kernel - Refactor smp collision statistics

[dragonfly.git] / sys / kern / kern_clock.c

/*
 * Copyright (c) 2003,2004 The DragonFly Project.  All rights reserved.
 * 
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@backplane.com>
 * 
 * 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.
 * 3. Neither the name of The DragonFly Project 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 COPYRIGHT HOLDERS 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
 * COPYRIGHT HOLDERS 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.
 * 
 * Copyright (c) 1997, 1998 Poul-Henning Kamp <phk@FreeBSD.org>
 * Copyright (c) 1982, 1986, 1991, 1993
 *      The Regents of the University of California.  All rights reserved.
 * (c) UNIX System Laboratories, Inc.
 * All or some portions of this file are derived from material licensed
 * to the University of California by American Telephone and Telegraph
 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
 * the permission of UNIX System Laboratories, Inc.
 *
 * 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.
 * 3. 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.
 *
 *      @(#)kern_clock.c        8.5 (Berkeley) 1/21/94
 * $FreeBSD: src/sys/kern/kern_clock.c,v 1.105.2.10 2002/10/17 13:19:40 maxim Exp $
 */

#include "opt_ntp.h"
#include "opt_pctrack.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/callout.h>
#include <sys/kernel.h>
#include <sys/kinfo.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/resource.h>
#include <sys/resourcevar.h>
#include <sys/signalvar.h>
#include <sys/priv.h>
#include <sys/timex.h>
#include <sys/timepps.h>
#include <sys/upmap.h>
#include <sys/lock.h>
#include <sys/sysctl.h>
#include <sys/kcollect.h>

#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_extern.h>

#include <sys/thread2.h>
#include <sys/spinlock2.h>

#include <machine/cpu.h>
#include <machine/limits.h>
#include <machine/smp.h>
#include <machine/cpufunc.h>
#include <machine/specialreg.h>
#include <machine/clock.h>

#ifdef GPROF
#include <sys/gmon.h>
#endif

#ifdef DEBUG_PCTRACK
static void do_pctrack(struct intrframe *frame, int which);
#endif

static void initclocks (void *dummy);
SYSINIT(clocks, SI_BOOT2_CLOCKS, SI_ORDER_FIRST, initclocks, NULL);

/*
 * Some of these don't belong here, but it's easiest to concentrate them.
 * Note that cpu_time counts in microseconds, but most userland programs
 * just compare relative times against the total by delta.
 */
struct kinfo_cputime cputime_percpu[MAXCPU];
#ifdef DEBUG_PCTRACK
struct kinfo_pcheader cputime_pcheader = { PCTRACK_SIZE, PCTRACK_ARYSIZE };
struct kinfo_pctrack cputime_pctrack[MAXCPU][PCTRACK_SIZE];
#endif

static int sniff_enable = 1;
static int sniff_target = -1;
SYSCTL_INT(_kern, OID_AUTO, sniff_enable, CTLFLAG_RW, &sniff_enable, 0 , "");
SYSCTL_INT(_kern, OID_AUTO, sniff_target, CTLFLAG_RW, &sniff_target, 0 , "");

static int
sysctl_cputime(SYSCTL_HANDLER_ARGS)
{
        int cpu, error = 0;
        int root_error;
        size_t size = sizeof(struct kinfo_cputime);
        struct kinfo_cputime tmp;

        /*
         * NOTE: For security reasons, only root can sniff %rip
         */
        root_error = priv_check_cred(curthread->td_ucred, PRIV_ROOT, 0);

        for (cpu = 0; cpu < ncpus; ++cpu) {
                tmp = cputime_percpu[cpu];
                if (root_error == 0) {
                        tmp.cp_sample_pc =
                                (int64_t)globaldata_find(cpu)->gd_sample_pc;
                        tmp.cp_sample_sp =
                                (int64_t)globaldata_find(cpu)->gd_sample_sp;
                }
                if ((error = SYSCTL_OUT(req, &tmp, size)) != 0)
                        break;
        }

        if (root_error == 0) {
                if (sniff_enable) {
                        int n = sniff_target;
                        if (n < 0)
                                smp_sniff();
                        else if (n < ncpus)
                                cpu_sniff(n);
                }
        }

        return (error);
}
SYSCTL_PROC(_kern, OID_AUTO, cputime, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
        sysctl_cputime, "S,kinfo_cputime", "CPU time statistics");

static int
sysctl_cp_time(SYSCTL_HANDLER_ARGS)
{
        long cpu_states[CPUSTATES] = {0};
        int cpu, error = 0;
        size_t size = sizeof(cpu_states);

        for (cpu = 0; cpu < ncpus; ++cpu) {
                cpu_states[CP_USER] += cputime_percpu[cpu].cp_user;
                cpu_states[CP_NICE] += cputime_percpu[cpu].cp_nice;
                cpu_states[CP_SYS] += cputime_percpu[cpu].cp_sys;
                cpu_states[CP_INTR] += cputime_percpu[cpu].cp_intr;
                cpu_states[CP_IDLE] += cputime_percpu[cpu].cp_idle;
        }

        error = SYSCTL_OUT(req, cpu_states, size);

        return (error);
}

SYSCTL_PROC(_kern, OID_AUTO, cp_time, (CTLTYPE_LONG|CTLFLAG_RD), 0, 0,
    sysctl_cp_time, "LU", "CPU time statistics");

static int
sysctl_cp_times(SYSCTL_HANDLER_ARGS)
{
        long cpu_states[CPUSTATES] = {0};
        int cpu, error;
        size_t size = sizeof(cpu_states);

        for (error = 0, cpu = 0; error == 0 && cpu < ncpus; ++cpu) {
                cpu_states[CP_USER] = cputime_percpu[cpu].cp_user;
                cpu_states[CP_NICE] = cputime_percpu[cpu].cp_nice;
                cpu_states[CP_SYS] = cputime_percpu[cpu].cp_sys;
                cpu_states[CP_INTR] = cputime_percpu[cpu].cp_intr;
                cpu_states[CP_IDLE] = cputime_percpu[cpu].cp_idle;
                error = SYSCTL_OUT(req, cpu_states, size);
        }

        return (error);
}

SYSCTL_PROC(_kern, OID_AUTO, cp_times, (CTLTYPE_LONG|CTLFLAG_RD), 0, 0,
    sysctl_cp_times, "LU", "per-CPU time statistics");

/*
 * boottime is used to calculate the 'real' uptime.  Do not confuse this with
 * microuptime().  microtime() is not drift compensated.  The real uptime
 * with compensation is nanotime() - bootime.  boottime is recalculated
 * whenever the real time is set based on the compensated elapsed time
 * in seconds (gd->gd_time_seconds).
 *
 * The gd_time_seconds and gd_cpuclock_base fields remain fairly monotonic.
 * Slight adjustments to gd_cpuclock_base are made to phase-lock it to
 * the real time.
 *
 * WARNING! time_second can backstep on time corrections. Also, unlike
 *          time_second, time_uptime is not a "real" time_t (seconds
 *          since the Epoch) but seconds since booting.
 */
struct timespec boottime;       /* boot time (realtime) for reference only */
time_t time_second;             /* read-only 'passive' realtime in seconds */
time_t time_uptime;             /* read-only 'passive' uptime in seconds */

