; 
+       spl_t                   s;
+       struct bintime          bt;
+       struct bintime          monotonic_bt;
+       struct latched_time     monotonic_time;
+       uint64_t                monotonic_usec_total;
+       clock_sec_t             sys2, monotonic_sec;
+        clock_usec_t            microsys2, monotonic_usec;
+        size_t                  size;
+
+       //Get PMU time with offset and corresponding sys time
+       PEGetUTCTimeOfDay(&secs, &microsecs);
+       clock_get_system_microtime(&sys, &microsys);
+
+       /*
+        * If the platform has a monotonic clock, use kern.monotonicclock_usecs
+        * to estimate the sleep/wake time, otherwise use the PMU and adjustments
+        * provided through settimeofday to estimate the sleep time.
+        * NOTE: the latter case relies that the kernel is the only component
+        * to set the PMU offset.
+        */
+       size = sizeof(monotonic_time);
+       if (kernel_sysctlbyname("kern.monotonicclock_usecs", &monotonic_time, &size, NULL, 0) != 0) {
+               has_monotonic_clock = 0;
+               os_log(OS_LOG_DEFAULT, "%s system does not have monotonic clock.\n", __func__);
+       } else {
+               has_monotonic_clock = 1;
+               monotonic_usec_total = monotonic_time.monotonic_time_usec;
+               absolutetime_to_microtime(monotonic_time.mach_time, &sys2, &microsys2);
+               os_log(OS_LOG_DEFAULT, "%s system has monotonic clock.\n", __func__);
+       }
 
        s = splclock();
        clock_lock();
 
 
        s = splclock();
        clock_lock();
 
-       interval = calend_set_adjustment(secs, microsecs);
-       if (interval != 0) {
-               calend_adjdeadline = mach_absolute_time() + interval;
-               if (!timer_call_enter(&calend_adjcall, calend_adjdeadline, TIMER_CALL_SYS_CRITICAL))
-                       calend_adjactive++;
-       }
-       else
-       if (timer_call_cancel(&calend_adjcall))
-               calend_adjactive--;
-
-       clock_unlock();
-       splx(s);
-}
+       commpage_disable_timestamp();
 
 
-static uint32_t
-calend_set_adjustment(
-       long                    *secs,
-       int                             *microsecs)
-{
-       uint64_t                now, t64;
-       int64_t                 total, ototal;
-       uint32_t                interval = 0;
+       utc_offset_secs = secs;
+       utc_offset_microsecs = microsecs;
 
 
-       /* 
-        * Compute the total adjustment time in nanoseconds.
-        */
-       total = ((int64_t)*secs * (int64_t)NSEC_PER_SEC) + (*microsecs * (int64_t)NSEC_PER_USEC);
+#if DEVELOPMENT || DEBUG
+       last_utc_sec = secs;
+       last_utc_usec = microsecs;
+       last_sys_sec = sys;
+       last_sys_usec = microsys;
+       if (secs > max_utc_sec)
+               max_utc_sec = secs;
+#endif
 
 
-       /* 
-        * Disable commpage gettimeofday().
+       /*
+        * We normally expect the UTC clock to be always-on and produce
+        * greater readings than the tick counter.  There may be corner cases
+        * due to differing clock resolutions (UTC clock is likely lower) and
+        * and errors reading the UTC clock (some implementations return 0
+        * on error) in which that doesn't hold true.  Bring the UTC measurements
+        * in-line with the tick counter measurements as a best effort in that case.
         */
         */
-       commpage_disable_timestamp();
+       //FIXME if the current time is prior than 1970 secs will be negative
+       if ((sys > secs) || ((sys == secs) && (microsys > microsecs))) {
+               os_log(OS_LOG_DEFAULT, "%s WARNING: PMU offset is less then sys PMU %lu s %d u sys %lu s %d u\n",
+                       __func__, (unsigned long) secs, microsecs, (unsigned long)sys, microsys);
+               secs = utc_offset_secs = sys;
+               microsecs = utc_offset_microsecs = microsys;
+       }
 
