xref: /src/sys/kern/kern_tc.c

/* $NetBSD: kern_tc.c,v 1.75 2023/07/28 10:37:28 riastradh Exp $ */

/*-
 * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Andrew Doran.
 *
 * 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.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. 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 FOUNDATION 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.
 */

/*-
 * ----------------------------------------------------------------------------
 * "THE BEER-WARE LICENSE" (Revision 42):
 * <phk (at) FreeBSD.ORG> wrote this file.  As long as you retain this notice you
 * can do whatever you want with this stuff. If we meet some day, and you think
 * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
 * ---------------------------------------------------------------------------
 */

/*
 * https://papers.freebsd.org/2002/phk-timecounters.files/timecounter.pdf
 */

#include <sys/cdefs.h>
/* __FBSDID("$FreeBSD: src/sys/kern/kern_tc.c,v 1.166 2005/09/19 22:16:31 andre Exp $"); */
__KERNEL_RCSID(0, "$NetBSD: kern_tc.c,v 1.75 2023/07/28 10:37:28 riastradh Exp $");

#ifdef _KERNEL_OPT
#include "opt_ntp.h"
#endif

#include <sys/param.h>

#include <sys/atomic.h>
#include <sys/evcnt.h>
#include <sys/kauth.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/reboot.h>	/* XXX just to get AB_VERBOSE */
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/systm.h>
#include <sys/timepps.h>
#include <sys/timetc.h>
#include <sys/timex.h>
#include <sys/xcall.h>

/*
 * A large step happens on boot.  This constant detects such steps.
 * It is relatively small so that ntp_update_second gets called enough
 * in the typical 'missed a couple of seconds' case, but doesn't loop
 * forever when the time step is large.
 */
#define LARGE_STEP	200

/*
 * Implement a dummy timecounter which we can use until we get a real one
 * in the air.  This allows the console and other early stuff to use
 * time services.
 */

static u_int
dummy_get_timecount(struct timecounter *tc)
{
	static u_int now;

	return ++now;
}

static struct timecounter dummy_timecounter = {
	.tc_get_timecount	= dummy_get_timecount,
	.tc_counter_mask	= ~0u,
	.tc_frequency		= 1000000,
	.tc_name		= "dummy",
	.tc_quality		= -1000000,
	.tc_priv		= NULL,
};

struct timehands {
	/* These fields must be initialized by the driver. */
	struct timecounter	*th_counter;     /* active timecounter */
	int64_t			th_adjustment;   /* frequency adjustment */
						 /* (NTP/adjtime) */
	uint64_t		th_scale;        /* scale factor (counter */
						 /* tick->time) */
	uint64_t 		th_offset_count; /* offset at last time */
						 /* update (tc_windup()) */
	struct bintime		th_offset;       /* bin (up)time at windup */
	struct timeval		th_microtime;    /* cached microtime */
	struct timespec		th_nanotime;     /* cached nanotime */
	/* Fields not to be copied in tc_windup start with th_generation. */
	volatile u_int		th_generation;   /* current genration */
	struct timehands	*th_next;        /* next timehand */
};

static struct timehands th0;
static struct timehands th9 = { .th_next = &th0, };
static struct timehands th8 = { .th_next = &th9, };
static struct timehands th7 = { .th_next = &th8, };
static struct timehands th6 = { .th_next = &th7, };
static struct timehands th5 = { .th_next = &th6, };
static struct timehands th4 = { .th_next = &th5, };
static struct timehands th3 = { .th_next = &th4, };
static struct timehands th2 = { .th_next = &th3, };
static struct timehands th1 = { .th_next = &th2, };
static struct timehands th0 = {
	.th_counter = &dummy_timecounter,
	.th_scale = (uint64_t)-1 / 1000000,
	.th_offset = { .sec = 1, .frac = 0 },
	.th_generation = 1,
	.th_next = &th1,
};

static struct timehands *volatile timehands = &th0;
struct timecounter *timecounter = &dummy_timecounter;
static struct timecounter *timecounters = &dummy_timecounter;

/* used by savecore(8) */
time_t time_second_legacy asm("time_second");

#ifdef __HAVE_ATOMIC64_LOADSTORE
volatile time_t time__second __cacheline_aligned = 1;
volatile time_t time__uptime __cacheline_aligned = 1;
#else
static volatile struct {
	uint32_t lo, hi;
} time__uptime32 __cacheline_aligned = {
	.lo = 1,
}, time__second32 __cacheline_aligned = {
	.lo = 1,
};
#endif

static struct {
	struct bintime bin;
	volatile unsigned gen;	/* even when stable, odd when changing */
} timebase __cacheline_aligned;

static int timestepwarnings;

kmutex_t timecounter_lock;
static u_int timecounter_mods;
static volatile int timecounter_removals = 1;
static u_int timecounter_bad;