/*
 * basetime is used to calculate the compensated real time of day.  The
 * basetime can be modified on a per-tick basis by the adjtime(), 
 * ntp_adjtime(), and sysctl-based time correction APIs.
 *
 * Note that frequency corrections can also be made by adjusting
 * gd_cpuclock_base.
 *
 * basetime is a tail-chasing FIFO, updated only by cpu #0.  The FIFO is
 * used on both SMP and UP systems to avoid MP races between cpu's and
 * interrupt races on UP systems.
 */
struct hardtime {
        __uint32_t time_second;
        sysclock_t cpuclock_base;
};

#define BASETIME_ARYSIZE        16
#define BASETIME_ARYMASK        (BASETIME_ARYSIZE - 1)
static struct timespec basetime[BASETIME_ARYSIZE];
static struct hardtime hardtime[BASETIME_ARYSIZE];
static volatile int basetime_index;

static int
sysctl_get_basetime(SYSCTL_HANDLER_ARGS)
{
        struct timespec *bt;
        int error;
        int index;

        /*
         * Because basetime data and index may be updated by another cpu,
         * a load fence is required to ensure that the data we read has
         * not been speculatively read relative to a possibly updated index.
         */
        index = basetime_index;
        cpu_lfence();
        bt = &basetime[index];
        error = SYSCTL_OUT(req, bt, sizeof(*bt));
        return (error);
}

SYSCTL_STRUCT(_kern, KERN_BOOTTIME, boottime, CTLFLAG_RD,
    &boottime, timespec, "System boottime");
SYSCTL_PROC(_kern, OID_AUTO, basetime, CTLTYPE_STRUCT|CTLFLAG_RD, 0, 0,
    sysctl_get_basetime, "S,timespec", "System basetime");

static void hardclock(systimer_t info, int, struct intrframe *frame);
static void statclock(systimer_t info, int, struct intrframe *frame);
static void schedclock(systimer_t info, int, struct intrframe *frame);
static void getnanotime_nbt(struct timespec *nbt, struct timespec *tsp);

int     ticks;                  /* system master ticks at hz */
int     clocks_running;         /* tsleep/timeout clocks operational */
int64_t nsec_adj;               /* ntpd per-tick adjustment in nsec << 32 */
int64_t nsec_acc;               /* accumulator */
int     sched_ticks;            /* global schedule clock ticks */

/* NTPD time correction fields */
int64_t ntp_tick_permanent;     /* per-tick adjustment in nsec << 32 */
int64_t ntp_tick_acc;           /* accumulator for per-tick adjustment */
int64_t ntp_delta;              /* one-time correction in nsec */
int64_t ntp_big_delta = 1000000000;
int32_t ntp_tick_delta;         /* current adjustment rate */
int32_t ntp_default_tick_delta; /* adjustment rate for ntp_delta */
time_t  ntp_leap_second;        /* time of next leap second */
int     ntp_leap_insert;        /* whether to insert or remove a second */
struct spinlock ntp_spin;

/*
 * Finish initializing clock frequencies and start all clocks running.
 */
/* ARGSUSED*/
static void
initclocks(void *dummy)
{
        /*psratio = profhz / stathz;*/
        spin_init(&ntp_spin, "ntp");
        initclocks_pcpu();
        clocks_running = 1;
        if (kpmap) {
            kpmap->tsc_freq = tsc_frequency;
            kpmap->tick_freq = hz;
        }
}

/*
 * Called on a per-cpu basis from the idle thread bootstrap on each cpu
 * during SMP initialization.
 *
 * This routine is called concurrently during low-level SMP initialization
 * and may not block in any way.  Meaning, among other things, we can't
 * acquire any tokens.
 */
void
initclocks_pcpu(void)
{
        struct globaldata *gd = mycpu;

        crit_enter();
        if (gd->gd_cpuid == 0) {
            gd->gd_time_seconds = 1;
            gd->gd_cpuclock_base = sys_cputimer->count();
            hardtime[0].time_second = gd->gd_time_seconds;
            hardtime[0].cpuclock_base = gd->gd_cpuclock_base;
        } else {
            gd->gd_time_seconds = globaldata_find(0)->gd_time_seconds;
            gd->gd_cpuclock_base = globaldata_find(0)->gd_cpuclock_base;
        }

        systimer_intr_enable();

        crit_exit();
}

/*
 * Called on a 10-second interval after the system is operational.
 * Return the collection data for USERPCT and install the data for
 * SYSTPCT and IDLEPCT.
 */
static
uint64_t
collect_cputime_callback(int n)
{
        static long cpu_base[CPUSTATES];
        long cpu_states[CPUSTATES];
        long total;
        long acc;
        long lsb;

        bzero(cpu_states, sizeof(cpu_states));
        for (n = 0; n < ncpus; ++n) {
                cpu_states[CP_USER] += cputime_percpu[n].cp_user;
                cpu_states[CP_NICE] += cputime_percpu[n].cp_nice;
                cpu_states[CP_SYS] += cputime_percpu[n].cp_sys;
                cpu_states[CP_INTR] += cputime_percpu[n].cp_intr;
                cpu_states[CP_IDLE] += cputime_percpu[n].cp_idle;
        }

        acc = 0;
        for (n = 0; n < CPUSTATES; ++n) {
                total = cpu_states[n] - cpu_base[n];
                cpu_base[n] = cpu_states[n];
                cpu_states[n] = total;
                acc += total;
        }
        if (acc == 0)           /* prevent degenerate divide by 0 */
                acc = 1;
        lsb = acc / (10000 * 2);
        kcollect_setvalue(KCOLLECT_SYSTPCT,
                          (cpu_states[CP_SYS] + lsb) * 10000 / acc);
        kcollect_setvalue(KCOLLECT_IDLEPCT,
                          (cpu_states[CP_IDLE] + lsb) * 10000 / acc);
        kcollect_setvalue(KCOLLECT_INTRPCT,
                          (cpu_states[CP_INTR] + lsb) * 10000 / acc);
        return((cpu_states[CP_USER] + cpu_states[CP_NICE] + lsb) * 10000 / acc);
}