 
-       /* 
-        * Get current absolute time.
-        */
-       now = mach_absolute_time();
+       // PMU time with offset - sys
+       // This macro stores the subtraction result in utc_offset_secs and utc_offset_microsecs
+       TIME_SUB(utc_offset_secs, sys, utc_offset_microsecs, microsys, USEC_PER_SEC);
 
 
-       /* 
-        * Save the old adjustment total for later return.
-        */
-       ototal = calend_adjtotal;
+       clock2bintime(&utc_offset_secs, &utc_offset_microsecs, &bt);
 
        /*
 
        /*
-        * Is a new correction specified?
+        *      Initialize the boot time based on the platform clock.
         */
         */
-       if (total != 0) {
-               /*
-                * Set delta to the standard, small, adjustment skew.
-                */
-               int32_t         delta = calend_adjskew;
-
-               if (total > 0) {
-                       /*
-                        * Positive adjustment. If greater than the preset 'big' 
-                        * threshold, slew at a faster rate, capping if necessary.
-                        */
-                       if (total > (int64_t) calend_adjbig)
-                               delta *= 10;
-                       if (delta > total)
-                               delta = (int32_t)total;
-
-                       /* 
-                        * Convert the delta back from ns to absolute time and store in adjoffset.
-                        */
-                       nanoseconds_to_absolutetime((uint64_t)delta, &t64);
-                       clock_calend.adjoffset = (uint32_t)t64;
-               }
-               else {
-                       /*
-                        * Negative adjustment; therefore, negate the delta. If 
-                        * greater than the preset 'big' threshold, slew at a faster 
-                        * rate, capping if necessary.
-                        */
-                       if (total < (int64_t) -calend_adjbig)
-                               delta *= 10;
-                       delta = -delta;
-                       if (delta < total)
-                               delta = (int32_t)total;
-
-                       /* 
-                        * Save the current absolute time. Subsequent time operations occuring
-                        * during this negative correction can make use of this value to ensure 
-                        * that time increases monotonically.
-                        */
-                       clock_calend.adjstart = now;
-
-                       /* 
-                        * Convert the delta back from ns to absolute time and store in adjoffset.
-                        */
-                       nanoseconds_to_absolutetime((uint64_t)-delta, &t64);
-                       clock_calend.adjoffset = (uint32_t)t64;
-               }
+       clock_boottime = secs;
+       clock_boottime_usec = microsecs;
+       commpage_update_boottime(clock_boottime * USEC_PER_SEC + clock_boottime_usec);
 
 
-               /* 
-                * Store the total adjustment time in ns. 
-                */
-               calend_adjtotal = total;
-               
-               /* 
-                * Store the delta for this adjustment period in ns. 
-                */
-               clock_calend.adjdelta = delta;
+       nanoseconds_to_absolutetime((uint64_t)NSEC_PER_SEC, &ticks_per_sec);
+       clock_calend.boottime = bt;
+       clock_calend.bintime = bt;
+       clock_calend.offset.sec = 0;
+       clock_calend.offset.frac = 0;
 
 
-               /* 
-                * Set the interval in absolute time for later return. 
-                */
-               interval = calend_adjinterval;
-       }
-       else {
-               /* 
-                * No change; clear any prior adjustment.
-                */
-               calend_adjtotal = clock_calend.adjdelta = 0;
-       }
+       clock_calend.tick_scale_x = (uint64_t)1 << 63;
+       clock_calend.tick_scale_x /= ticks_per_sec;
+       clock_calend.tick_scale_x *= 2;
 