#ifdef __HAVE_ATOMIC64_LOADSTORE

static inline void
setrealuptime(time_t second, time_t uptime)
{

	time_second_legacy = second;

	atomic_store_relaxed(&time__second, second);
	atomic_store_relaxed(&time__uptime, uptime);
}

#else

static inline void
setrealuptime(time_t second, time_t uptime)
{
	uint32_t seclo = second & 0xffffffff, sechi = second >> 32;
	uint32_t uplo = uptime & 0xffffffff, uphi = uptime >> 32;

	KDASSERT(mutex_owned(&timecounter_lock));

	time_second_legacy = second;

	/*
	 * Fast path -- no wraparound, just updating the low bits, so
	 * no need for seqlocked access.
	 */
	if (__predict_true(sechi == time__second32.hi) &&
	    __predict_true(uphi == time__uptime32.hi)) {
		atomic_store_relaxed(&time__second32.lo, seclo);
		atomic_store_relaxed(&time__uptime32.lo, uplo);
		return;
	}

	atomic_store_relaxed(&time__second32.hi, 0xffffffff);
	atomic_store_relaxed(&time__uptime32.hi, 0xffffffff);
	membar_producer();
	atomic_store_relaxed(&time__second32.lo, seclo);
	atomic_store_relaxed(&time__uptime32.lo, uplo);
	membar_producer();
	atomic_store_relaxed(&time__second32.hi, sechi);
	atomic_store_relaxed(&time__uptime32.hi, uphi);
}

time_t
getrealtime(void)
{
	uint32_t lo, hi;

	do {
		for (;;) {
			hi = atomic_load_relaxed(&time__second32.hi);
			if (__predict_true(hi != 0xffffffff))
				break;
			SPINLOCK_BACKOFF_HOOK;
		}
		membar_consumer();
		lo = atomic_load_relaxed(&time__second32.lo);
		membar_consumer();
	} while (hi != atomic_load_relaxed(&time__second32.hi));

	return ((time_t)hi << 32) | lo;
}

time_t
getuptime(void)
{
	uint32_t lo, hi;

	do {
		for (;;) {
			hi = atomic_load_relaxed(&time__uptime32.hi);
			if (__predict_true(hi != 0xffffffff))
				break;
			SPINLOCK_BACKOFF_HOOK;
		}
		membar_consumer();
		lo = atomic_load_relaxed(&time__uptime32.lo);
		membar_consumer();
	} while (hi != atomic_load_relaxed(&time__uptime32.hi));

	return ((time_t)hi << 32) | lo;
}

time_t
getboottime(void)
{

	return getrealtime() - getuptime();
}

uint32_t
getuptime32(void)
{

	return atomic_load_relaxed(&time__uptime32.lo);
}

#endif	/* !defined(__HAVE_ATOMIC64_LOADSTORE) */

/*
 * sysctl helper routine for kern.timercounter.hardware
 */
static int
sysctl_kern_timecounter_hardware(SYSCTLFN_ARGS)
{
	struct sysctlnode node;
	int error;
	char newname[MAX_TCNAMELEN];
	struct timecounter *newtc, *tc;

	tc = timecounter;

	strlcpy(newname, tc->tc_name, sizeof(newname));

	node = *rnode;
	node.sysctl_data = newname;
	node.sysctl_size = sizeof(newname);

	error = sysctl_lookup(SYSCTLFN_CALL(&node));

	if (error ||
	    newp == NULL ||
	    strncmp(newname, tc->tc_name, sizeof(newname)) == 0)
		return error;

	if (l != NULL && (error = kauth_authorize_system(l->l_cred,
	    KAUTH_SYSTEM_TIME, KAUTH_REQ_SYSTEM_TIME_TIMECOUNTERS, newname,
	    NULL, NULL)) != 0)
		return error;

	if (!cold)
		mutex_spin_enter(&timecounter_lock);
	error = EINVAL;
	for (newtc = timecounters; newtc != NULL; newtc = newtc->tc_next) {
		if (strcmp(newname, newtc->tc_name) != 0)
			continue;
		/* Warm up new timecounter. */
		(void)newtc->tc_get_timecount(newtc);
		(void)newtc->tc_get_timecount(newtc);
		timecounter = newtc;
		error = 0;
		break;
	}
	if (!cold)
		mutex_spin_exit(&timecounter_lock);
	return error;
}