/*
 * This routine is called on just the BSP, just after SMP initialization
 * completes to * finish initializing any clocks that might contend/block
 * (e.g. like on a token).  We can't do this in initclocks_pcpu() because
 * that function is called from the idle thread bootstrap for each cpu and
 * not allowed to block at all.
 */
static
void
initclocks_other(void *dummy)
{
        struct globaldata *ogd = mycpu;
        struct globaldata *gd;
        int n;

        for (n = 0; n < ncpus; ++n) {
                lwkt_setcpu_self(globaldata_find(n));
                gd = mycpu;

                /*
                 * Use a non-queued periodic systimer to prevent multiple
                 * ticks from building up if the sysclock jumps forward
                 * (8254 gets reset).  The sysclock will never jump backwards.
                 * Our time sync is based on the actual sysclock, not the
                 * ticks count.
                 *
                 * Install statclock before hardclock to prevent statclock
                 * from misinterpreting gd_flags for tick assignment when
                 * they overlap.
                 */
                systimer_init_periodic_flags(&gd->gd_statclock, statclock,
                                          NULL, stathz,
                                          SYSTF_MSSYNC | SYSTF_FIRST);
                systimer_init_periodic_flags(&gd->gd_hardclock, hardclock,
                                          NULL, hz, SYSTF_MSSYNC);
                /* XXX correct the frequency for scheduler / estcpu tests */
                systimer_init_periodic_flags(&gd->gd_schedclock, schedclock,
                                          NULL, ESTCPUFREQ, SYSTF_MSSYNC);
        }
        lwkt_setcpu_self(ogd);

        /*
         * Regular data collection
         */
        kcollect_register(KCOLLECT_USERPCT, "user", collect_cputime_callback,
                          KCOLLECT_SCALE(KCOLLECT_USERPCT_FORMAT, 0));
        kcollect_register(KCOLLECT_SYSTPCT, "syst", NULL,
                          KCOLLECT_SCALE(KCOLLECT_SYSTPCT_FORMAT, 0));
        kcollect_register(KCOLLECT_IDLEPCT, "idle", NULL,
                          KCOLLECT_SCALE(KCOLLECT_IDLEPCT_FORMAT, 0));
}
SYSINIT(clocks2, SI_BOOT2_POST_SMP, SI_ORDER_ANY, initclocks_other, NULL);

/*
 * This sets the current real time of day.  Timespecs are in seconds and
 * nanoseconds.  We do not mess with gd_time_seconds and gd_cpuclock_base,
 * instead we adjust basetime so basetime + gd_* results in the current
 * time of day.  This way the gd_* fields are guaranteed to represent
 * a monotonically increasing 'uptime' value.
 *
 * When set_timeofday() is called from userland, the system call forces it
 * onto cpu #0 since only cpu #0 can update basetime_index.
 */
void
set_timeofday(struct timespec *ts)
{
        struct timespec *nbt;
        int ni;

        /*
         * XXX SMP / non-atomic basetime updates
         */
        crit_enter();
        ni = (basetime_index + 1) & BASETIME_ARYMASK;
        cpu_lfence();
        nbt = &basetime[ni];
        nanouptime(nbt);
        nbt->tv_sec = ts->tv_sec - nbt->tv_sec;
        nbt->tv_nsec = ts->tv_nsec - nbt->tv_nsec;
        if (nbt->tv_nsec < 0) {
            nbt->tv_nsec += 1000000000;
            --nbt->tv_sec;
        }

        /*
         * Note that basetime diverges from boottime as the clock drift is
         * compensated for, so we cannot do away with boottime.  When setting
         * the absolute time of day the drift is 0 (for an instant) and we
         * can simply assign boottime to basetime.  
         *
         * Note that nanouptime() is based on gd_time_seconds which is drift
         * compensated up to a point (it is guaranteed to remain monotonically
         * increasing).  gd_time_seconds is thus our best uptime guess and
         * suitable for use in the boottime calculation.  It is already taken
         * into account in the basetime calculation above.
         */
        spin_lock(&ntp_spin);
        boottime.tv_sec = nbt->tv_sec;
        ntp_delta = 0;

        /*
         * We now have a new basetime, make sure all other cpus have it,
         * then update the index.
         */
        cpu_sfence();
        basetime_index = ni;
        spin_unlock(&ntp_spin);

        crit_exit();
}
        
/*
 * Each cpu has its own hardclock, but we only increments ticks and softticks
 * on cpu #0.
 *
 * NOTE! systimer! the MP lock might not be held here.  We can only safely
 * manipulate objects owned by the current cpu.
 */
static void
hardclock(systimer_t info, int in_ipi, struct intrframe *frame)
{
        sysclock_t cputicks;
        struct proc *p;
        struct globaldata *gd = mycpu;

        if ((gd->gd_reqflags & RQF_IPIQ) == 0 && lwkt_need_ipiq_process(gd)) {
                /* Defer to doreti on passive IPIQ processing */
                need_ipiq();
        }

        /*
         * We update the compensation base to calculate fine-grained time
         * from the sys_cputimer on a per-cpu basis in order to avoid
         * having to mess around with locks.  sys_cputimer is assumed to
         * be consistent across all cpus.  CPU N copies the base state from
         * CPU 0 using the same FIFO trick that we use for basetime (so we
         * don't catch a CPU 0 update in the middle).
         *
         * Note that we never allow info->time (aka gd->gd_hardclock.time)
         * to reverse index gd_cpuclock_base, but that it is possible for
         * it to temporarily get behind in the seconds if something in the
         * system locks interrupts for a long period of time.  Since periodic
         * timers count events, though everything should resynch again
         * immediately.
         */
        if (gd->gd_cpuid == 0) {
                int ni;

                cputicks = info->time - gd->gd_cpuclock_base;
                if (cputicks >= sys_cputimer->freq) {
                        cputicks /= sys_cputimer->freq;
                        if (cputicks != 0 && cputicks != 1)
                                kprintf("Warning: hardclock missed > 1 sec\n");
                        gd->gd_time_seconds += cputicks;
                        gd->gd_cpuclock_base += sys_cputimer->freq * cputicks;
                        /* uncorrected monotonic 1-sec gran */
                        time_uptime += cputicks;
                }
                ni = (basetime_index + 1) & BASETIME_ARYMASK;
                hardtime[ni].time_second = gd->gd_time_seconds;
                hardtime[ni].cpuclock_base = gd->gd_cpuclock_base;
        } else {
                int ni;

                ni = basetime_index;
                cpu_lfence();
                gd->gd_time_seconds = hardtime[ni].time_second;
                gd->gd_cpuclock_base = hardtime[ni].cpuclock_base;
        }

        /*
         * The system-wide ticks counter and NTP related timedelta/tickdelta
         * adjustments only occur on cpu #0.  NTP adjustments are accomplished
         * by updating basetime.
         */
        if (gd->gd_cpuid == 0) {
            struct timespec *nbt;
            struct timespec nts;
            int leap;
            int ni;

            ++ticks;

#if 0
            if (tco->tc_poll_pps) 
                tco->tc_poll_pps(tco);
#endif