 
-       /* 
-        * If an prior correction was in progress, return the
-        * remaining uncorrected time from it. 
-        */
-       if (ototal != 0) {
-               *secs = (long)(ototal / (long)NSEC_PER_SEC);
-               *microsecs = (int)((ototal % (int)NSEC_PER_SEC) / (int)NSEC_PER_USEC);
+       clock_calend.s_scale_ns = NSEC_PER_SEC;
+       clock_calend.s_adj_nsx = 0;
+
+       if (has_monotonic_clock) {
+
+               monotonic_sec = monotonic_usec_total / (clock_sec_t)USEC_PER_SEC;
+               monotonic_usec = monotonic_usec_total % (clock_usec_t)USEC_PER_SEC;
+
+               // PMU time without offset - sys
+               // This macro stores the subtraction result in monotonic_sec and monotonic_usec
+               TIME_SUB(monotonic_sec, sys2, monotonic_usec, microsys2, USEC_PER_SEC);
+               clock2bintime(&monotonic_sec, &monotonic_usec, &monotonic_bt);
+
+               // set the baseleep as the difference between monotonic clock - sys
+               clock_calend.basesleep = monotonic_bt;
+       } else {
+               // set the baseleep as the difference between PMU clock - sys
+               clock_calend.basesleep = bt;
        }
        }
-       else
-               *secs = *microsecs = 0;
+       commpage_update_mach_continuous_time(mach_absolutetime_asleep);
+
+#if DEVELOPMENT || DEBUG
+       struct clock_calend clock_calend_cp = clock_calend;
+#endif
+
+       clock_unlock();
+       splx(s);
 
 
+        print_all_clock_variables(__func__, &secs, &microsecs, &sys, &microsys, &clock_calend_cp);
+
+       /*
+        *      Send host notifications.
+        */
+       host_notify_calendar_change();
+       
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
-       
-       return (interval);
 }
 
 }
 
-static void
-calend_adjust_call(void)
-{
-       uint32_t        interval;
-       spl_t           s;
 
 
-       s = splclock();
-       clock_lock();
-
-       if (--calend_adjactive == 0) {
-               interval = calend_adjust();
-               if (interval != 0) {
-                       clock_deadline_for_periodic_event(interval, mach_absolute_time(), &calend_adjdeadline);
+void
+clock_wakeup_calendar(void)
+{
+       clock_sec_t             sys;
+       clock_sec_t             secs;
+       clock_usec_t            microsys;
+       clock_usec_t            microsecs;
+       spl_t                   s;
+       struct bintime          bt, last_sleep_bt;
+       clock_sec_t             basesleep_s, last_sleep_sec;
+       clock_usec_t            basesleep_us, last_sleep_usec;
+       struct latched_time     monotonic_time;
+       uint64_t                monotonic_usec_total;
+       size_t                  size;
+       clock_sec_t secs_copy;
+        clock_usec_t microsecs_copy;
+#if DEVELOPMENT || DEBUG
+       clock_sec_t utc_sec;
+       clock_usec_t utc_usec;
+       PEGetUTCTimeOfDay(&utc_sec, &utc_usec);
+#endif
 
 
-                       if (!timer_call_enter(&calend_adjcall, calend_adjdeadline, TIMER_CALL_SYS_CRITICAL))
-                               calend_adjactive++;
+       /*
+        * If the platform has the monotonic clock use that to
+        * compute the sleep time. The monotonic clock does not have an offset
+        * that can be modified, so nor kernel or userspace can change the time
+        * of this clock, it can only monotonically increase over time.
+        * During sleep mach_absolute_time does not tick,
+        * so the sleep time is the difference betwen the current monotonic time
+        * less the absolute time and the previous difference stored at wake time.
+        *
+        * basesleep = monotonic - sys ---> computed at last wake
+        * sleep_time = (monotonic - sys) - basesleep
+        *
+        * If the platform does not support monotonic time we use the PMU time
+        * to compute the last sleep.
+        * The PMU time is the monotonic clock + an offset that can be set
+        * by kernel.
+        *
+        * IMPORTANT:
+        * We assume that only the kernel is setting the offset of the PMU and that
+        * it is doing it only througth the settimeofday interface.
+        *
+        * basesleep is the different between the PMU time and the mach_absolute_time
+        * at wake.
+        * During awake time settimeofday can change the PMU offset by a delta,
+        * and basesleep is shifted by the same delta applyed to the PMU. So the sleep
+        * time computation becomes:
+        *
+        * PMU = monotonic + PMU_offset
+        * basesleep = PMU - sys ---> computed at last wake
+        * basesleep += settimeofday_delta
+        * PMU_offset += settimeofday_delta
+        * sleep_time = (PMU - sys) - basesleep
+        */
+       if (has_monotonic_clock) {
+               //Get monotonic time with corresponding sys time
+               size = sizeof(monotonic_time);
+               if (kernel_sysctlbyname("kern.monotonicclock_usecs", &monotonic_time, &size, NULL, 0) != 0) {
+                       panic("%s: could not call kern.monotonicclock_usecs", __func__);
                }
                }
-       }
+               monotonic_usec_total = monotonic_time.monotonic_time_usec;
+               absolutetime_to_microtime(monotonic_time.mach_time, &sys, &microsys);
 