static int
sysctl_kern_timecounter_choice(SYSCTLFN_ARGS)
{
	char buf[MAX_TCNAMELEN+48];
	char *where;
	const char *spc;
	struct timecounter *tc;
	size_t needed, left, slen;
	int error, mods;

	if (newp != NULL)
		return EPERM;
	if (namelen != 0)
		return EINVAL;

	mutex_spin_enter(&timecounter_lock);
 retry:
	spc = "";
	error = 0;
	needed = 0;
	left = *oldlenp;
	where = oldp;
	for (tc = timecounters; error == 0 && tc != NULL; tc = tc->tc_next) {
		if (where == NULL) {
			needed += sizeof(buf);  /* be conservative */
		} else {
			slen = snprintf(buf, sizeof(buf), "%s%s(q=%d, f=%" PRId64
					" Hz)", spc, tc->tc_name, tc->tc_quality,
					tc->tc_frequency);
			if (left < slen + 1)
				break;
		 	mods = timecounter_mods;
			mutex_spin_exit(&timecounter_lock);
			error = copyout(buf, where, slen + 1);
			mutex_spin_enter(&timecounter_lock);
			if (mods != timecounter_mods) {
				goto retry;
			}
			spc = " ";
			where += slen;
			needed += slen;
			left -= slen;
		}
	}
	mutex_spin_exit(&timecounter_lock);

	*oldlenp = needed;
	return error;
}

SYSCTL_SETUP(sysctl_timecounter_setup, "sysctl timecounter setup")
{
	const struct sysctlnode *node;

	sysctl_createv(clog, 0, NULL, &node,
		       CTLFLAG_PERMANENT,
		       CTLTYPE_NODE, "timecounter",
		       SYSCTL_DESCR("time counter information"),
		       NULL, 0, NULL, 0,
		       CTL_KERN, CTL_CREATE, CTL_EOL);

	if (node != NULL) {
		sysctl_createv(clog, 0, NULL, NULL,
			       CTLFLAG_PERMANENT,
			       CTLTYPE_STRING, "choice",
			       SYSCTL_DESCR("available counters"),
			       sysctl_kern_timecounter_choice, 0, NULL, 0,
			       CTL_KERN, node->sysctl_num, CTL_CREATE, CTL_EOL);

		sysctl_createv(clog, 0, NULL, NULL,
			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			       CTLTYPE_STRING, "hardware",
			       SYSCTL_DESCR("currently active time counter"),
			       sysctl_kern_timecounter_hardware, 0, NULL, MAX_TCNAMELEN,
			       CTL_KERN, node->sysctl_num, CTL_CREATE, CTL_EOL);

		sysctl_createv(clog, 0, NULL, NULL,
			       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			       CTLTYPE_INT, "timestepwarnings",
			       SYSCTL_DESCR("log time steps"),
			       NULL, 0, &timestepwarnings, 0,
			       CTL_KERN, node->sysctl_num, CTL_CREATE, CTL_EOL);
	}
}

#ifdef TC_COUNTERS
#define	TC_STATS(name)							\
static struct evcnt n##name =						\
    EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, "timecounter", #name);	\
EVCNT_ATTACH_STATIC(n##name)
TC_STATS(binuptime);    TC_STATS(nanouptime);    TC_STATS(microuptime);
TC_STATS(bintime);      TC_STATS(nanotime);      TC_STATS(microtime);
TC_STATS(getbinuptime); TC_STATS(getnanouptime); TC_STATS(getmicrouptime);
TC_STATS(getbintime);   TC_STATS(getnanotime);   TC_STATS(getmicrotime);
TC_STATS(setclock);
#define	TC_COUNT(var)	var.ev_count++
#undef TC_STATS
#else
#define	TC_COUNT(var)	/* nothing */
#endif	/* TC_COUNTERS */

static void tc_windup(void);

/*
 * Return the difference between the timehands' counter value now and what
 * was when we copied it to the timehands' offset_count.
 */
static inline u_int
tc_delta(struct timehands *th)
{
	struct timecounter *tc;

	tc = th->th_counter;
	return (tc->tc_get_timecount(tc) -
		 th->th_offset_count) & tc->tc_counter_mask;
}