            /*
             * Calculate the new basetime index.  We are in a critical section
             * on cpu #0 and can safely play with basetime_index.  Start
             * with the current basetime and then make adjustments.
             */
            ni = (basetime_index + 1) & BASETIME_ARYMASK;
            nbt = &basetime[ni];
            *nbt = basetime[basetime_index];

            /*
             * ntp adjustments only occur on cpu 0 and are protected by
             * ntp_spin.  This spinlock virtually never conflicts.
             */
            spin_lock(&ntp_spin);

            /*
             * Apply adjtime corrections.  (adjtime() API)
             *
             * adjtime() only runs on cpu #0 so our critical section is
             * sufficient to access these variables.
             */
            if (ntp_delta != 0) {
                nbt->tv_nsec += ntp_tick_delta;
                ntp_delta -= ntp_tick_delta;
                if ((ntp_delta > 0 && ntp_delta < ntp_tick_delta) ||
                    (ntp_delta < 0 && ntp_delta > ntp_tick_delta)) {
                        ntp_tick_delta = ntp_delta;
                }
            }

            /*
             * Apply permanent frequency corrections.  (sysctl API)
             */
            if (ntp_tick_permanent != 0) {
                ntp_tick_acc += ntp_tick_permanent;
                if (ntp_tick_acc >= (1LL << 32)) {
                    nbt->tv_nsec += ntp_tick_acc >> 32;
                    ntp_tick_acc -= (ntp_tick_acc >> 32) << 32;
                } else if (ntp_tick_acc <= -(1LL << 32)) {
                    /* Negate ntp_tick_acc to avoid shifting the sign bit. */
                    nbt->tv_nsec -= (-ntp_tick_acc) >> 32;
                    ntp_tick_acc += ((-ntp_tick_acc) >> 32) << 32;
                }
            }

            if (nbt->tv_nsec >= 1000000000) {
                    nbt->tv_sec++;
                    nbt->tv_nsec -= 1000000000;
            } else if (nbt->tv_nsec < 0) {
                    nbt->tv_sec--;
                    nbt->tv_nsec += 1000000000;
            }

            /*
             * Another per-tick compensation.  (for ntp_adjtime() API)
             */
            if (nsec_adj != 0) {
                nsec_acc += nsec_adj;
                if (nsec_acc >= 0x100000000LL) {
                    nbt->tv_nsec += nsec_acc >> 32;
                    nsec_acc = (nsec_acc & 0xFFFFFFFFLL);
                } else if (nsec_acc <= -0x100000000LL) {
                    nbt->tv_nsec -= -nsec_acc >> 32;
                    nsec_acc = -(-nsec_acc & 0xFFFFFFFFLL);
                }
                if (nbt->tv_nsec >= 1000000000) {
                    nbt->tv_nsec -= 1000000000;
                    ++nbt->tv_sec;
                } else if (nbt->tv_nsec < 0) {
                    nbt->tv_nsec += 1000000000;
                    --nbt->tv_sec;
                }
            }
            spin_unlock(&ntp_spin);

            /************************************************************
             *                  LEAP SECOND CORRECTION                  *
             ************************************************************
             *
             * Taking into account all the corrections made above, figure
             * out the new real time.  If the seconds field has changed
             * then apply any pending leap-second corrections.
             */
            getnanotime_nbt(nbt, &nts);

            if (time_second != nts.tv_sec) {
                /*
                 * Apply leap second (sysctl API).  Adjust nts for changes
                 * so we do not have to call getnanotime_nbt again.
                 */
                if (ntp_leap_second) {
                    if (ntp_leap_second == nts.tv_sec) {
                        if (ntp_leap_insert) {
                            nbt->tv_sec++;
                            nts.tv_sec++;
                        } else {
                            nbt->tv_sec--;
                            nts.tv_sec--;
                        }
                        ntp_leap_second--;
                    }
                }

                /*
                 * Apply leap second (ntp_adjtime() API), calculate a new
                 * nsec_adj field.  ntp_update_second() returns nsec_adj
                 * as a per-second value but we need it as a per-tick value.
                 */
                leap = ntp_update_second(time_second, &nsec_adj);
                nsec_adj /= hz;
                nbt->tv_sec += leap;
                nts.tv_sec += leap;

                /*
                 * Update the time_second 'approximate time' global.
                 */
                time_second = nts.tv_sec;
            }

            /*
             * Finally, our new basetime is ready to go live!
             */
            cpu_sfence();
            basetime_index = ni;

            /*
             * Update kpmap on each tick.  TS updates are integrated with
             * fences and upticks allowing userland to read the data
             * deterministically.
             */
            if (kpmap) {
                int w;

                w = (kpmap->upticks + 1) & 1;
                getnanouptime(&kpmap->ts_uptime[w]);
                getnanotime(&kpmap->ts_realtime[w]);
                cpu_sfence();
                ++kpmap->upticks;
                cpu_sfence();
            }
        }

        /*
         * lwkt thread scheduler fair queueing
         */
        lwkt_schedulerclock(curthread);

        /*
         * softticks are handled for all cpus
         */
        hardclock_softtick(gd);

        /*
         * Rollup accumulated vmstats, copy-back for critical path checks.
         */
        vmstats_rollup_cpu(gd);
        mycpu->gd_vmstats = vmstats;

        /*
         * ITimer handling is per-tick, per-cpu.
         *
         * We must acquire the per-process token in order for ksignal()
         * to be non-blocking.  For the moment this requires an AST fault,
         * the ksignal() cannot be safely issued from this hard interrupt.
         *
         * XXX Even the trytoken here isn't right, and itimer operation in
         *     a multi threaded environment is going to be weird at the
         *     very least.
         */
        if ((p = curproc) != NULL && lwkt_trytoken(&p->p_token)) {
                crit_enter_hard();
                if (p->p_upmap)
                        ++p->p_upmap->runticks;

                if (frame && CLKF_USERMODE(frame) &&
                    timevalisset(&p->p_timer[ITIMER_VIRTUAL].it_value) &&
                    itimerdecr(&p->p_timer[ITIMER_VIRTUAL], ustick) == 0) {
                        p->p_flags |= P_SIGVTALRM;
                        need_user_resched();
                }
                if (timevalisset(&p->p_timer[ITIMER_PROF].it_value) &&
                    itimerdecr(&p->p_timer[ITIMER_PROF], ustick) == 0) {
                        p->p_flags |= P_SIGPROF;
                        need_user_resched();
                }
                crit_exit_hard();
                lwkt_reltoken(&p->p_token);
        }
        setdelayed();
}