 
-       clock_unlock();
-       splx(s);
-}
+               secs = monotonic_usec_total / (clock_sec_t)USEC_PER_SEC;
+               microsecs = monotonic_usec_total % (clock_usec_t)USEC_PER_SEC;
+       } else {
+               //Get PMU time with offset and corresponding sys time
+               PEGetUTCTimeOfDay(&secs, &microsecs);
+               clock_get_system_microtime(&sys, &microsys);
 
 
-static uint32_t
-calend_adjust(void)
-{
-       uint64_t                now, t64;
-       int32_t                 delta;
-       uint32_t                interval = 0;
+       }
 
 
+       s = splclock();
+       clock_lock();
+       
        commpage_disable_timestamp();
 
        commpage_disable_timestamp();
 
-       now = mach_absolute_time();
+       secs_copy = secs;
+       microsecs_copy = microsecs;
 
 
-       delta = clock_calend.adjdelta;
+#if DEVELOPMENT || DEBUG
+       struct clock_calend clock_calend_cp1 = clock_calend;
+#endif /* DEVELOPMENT || DEBUG */
 
 
-       if (delta > 0) {
-               clock_calend.offset += clock_calend.adjoffset;
+#if DEVELOPMENT || DEBUG
+       last_utc_sec = secs;
+       last_utc_usec = microsecs;
+       last_sys_sec = sys;
+       last_sys_usec = microsys;
+       if (secs > max_utc_sec)
+               max_utc_sec = secs;
+#endif
+       /*
+        * We normally expect the UTC clock to be always-on and produce
+        * greater readings than the tick counter.  There may be corner cases
+        * due to differing clock resolutions (UTC clock is likely lower) and
+        * and errors reading the UTC clock (some implementations return 0
+        * on error) in which that doesn't hold true.  Bring the UTC measurements
+        * in-line with the tick counter measurements as a best effort in that case.
+        */
+       //FIXME if the current time is prior than 1970 secs will be negative
+       if ((sys > secs) || ((sys == secs) && (microsys > microsecs))) {
+               os_log(OS_LOG_DEFAULT, "%s WARNING: %s is less then sys %s %lu s %d u sys %lu s %d u\n",
+                       __func__, (has_monotonic_clock)?"monotonic":"PMU", (has_monotonic_clock)?"monotonic":"PMU", (unsigned long)secs, microsecs, (unsigned long)sys, microsys);
+               secs = sys;
+               microsecs = microsys;
+       }
 
 
-               calend_adjtotal -= delta;
-               if (delta > calend_adjtotal) {
-                       clock_calend.adjdelta = delta = (int32_t)calend_adjtotal;
+       // PMU or monotonic - sys
+       // This macro stores the subtraction result in secs and microsecs
+       TIME_SUB(secs, sys, microsecs, microsys, USEC_PER_SEC);
+       clock2bintime(&secs, &microsecs, &bt);
 