/*
 * The statistics clock typically runs at a 125Hz rate, and is intended
 * to be frequency offset from the hardclock (typ 100Hz).  It is per-cpu.
 *
 * NOTE! systimer! the MP lock might not be held here.  We can only safely
 * manipulate objects owned by the current cpu.
 *
 * The stats clock is responsible for grabbing a profiling sample.
 * Most of the statistics are only used by user-level statistics programs.
 * The main exceptions are p->p_uticks, p->p_sticks, p->p_iticks, and
 * p->p_estcpu.
 *
 * Like the other clocks, the stat clock is called from what is effectively
 * a fast interrupt, so the context should be the thread/process that got
 * interrupted.
 */
static void
statclock(systimer_t info, int in_ipi, struct intrframe *frame)
{
#ifdef GPROF
        struct gmonparam *g;
        int i;
#endif
        globaldata_t gd = mycpu;
        thread_t td;
        struct proc *p;
        int bump;
        sysclock_t cv;
        sysclock_t scv;

        /*
         * How big was our timeslice relative to the last time?  Calculate
         * in microseconds.
         *
         * NOTE: Use of microuptime() is typically MPSAFE, but usually not
         *       during early boot.  Just use the systimer count to be nice
         *       to e.g. qemu.  The systimer has a better chance of being
         *       MPSAFE at early boot.
         */
        cv = sys_cputimer->count();
        scv = gd->statint.gd_statcv;
        if (scv == 0) {
                bump = 1;
        } else {
                bump = (sys_cputimer->freq64_usec * (cv - scv)) >> 32;
                if (bump < 0)
                        bump = 0;
                if (bump > 1000000)
                        bump = 1000000;
        }
        gd->statint.gd_statcv = cv;

#if 0
        stv = &gd->gd_stattv;
        if (stv->tv_sec == 0) {
            bump = 1;
        } else {
            bump = tv.tv_usec - stv->tv_usec +
                (tv.tv_sec - stv->tv_sec) * 1000000;
            if (bump < 0)
                bump = 0;
            if (bump > 1000000)
                bump = 1000000;
        }
        *stv = tv;
#endif

        td = curthread;
        p = td->td_proc;

        if (frame && CLKF_USERMODE(frame)) {
                /*
                 * Came from userland, handle user time and deal with
                 * possible process.
                 */
                if (p && (p->p_flags & P_PROFIL))
                        addupc_intr(p, CLKF_PC(frame), 1);
                td->td_uticks += bump;

                /*
                 * Charge the time as appropriate
                 */
                if (p && p->p_nice > NZERO)
                        cpu_time.cp_nice += bump;
                else
                        cpu_time.cp_user += bump;
        } else {
                int intr_nest = gd->gd_intr_nesting_level;

                if (in_ipi) {
                        /*
                         * IPI processing code will bump gd_intr_nesting_level
                         * up by one, which breaks following CLKF_INTR testing,
                         * so we subtract it by one here.
                         */
                        --intr_nest;
                }
#ifdef GPROF
                /*
                 * Kernel statistics are just like addupc_intr, only easier.
                 */
                g = &_gmonparam;
                if (g->state == GMON_PROF_ON && frame) {
                        i = CLKF_PC(frame) - g->lowpc;
                        if (i < g->textsize) {
                                i /= HISTFRACTION * sizeof(*g->kcount);
                                g->kcount[i]++;
                        }
                }
#endif

#define IS_INTR_RUNNING ((frame && CLKF_INTR(intr_nest)) || CLKF_INTR_TD(td))

                /*
                 * Came from kernel mode, so we were:
                 * - handling an interrupt,
                 * - doing syscall or trap work on behalf of the current
                 *   user process, or
                 * - spinning in the idle loop.
                 * Whichever it is, charge the time as appropriate.
                 * Note that we charge interrupts to the current process,
                 * regardless of whether they are ``for'' that process,
                 * so that we know how much of its real time was spent
                 * in ``non-process'' (i.e., interrupt) work.
                 *
                 * XXX assume system if frame is NULL.  A NULL frame 
                 * can occur if ipi processing is done from a crit_exit().
                 */
                if (IS_INTR_RUNNING ||
                    (gd->gd_reqflags & RQF_INTPEND)) {
                        /*
                         * If we interrupted an interrupt thread, well,
                         * count it as interrupt time.
                         */
                        td->td_iticks += bump;
#ifdef DEBUG_PCTRACK
                        if (frame)
                                do_pctrack(frame, PCTRACK_INT);
#endif
                        cpu_time.cp_intr += bump;
                } else if (gd->gd_flags & GDF_VIRTUSER) {
                        /*
                         * The vkernel doesn't do a good job providing trap
                         * frames that we can test.  If the GDF_VIRTUSER
                         * flag is set we probably interrupted user mode.
                         *
                         * We also use this flag on the host when entering
                         * VMM mode.
                         */
                        td->td_uticks += bump;

                        /*
                         * Charge the time as appropriate
                         */
                        if (p && p->p_nice > NZERO)
                                cpu_time.cp_nice += bump;
                        else
                                cpu_time.cp_user += bump;
                } else {
                        td->td_sticks += bump;
                        if (td == &gd->gd_idlethread) {
                                /*
                                 * We want to count token contention as
                                 * system time.  When token contention occurs
                                 * the cpu may only be outside its critical
                                 * section while switching through the idle
                                 * thread.  In this situation, various flags
                                 * will be set in gd_reqflags.
                                 */
                                if (gd->gd_reqflags & RQF_IDLECHECK_WK_MASK)
                                        cpu_time.cp_sys += bump;
                                else
                                        cpu_time.cp_idle += bump;
                        } else {
                                /*
                                 * System thread was running.
                                 */
#ifdef DEBUG_PCTRACK
                                if (frame)
                                        do_pctrack(frame, PCTRACK_SYS);
#endif
                                cpu_time.cp_sys += bump;
                        }
                }

#undef IS_INTR_RUNNING
        }
}

#ifdef DEBUG_PCTRACK
/*
 * Sample the PC when in the kernel or in an interrupt.  User code can
 * retrieve the information and generate a histogram or other output.
 */

static void
do_pctrack(struct intrframe *frame, int which)
{
        struct kinfo_pctrack *pctrack;

        pctrack = &cputime_pctrack[mycpu->gd_cpuid][which];
        pctrack->pc_array[pctrack->pc_index & PCTRACK_ARYMASK] = 
                (void *)CLKF_PC(frame);
        ++pctrack->pc_index;
}

static int
sysctl_pctrack(SYSCTL_HANDLER_ARGS)
{
        struct kinfo_pcheader head;
        int error;
        int cpu;
        int ntrack;

        head.pc_ntrack = PCTRACK_SIZE;
        head.pc_arysize = PCTRACK_ARYSIZE;

        if ((error = SYSCTL_OUT(req, &head, sizeof(head))) != 0)
                return (error);

        for (cpu = 0; cpu < ncpus; ++cpu) {
                for (ntrack = 0; ntrack < PCTRACK_SIZE; ++ntrack) {
                        error = SYSCTL_OUT(req, &cputime_pctrack[cpu][ntrack],
                                           sizeof(struct kinfo_pctrack));
                        if (error)
                                break;
                }
                if (error)
                        break;
        }
        return (error);
}
SYSCTL_PROC(_kern, OID_AUTO, pctrack, (CTLTYPE_OPAQUE|CTLFLAG_RD), 0, 0,
        sysctl_pctrack, "S,kinfo_pcheader", "CPU PC tracking");

#endif

/*
 * The scheduler clock typically runs at a 50Hz rate.  NOTE! systimer,
 * the MP lock might not be held.  We can safely manipulate parts of curproc
 * but that's about it.
 *
 * Each cpu has its own scheduler clock.
 */
static void
schedclock(systimer_t info, int in_ipi __unused, struct intrframe *frame)
{
        struct lwp *lp;
        struct rusage *ru;
        struct vmspace *vm;
        long rss;

        if ((lp = lwkt_preempted_proc()) != NULL) {
                /*
                 * Account for cpu time used and hit the scheduler.  Note
                 * that this call MUST BE MP SAFE, and the BGL IS NOT HELD
                 * HERE.
                 */
                ++lp->lwp_cpticks;
                usched_schedulerclock(lp, info->periodic, info->time);
        } else {
                usched_schedulerclock(NULL, info->periodic, info->time);
        }
        if ((lp = curthread->td_lwp) != NULL) {
                /*
                 * Update resource usage integrals and maximums.
                 */
                if ((ru = &lp->lwp_proc->p_ru) &&
                    (vm = lp->lwp_proc->p_vmspace) != NULL) {
                        ru->ru_ixrss += pgtok(vm->vm_tsize);
                        ru->ru_idrss += pgtok(vm->vm_dsize);
                        ru->ru_isrss += pgtok(vm->vm_ssize);
                        if (lwkt_trytoken(&vm->vm_map.token)) {
                                rss = pgtok(vmspace_resident_count(vm));
                                if (ru->ru_maxrss < rss)
                                        ru->ru_maxrss = rss;
                                lwkt_reltoken(&vm->vm_map.token);
                        }
                }
        }
        /* Increment the global sched_ticks */
        if (mycpu->gd_cpuid == 0)
                ++sched_ticks;
}

/*
 * Compute number of ticks for the specified amount of time.  The 
 * return value is intended to be used in a clock interrupt timed
 * operation and guaranteed to meet or exceed the requested time.
 * If the representation overflows, return INT_MAX.  The minimum return
 * value is 1 ticks and the function will average the calculation up.
 * If any value greater then 0 microseconds is supplied, a value
 * of at least 2 will be returned to ensure that a near-term clock
 * interrupt does not cause the timeout to occur (degenerately) early.
 *
 * Note that limit checks must take into account microseconds, which is
 * done simply by using the smaller signed long maximum instead of
 * the unsigned long maximum.
 *
 * If ints have 32 bits, then the maximum value for any timeout in
 * 10ms ticks is 248 days.
 */
int
tvtohz_high(struct timeval *tv)
{
        int ticks;
        long sec, usec;

        sec = tv->tv_sec;
        usec = tv->tv_usec;
        if (usec < 0) {
                sec--;
                usec += 1000000;
        }
        if (sec < 0) {
#ifdef DIAGNOSTIC
                if (usec > 0) {
                        sec++;
                        usec -= 1000000;
                }
                kprintf("tvtohz_high: negative time difference "
                        "%ld sec %ld usec\n",
                        sec, usec);
#endif
                ticks = 1;
        } else if (sec <= INT_MAX / hz) {
                ticks = (int)(sec * hz + 
                            ((u_long)usec + (ustick - 1)) / ustick) + 1;
        } else {
                ticks = INT_MAX;
        }
        return (ticks);
}

int
tstohz_high(struct timespec *ts)
{
        int ticks;
        long sec, nsec;

        sec = ts->tv_sec;
        nsec = ts->tv_nsec;
        if (nsec < 0) {
                sec--;
                nsec += 1000000000;
        }
        if (sec < 0) {
#ifdef DIAGNOSTIC
                if (nsec > 0) {
                        sec++;
                        nsec -= 1000000000;
                }
                kprintf("tstohz_high: negative time difference "
                        "%ld sec %ld nsec\n",
                        sec, nsec);
#endif
                ticks = 1;
        } else if (sec <= INT_MAX / hz) {
                ticks = (int)(sec * hz +
                            ((u_long)nsec + (nstick - 1)) / nstick) + 1;
        } else {
                ticks = INT_MAX;
        }
        return (ticks);
}


/*
 * Compute number of ticks for the specified amount of time, erroring on
 * the side of it being too low to ensure that sleeping the returned number
 * of ticks will not result in a late return.
 *
 * The supplied timeval may not be negative and should be normalized.  A
 * return value of 0 is possible if the timeval converts to less then
 * 1 tick.
 *
 * If ints have 32 bits, then the maximum value for any timeout in
 * 10ms ticks is 248 days.
 */
int
tvtohz_low(struct timeval *tv)
{
        int ticks;
        long sec;

        sec = tv->tv_sec;
        if (sec <= INT_MAX / hz)
                ticks = (int)(sec * hz + (u_long)tv->tv_usec / ustick);
        else
                ticks = INT_MAX;
        return (ticks);
}

int
tstohz_low(struct timespec *ts)
{
        int ticks;
        long sec;

        sec = ts->tv_sec;
        if (sec <= INT_MAX / hz)
                ticks = (int)(sec * hz + (u_long)ts->tv_nsec / nstick);
        else
                ticks = INT_MAX;
        return (ticks);
}

/*
 * Start profiling on a process.
 *
 * Caller must hold p->p_token();
 *
 * Kernel profiling passes proc0 which never exits and hence
 * keeps the profile clock running constantly.
 */
void
startprofclock(struct proc *p)
{
        if ((p->p_flags & P_PROFIL) == 0) {
                p->p_flags |= P_PROFIL;
#if 0   /* XXX */
                if (++profprocs == 1 && stathz != 0) {
                        crit_enter();
                        psdiv = psratio;
                        setstatclockrate(profhz);
                        crit_exit();
                }
#endif
        }
}

/*
 * Stop profiling on a process.
 *
 * caller must hold p->p_token
 */
void
stopprofclock(struct proc *p)
{
        if (p->p_flags & P_PROFIL) {
                p->p_flags &= ~P_PROFIL;
#if 0   /* XXX */
                if (--profprocs == 0 && stathz != 0) {
                        crit_enter();
                        psdiv = 1;
                        setstatclockrate(stathz);
                        crit_exit();
                }
#endif
        }
}