 
-                       nanoseconds_to_absolutetime((uint64_t)delta, &t64);
-                       clock_calend.adjoffset = (uint32_t)t64;
-               }
+       /*
+        * Safety belt: the UTC clock will likely have a lower resolution than the tick counter.
+        * It's also possible that the device didn't fully transition to the powered-off state on
+        * the most recent sleep, so the tick counter may not have reset or may have only briefly
+        * tured off.  In that case it's possible for the difference between the UTC clock and the
+        * tick counter to be less than the previously recorded value in clock.calend.basesleep.
+        * In that case simply record that we slept for 0 ticks.
+        */ 
+       if ((bt.sec > clock_calend.basesleep.sec) ||
+           ((bt.sec == clock_calend.basesleep.sec) && (bt.frac > clock_calend.basesleep.frac))) {
+
+               //last_sleep is the difference between current PMU or monotonic - abs and last wake PMU or monotonic - abs
+               last_sleep_bt = bt;
+               bintime_sub(&last_sleep_bt, &clock_calend.basesleep);
+
+               //set baseseep to current PMU or monotonic - abs
+               clock_calend.basesleep = bt;
+               bintime2usclock(&last_sleep_bt, &last_sleep_sec, &last_sleep_usec);
+               bintime2absolutetime(&last_sleep_bt, &mach_absolutetime_last_sleep);
+               mach_absolutetime_asleep += mach_absolutetime_last_sleep;
+
+               bintime_add(&clock_calend.offset, &last_sleep_bt);
+               bintime_add(&clock_calend.bintime, &last_sleep_bt);
+
+       } else{
+               mach_absolutetime_last_sleep = 0;
+               last_sleep_sec = last_sleep_usec = 0;
+               bintime2usclock(&clock_calend.basesleep, &basesleep_s, &basesleep_us);
+               os_log(OS_LOG_DEFAULT, "%s WARNING: basesleep (%lu s %d u)  > %s-sys (%lu s %d u) \n",
+                       __func__, (unsigned long) basesleep_s, basesleep_us, (has_monotonic_clock)?"monotonic":"PMU", (unsigned long) secs_copy, microsecs_copy );
        }
        }
-       else
-               if (delta < 0) {
-                       clock_calend.offset -= clock_calend.adjoffset;
 
 
-                       calend_adjtotal -= delta;
-                       if (delta < calend_adjtotal) {
-                               clock_calend.adjdelta = delta = (int32_t)calend_adjtotal;
+       KERNEL_DEBUG_CONSTANT(
+                 MACHDBG_CODE(DBG_MACH_CLOCK,MACH_EPOCH_CHANGE) | DBG_FUNC_NONE,
+                 (uintptr_t) mach_absolutetime_last_sleep,
+                 (uintptr_t) mach_absolutetime_asleep,
+                 (uintptr_t) (mach_absolutetime_last_sleep >> 32),
+                 (uintptr_t) (mach_absolutetime_asleep >> 32),
+                 0);
 
 
-                               nanoseconds_to_absolutetime((uint64_t)-delta, &t64);
-                               clock_calend.adjoffset = (uint32_t)t64;
-                       }
+       commpage_update_mach_continuous_time(mach_absolutetime_asleep);
+       adjust_cont_time_thread_calls();
 
 
-                       if (clock_calend.adjdelta != 0)
-                               clock_calend.adjstart = now;
-               }
+#if DEVELOPMENT || DEBUG
+       struct clock_calend clock_calend_cp = clock_calend;
+#endif
+
+       clock_unlock();
+       splx(s);
+
+#if DEVELOPMENT || DEBUG
+       if (g_should_log_clock_adjustments) {
+               os_log(OS_LOG_DEFAULT, "PMU was %lu s %d u\n",(unsigned long) utc_sec, utc_usec);
+               os_log(OS_LOG_DEFAULT, "last sleep was %lu s %d u\n",(unsigned long) last_sleep_sec, last_sleep_usec);
+               print_all_clock_variables("clock_wakeup_calendar:BEFORE",
+                                 &secs_copy, &microsecs_copy, &sys, &microsys, &clock_calend_cp1);
+               print_all_clock_variables("clock_wakeup_calendar:AFTER", NULL, NULL, NULL, NULL, &clock_calend_cp);
+       }
+#endif /* DEVELOPMENT || DEBUG */
 
 
-       if (clock_calend.adjdelta != 0)
-               interval = calend_adjinterval;
+       host_notify_calendar_change();
 
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
 
 #if CONFIG_DTRACE
        clock_track_calend_nowait();
 #endif
+}
+
 
 
-       return (interval);
+/*
+ *     clock_get_boottime_nanotime:
+ *
+ *     Return the boottime, used by sysctl.
+ */
+void
+clock_get_boottime_nanotime(
+       clock_sec_t                     *secs,
+       clock_nsec_t            *nanosecs)
+{
+       spl_t   s;
+
+       s = splclock();
+       clock_lock();
+
+       *secs = (clock_sec_t)clock_boottime;
+       *nanosecs = (clock_nsec_t)clock_boottime_usec * NSEC_PER_USEC;
+
+       clock_unlock();
+       splx(s);
 }
 