/*
 * Return information about system clocks.
 */
static int
sysctl_kern_clockrate(SYSCTL_HANDLER_ARGS)
{
        struct kinfo_clockinfo clkinfo;
        /*
         * Construct clockinfo structure.
         */
        clkinfo.ci_hz = hz;
        clkinfo.ci_tick = ustick;
        clkinfo.ci_tickadj = ntp_default_tick_delta / 1000;
        clkinfo.ci_profhz = profhz;
        clkinfo.ci_stathz = stathz ? stathz : hz;
        return (sysctl_handle_opaque(oidp, &clkinfo, sizeof clkinfo, req));
}

SYSCTL_PROC(_kern, KERN_CLOCKRATE, clockrate, CTLTYPE_STRUCT|CTLFLAG_RD,
        0, 0, sysctl_kern_clockrate, "S,clockinfo","");

/*
 * We have eight functions for looking at the clock, four for
 * microseconds and four for nanoseconds.  For each there is fast
 * but less precise version "get{nano|micro}[up]time" which will
 * return a time which is up to 1/HZ previous to the call, whereas
 * the raw version "{nano|micro}[up]time" will return a timestamp
 * which is as precise as possible.  The "up" variants return the
 * time relative to system boot, these are well suited for time
 * interval measurements.
 *
 * Each cpu independently maintains the current time of day, so all
 * we need to do to protect ourselves from changes is to do a loop
 * check on the seconds field changing out from under us.
 *
 * The system timer maintains a 32 bit count and due to various issues
 * it is possible for the calculated delta to occasionally exceed
 * sys_cputimer->freq.  If this occurs the sys_cputimer->freq64_nsec
 * multiplication can easily overflow, so we deal with the case.  For
 * uniformity we deal with the case in the usec case too.
 *
 * All the [get][micro,nano][time,uptime]() routines are MPSAFE.
 */
void
getmicrouptime(struct timeval *tvp)
{
        struct globaldata *gd = mycpu;
        sysclock_t delta;

        do {
                tvp->tv_sec = gd->gd_time_seconds;
                delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
        } while (tvp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tvp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
        if (tvp->tv_usec >= 1000000) {
                tvp->tv_usec -= 1000000;
                ++tvp->tv_sec;
        }
}

void
getnanouptime(struct timespec *tsp)
{
        struct globaldata *gd = mycpu;
        sysclock_t delta;

        do {
                tsp->tv_sec = gd->gd_time_seconds;
                delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
        } while (tsp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tsp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
}

void
microuptime(struct timeval *tvp)
{
        struct globaldata *gd = mycpu;
        sysclock_t delta;

        do {
                tvp->tv_sec = gd->gd_time_seconds;
                delta = sys_cputimer->count() - gd->gd_cpuclock_base;
        } while (tvp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tvp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;
}

void
nanouptime(struct timespec *tsp)
{
        struct globaldata *gd = mycpu;
        sysclock_t delta;

        do {
                tsp->tv_sec = gd->gd_time_seconds;
                delta = sys_cputimer->count() - gd->gd_cpuclock_base;
        } while (tsp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tsp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
}

/*
 * realtime routines
 */
void
getmicrotime(struct timeval *tvp)
{
        struct globaldata *gd = mycpu;
        struct timespec *bt;
        sysclock_t delta;

        do {
                tvp->tv_sec = gd->gd_time_seconds;
                delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
        } while (tvp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tvp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;

        bt = &basetime[basetime_index];
        cpu_lfence();
        tvp->tv_sec += bt->tv_sec;
        tvp->tv_usec += bt->tv_nsec / 1000;
        while (tvp->tv_usec >= 1000000) {
                tvp->tv_usec -= 1000000;
                ++tvp->tv_sec;
        }
}

void
getnanotime(struct timespec *tsp)
{
        struct globaldata *gd = mycpu;
        struct timespec *bt;
        sysclock_t delta;

        do {
                tsp->tv_sec = gd->gd_time_seconds;
                delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
        } while (tsp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tsp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;

        bt = &basetime[basetime_index];
        cpu_lfence();
        tsp->tv_sec += bt->tv_sec;
        tsp->tv_nsec += bt->tv_nsec;
        while (tsp->tv_nsec >= 1000000000) {
                tsp->tv_nsec -= 1000000000;
                ++tsp->tv_sec;
        }
}

static void
getnanotime_nbt(struct timespec *nbt, struct timespec *tsp)
{
        struct globaldata *gd = mycpu;
        sysclock_t delta;

        do {
                tsp->tv_sec = gd->gd_time_seconds;
                delta = gd->gd_hardclock.time - gd->gd_cpuclock_base;
        } while (tsp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tsp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;

        tsp->tv_sec += nbt->tv_sec;
        tsp->tv_nsec += nbt->tv_nsec;
        while (tsp->tv_nsec >= 1000000000) {
                tsp->tv_nsec -= 1000000000;
                ++tsp->tv_sec;
        }
}


void
microtime(struct timeval *tvp)
{
        struct globaldata *gd = mycpu;
        struct timespec *bt;
        sysclock_t delta;

        do {
                tvp->tv_sec = gd->gd_time_seconds;
                delta = sys_cputimer->count() - gd->gd_cpuclock_base;
        } while (tvp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tvp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tvp->tv_usec = (sys_cputimer->freq64_usec * delta) >> 32;

        bt = &basetime[basetime_index];
        cpu_lfence();
        tvp->tv_sec += bt->tv_sec;
        tvp->tv_usec += bt->tv_nsec / 1000;
        while (tvp->tv_usec >= 1000000) {
                tvp->tv_usec -= 1000000;
                ++tvp->tv_sec;
        }
}

void
nanotime(struct timespec *tsp)
{
        struct globaldata *gd = mycpu;
        struct timespec *bt;
        sysclock_t delta;

        do {
                tsp->tv_sec = gd->gd_time_seconds;
                delta = sys_cputimer->count() - gd->gd_cpuclock_base;
        } while (tsp->tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                tsp->tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        tsp->tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;

        bt = &basetime[basetime_index];
        cpu_lfence();
        tsp->tv_sec += bt->tv_sec;
        tsp->tv_nsec += bt->tv_nsec;
        while (tsp->tv_nsec >= 1000000000) {
                tsp->tv_nsec -= 1000000000;
                ++tsp->tv_sec;
        }
}