 /*
 }
 
 /*
- *     clock_wakeup_calendar:
+ *     clock_get_boottime_nanotime:
  *
  *
- *     Interface to power management, used
- *     to initiate the reset of the calendar
- *     on wake from sleep event.
+ *     Return the boottime, used by sysctl.
  */
 void
  */
 void
-clock_wakeup_calendar(void)
+clock_get_boottime_microtime(
+       clock_sec_t                     *secs,
+       clock_usec_t            *microsecs)
 {
 {
-       thread_call_enter(&calend_wakecall);
+       spl_t   s;
+
+       s = splclock();
+       clock_lock();
+
+       *secs = (clock_sec_t)clock_boottime;
+       *microsecs = (clock_nsec_t)clock_boottime_usec;
+
+       clock_unlock();
+       splx(s);
 }
 
 }
 
+
 /*
  *     Wait / delay routines.
  */
 /*
  *     Wait / delay routines.
  */
@@ -843,6 +1468,19 @@ _clock_delay_until_deadline(
        uint64_t                interval,
        uint64_t                deadline)
 {
        uint64_t                interval,
        uint64_t                deadline)
 {
+       _clock_delay_until_deadline_with_leeway(interval, deadline, 0);
+}
+
+/*
+ * Like _clock_delay_until_deadline, but it accepts a
+ * leeway value.
+ */
+void
+_clock_delay_until_deadline_with_leeway(
+       uint64_t                interval,
+       uint64_t                deadline,
+       uint64_t                leeway)
+{
 
        if (interval == 0)
                return;
 
        if (interval == 0)
                return;
@@ -852,13 +1490,21 @@ _clock_delay_until_deadline(
                        ml_get_interrupts_enabled() == FALSE    ) {
                machine_delay_until(interval, deadline);
        } else {
                        ml_get_interrupts_enabled() == FALSE    ) {
                machine_delay_until(interval, deadline);
        } else {
-               assert_wait_deadline((event_t)clock_delay_until, THREAD_UNINT, deadline);
+               /*
+                * For now, assume a leeway request of 0 means the client does not want a leeway
+                * value. We may want to change this interpretation in the future.
+                */
+
+               if (leeway) {
+                       assert_wait_deadline_with_leeway((event_t)clock_delay_until, THREAD_UNINT, TIMEOUT_URGENCY_LEEWAY, deadline, leeway);
+               } else {
+                       assert_wait_deadline((event_t)clock_delay_until, THREAD_UNINT, deadline);
+               }
 
                thread_block(THREAD_CONTINUE_NULL);
        }
 }
 
 
                thread_block(THREAD_CONTINUE_NULL);
        }
 }
 
-
 void
 delay_for_interval(
        uint32_t                interval,
 void
 delay_for_interval(
        uint32_t                interval,
@@ -871,6 +1517,21 @@ delay_for_interval(
        _clock_delay_until_deadline(abstime, mach_absolute_time() + abstime);
 }
 
        _clock_delay_until_deadline(abstime, mach_absolute_time() + abstime);
 }
 
+void
+delay_for_interval_with_leeway(
+       uint32_t                interval,
+       uint32_t                leeway,
+       uint32_t                scale_factor)
+{
+       uint64_t                abstime_interval;
+       uint64_t                abstime_leeway;
+
+       clock_interval_to_absolutetime_interval(interval, scale_factor, &abstime_interval);
+       clock_interval_to_absolutetime_interval(leeway, scale_factor, &abstime_leeway);
+
+       _clock_delay_until_deadline_with_leeway(abstime_interval, mach_absolute_time() + abstime_interval, abstime_leeway);
+}
+
 void
 delay(
        int             usec)
 void
 delay(
        int             usec)
@@ -902,6 +1563,14 @@ clock_absolutetime_interval_to_deadline(
        *result = mach_absolute_time() + abstime;
 }
 
        *result = mach_absolute_time() + abstime;
 }
 
+void
+clock_continuoustime_interval_to_deadline(
+       uint64_t                        conttime,
+       uint64_t                        *result)
+{
+       *result = mach_continuous_time() + conttime;
+}
+
 void
 clock_get_uptime(
        uint64_t        *result)
 void
 clock_get_uptime(
        uint64_t        *result)
@@ -928,6 +1597,61 @@ clock_deadline_for_periodic_event(
        }
 }
 
        }
 }
 
+uint64_t
+mach_continuous_time(void)
+{
+       while(1) {      
+               uint64_t read1 = mach_absolutetime_asleep;
+               uint64_t absolute = mach_absolute_time();
+               OSMemoryBarrier();
+               uint64_t read2 = mach_absolutetime_asleep;
+
+               if(__builtin_expect(read1 == read2, 1)) {
+                       return absolute + read1;
+               }
+       }
+}
+
+uint64_t
+mach_continuous_approximate_time(void)
+{
+       while(1) {
+               uint64_t read1 = mach_absolutetime_asleep;
+               uint64_t absolute = mach_approximate_time();
+               OSMemoryBarrier();
+               uint64_t read2 = mach_absolutetime_asleep;
+
+               if(__builtin_expect(read1 == read2, 1)) {
+                       return absolute + read1;
+               }
+       }
+}
+
+/*
+ * continuoustime_to_absolutetime
+ * Must be called with interrupts disabled
+ * Returned value is only valid until the next update to
+ * mach_continuous_time 
+ */
+uint64_t
+continuoustime_to_absolutetime(uint64_t conttime) {
+       if (conttime <= mach_absolutetime_asleep)
+               return 0;
+       else
+               return conttime - mach_absolutetime_asleep;
+}
+
+/*
+ * absolutetime_to_continuoustime
+ * Must be called with interrupts disabled
+ * Returned value is only valid until the next update to
+ * mach_continuous_time 
+ */
+uint64_t
+absolutetime_to_continuoustime(uint64_t abstime) {
+       return abstime + mach_absolutetime_asleep;
+}
+
 #if    CONFIG_DTRACE
 
 /*
 #if    CONFIG_DTRACE
 
 /*
@@ -950,6 +1674,7 @@ clock_get_calendar_nanotime_nowait(
        int i = 0;
        uint64_t                now;
        struct unlocked_clock_calend stable;
        int i = 0;
        uint64_t                now;
        struct unlocked_clock_calend stable;
+       struct bintime bt;
 
        for (;;) {
                stable = flipflop[i];           /* take snapshot */
 
        for (;;) {
                stable = flipflop[i];           /* take snapshot */
@@ -970,31 +1695,17 @@ clock_get_calendar_nanotime_nowait(
                if (flipflop[i].gen == stable.gen)
                        break;
 
                if (flipflop[i].gen == stable.gen)
                        break;
 
-               /* Switch to the oher element of the flipflop, and try again. */
+               /* Switch to the other element of the flipflop, and try again. */
                i ^= 1;
        }
 
        now = mach_absolute_time();
 
                i ^= 1;
        }
 
        now = mach_absolute_time();
 
-       if (stable.calend.adjdelta < 0) {
-               uint32_t        t32;
-
-               if (now > stable.calend.adjstart) {
-                       t32 = (uint32_t)(now - stable.calend.adjstart);
-
-                       if (t32 > stable.calend.adjoffset)
-                               now -= stable.calend.adjoffset;
-                       else
-                               now = stable.calend.adjstart;
-               }
-       }
-
-       now += stable.calend.offset;
+       bt = get_scaled_time(now);
 
 
-       absolutetime_to_microtime(now, secs, nanosecs);
-       *nanosecs *= NSEC_PER_USEC;
+       bintime_add(&bt, &clock_calend.bintime);
 
 
-       *secs += (clock_sec_t)stable.calend.epoch;
+       bintime2nsclock(&bt, secs, nanosecs);
 }
 
 static void 
 }
 
 static void