/*
 * Get an approximate time_t.  It does not have to be accurate.  This
 * function is called only from KTR and can be called with the system in
 * any state so do not use a critical section or other complex operation
 * here.
 *
 * NOTE: This is not exactly synchronized with real time.  To do that we
 *       would have to do what microtime does and check for a nanoseconds
 *       overflow.
 */
time_t
get_approximate_time_t(void)
{
        struct globaldata *gd = mycpu;
        struct timespec *bt;

        bt = &basetime[basetime_index];
        return(gd->gd_time_seconds + bt->tv_sec);
}

int
pps_ioctl(u_long cmd, caddr_t data, struct pps_state *pps)
{
        pps_params_t *app;
        struct pps_fetch_args *fapi;
#ifdef PPS_SYNC
        struct pps_kcbind_args *kapi;
#endif

        switch (cmd) {
        case PPS_IOC_CREATE:
                return (0);
        case PPS_IOC_DESTROY:
                return (0);
        case PPS_IOC_SETPARAMS:
                app = (pps_params_t *)data;
                if (app->mode & ~pps->ppscap)
                        return (EINVAL);
                pps->ppsparam = *app;         
                return (0);
        case PPS_IOC_GETPARAMS:
                app = (pps_params_t *)data;
                *app = pps->ppsparam;
                app->api_version = PPS_API_VERS_1;
                return (0);
        case PPS_IOC_GETCAP:
                *(int*)data = pps->ppscap;
                return (0);
        case PPS_IOC_FETCH:
                fapi = (struct pps_fetch_args *)data;
                if (fapi->tsformat && fapi->tsformat != PPS_TSFMT_TSPEC)
                        return (EINVAL);
                if (fapi->timeout.tv_sec || fapi->timeout.tv_nsec)
                        return (EOPNOTSUPP);
                pps->ppsinfo.current_mode = pps->ppsparam.mode;         
                fapi->pps_info_buf = pps->ppsinfo;
                return (0);
        case PPS_IOC_KCBIND:
#ifdef PPS_SYNC
                kapi = (struct pps_kcbind_args *)data;
                /* XXX Only root should be able to do this */
                if (kapi->tsformat && kapi->tsformat != PPS_TSFMT_TSPEC)
                        return (EINVAL);
                if (kapi->kernel_consumer != PPS_KC_HARDPPS)
                        return (EINVAL);
                if (kapi->edge & ~pps->ppscap)
                        return (EINVAL);
                pps->kcmode = kapi->edge;
                return (0);
#else
                return (EOPNOTSUPP);
#endif
        default:
                return (ENOTTY);
        }
}

void
pps_init(struct pps_state *pps)
{
        pps->ppscap |= PPS_TSFMT_TSPEC;
        if (pps->ppscap & PPS_CAPTUREASSERT)
                pps->ppscap |= PPS_OFFSETASSERT;
        if (pps->ppscap & PPS_CAPTURECLEAR)
                pps->ppscap |= PPS_OFFSETCLEAR;
}

void
pps_event(struct pps_state *pps, sysclock_t count, int event)
{
        struct globaldata *gd;
        struct timespec *tsp;
        struct timespec *osp;
        struct timespec *bt;
        struct timespec ts;
        sysclock_t *pcount;
#ifdef PPS_SYNC
        sysclock_t tcount;
#endif
        sysclock_t delta;
        pps_seq_t *pseq;
        int foff;
#ifdef PPS_SYNC
        int fhard;
#endif
        int ni;

        gd = mycpu;

        /* Things would be easier with arrays... */
        if (event == PPS_CAPTUREASSERT) {
                tsp = &pps->ppsinfo.assert_timestamp;
                osp = &pps->ppsparam.assert_offset;
                foff = pps->ppsparam.mode & PPS_OFFSETASSERT;
#ifdef PPS_SYNC
                fhard = pps->kcmode & PPS_CAPTUREASSERT;
#endif
                pcount = &pps->ppscount[0];
                pseq = &pps->ppsinfo.assert_sequence;
        } else {
                tsp = &pps->ppsinfo.clear_timestamp;
                osp = &pps->ppsparam.clear_offset;
                foff = pps->ppsparam.mode & PPS_OFFSETCLEAR;
#ifdef PPS_SYNC
                fhard = pps->kcmode & PPS_CAPTURECLEAR;
#endif
                pcount = &pps->ppscount[1];
                pseq = &pps->ppsinfo.clear_sequence;
        }

        /* Nothing really happened */
        if (*pcount == count)
                return;

        *pcount = count;

        do {
                ts.tv_sec = gd->gd_time_seconds;
                delta = count - gd->gd_cpuclock_base;
        } while (ts.tv_sec != gd->gd_time_seconds);

        if (delta >= sys_cputimer->freq) {
                ts.tv_sec += delta / sys_cputimer->freq;
                delta %= sys_cputimer->freq;
        }
        ts.tv_nsec = (sys_cputimer->freq64_nsec * delta) >> 32;
        ni = basetime_index;
        cpu_lfence();
        bt = &basetime[ni];
        ts.tv_sec += bt->tv_sec;
        ts.tv_nsec += bt->tv_nsec;
        while (ts.tv_nsec >= 1000000000) {
                ts.tv_nsec -= 1000000000;
                ++ts.tv_sec;
        }

        (*pseq)++;
        *tsp = ts;

        if (foff) {
                timespecadd(tsp, osp);
                if (tsp->tv_nsec < 0) {
                        tsp->tv_nsec += 1000000000;
                        tsp->tv_sec -= 1;
                }
        }
#ifdef PPS_SYNC
        if (fhard) {
                /* magic, at its best... */
                tcount = count - pps->ppscount[2];
                pps->ppscount[2] = count;
                if (tcount >= sys_cputimer->freq) {
                        delta = (1000000000 * (tcount / sys_cputimer->freq) +
                                 sys_cputimer->freq64_nsec * 
                                 (tcount % sys_cputimer->freq)) >> 32;
                } else {
                        delta = (sys_cputimer->freq64_nsec * tcount) >> 32;
                }
                hardpps(tsp, delta);
        }
#endif
}

/*
 * Return the tsc target value for a delay of (ns).
 *
 * Returns -1 if the TSC is not supported.
 */
tsc_uclock_t
tsc_get_target(int ns)
{
#if defined(_RDTSC_SUPPORTED_)
        if (cpu_feature & CPUID_TSC) {
                return (rdtsc() + tsc_frequency * ns / (int64_t)1000000000);
        }
#endif
        return(-1);
}

/*
 * Compare the tsc against the passed target
 *
 * Returns +1 if the target has been reached
 * Returns  0 if the target has not yet been reached
 * Returns -1 if the TSC is not supported.
 *
 * Typical use:         while (tsc_test_target(target) == 0) { ...poll... }
 */
int
tsc_test_target(int64_t target)
{
#if defined(_RDTSC_SUPPORTED_)
        if (cpu_feature & CPUID_TSC) {
                if ((int64_t)(target - rdtsc()) <= 0)
                        return(1);
                return(0);
        }
#endif
        return(-1);
}

/*
 * Delay the specified number of nanoseconds using the tsc.  This function
 * returns immediately if the TSC is not supported.  At least one cpu_pause()
 * will be issued.
 */
void
tsc_delay(int ns)
{
        int64_t clk;

        clk = tsc_get_target(ns);
        cpu_pause();
        while (tsc_test_target(clk) == 0)
                cpu_pause();
}