/*
 * Functions for reading the time.  We have to loop until we are sure that
 * the timehands that we operated on was not updated under our feet.  See
 * the comment in <sys/timevar.h> for a description of these 12 functions.
 */

void
binuptime(struct bintime *bt)
{
	struct timehands *th;
	lwp_t *l;
	u_int lgen, gen;

	TC_COUNT(nbinuptime);

	/*
	 * Provide exclusion against tc_detach().
	 *
	 * We record the number of timecounter removals before accessing
	 * timecounter state.  Note that the LWP can be using multiple
	 * "generations" at once, due to interrupts (interrupted while in
	 * this function).  Hardware interrupts will borrow the interrupted
	 * LWP's l_tcgen value for this purpose, and can themselves be
	 * interrupted by higher priority interrupts.  In this case we need
	 * to ensure that the oldest generation in use is recorded.
	 *
	 * splsched() is too expensive to use, so we take care to structure
	 * this code in such a way that it is not required.  Likewise, we
	 * do not disable preemption.
	 *
	 * Memory barriers are also too expensive to use for such a
	 * performance critical function.  The good news is that we do not
	 * need memory barriers for this type of exclusion, as the thread
	 * updating timecounter_removals will issue a broadcast cross call
	 * before inspecting our l_tcgen value (this elides memory ordering
	 * issues).
	 *
	 * XXX If the author of the above comment knows how to make it
	 * safe to avoid memory barriers around the access to
	 * th->th_generation, I'm all ears.
	 */
	l = curlwp;
	lgen = l->l_tcgen;
	if (__predict_true(lgen == 0)) {
		l->l_tcgen = timecounter_removals;
	}
	__insn_barrier();

	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		*bt = th->th_offset;
		bintime_addx(bt, th->th_scale * tc_delta(th));
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);

	__insn_barrier();
	l->l_tcgen = lgen;
}

void
nanouptime(struct timespec *tsp)
{
	struct bintime bt;

	TC_COUNT(nnanouptime);
	binuptime(&bt);
	bintime2timespec(&bt, tsp);
}

void
microuptime(struct timeval *tvp)
{
	struct bintime bt;

	TC_COUNT(nmicrouptime);
	binuptime(&bt);
	bintime2timeval(&bt, tvp);
}

void
bintime(struct bintime *bt)
{
	struct bintime boottime;

	TC_COUNT(nbintime);
	binuptime(bt);
	getbinboottime(&boottime);
	bintime_add(bt, &boottime);
}

void
nanotime(struct timespec *tsp)
{
	struct bintime bt;

	TC_COUNT(nnanotime);
	bintime(&bt);
	bintime2timespec(&bt, tsp);
}

void
microtime(struct timeval *tvp)
{
	struct bintime bt;

	TC_COUNT(nmicrotime);
	bintime(&bt);
	bintime2timeval(&bt, tvp);
}

void
getbinuptime(struct bintime *bt)
{
	struct timehands *th;
	u_int gen;

	TC_COUNT(ngetbinuptime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		*bt = th->th_offset;
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
}

void
getnanouptime(struct timespec *tsp)
{
	struct timehands *th;
	u_int gen;

	TC_COUNT(ngetnanouptime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		bintime2timespec(&th->th_offset, tsp);
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
}

void
getmicrouptime(struct timeval *tvp)
{
	struct timehands *th;
	u_int gen;

	TC_COUNT(ngetmicrouptime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		bintime2timeval(&th->th_offset, tvp);
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
}

void
getbintime(struct bintime *bt)
{
	struct timehands *th;
	struct bintime boottime;
	u_int gen;

	TC_COUNT(ngetbintime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		*bt = th->th_offset;
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
	getbinboottime(&boottime);
	bintime_add(bt, &boottime);
}

static inline void
dogetnanotime(struct timespec *tsp)
{
	struct timehands *th;
	u_int gen;

	TC_COUNT(ngetnanotime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		*tsp = th->th_nanotime;
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
}

void
getnanotime(struct timespec *tsp)
{

	dogetnanotime(tsp);
}

void dtrace_getnanotime(struct timespec *tsp);

void
dtrace_getnanotime(struct timespec *tsp)
{

	dogetnanotime(tsp);
}

void
getmicrotime(struct timeval *tvp)
{
	struct timehands *th;
	u_int gen;

	TC_COUNT(ngetmicrotime);
	do {
		th = atomic_load_consume(&timehands);
		gen = th->th_generation;
		membar_consumer();
		*tvp = th->th_microtime;
		membar_consumer();
	} while (gen == 0 || gen != th->th_generation);
}

void
getnanoboottime(struct timespec *tsp)
{
	struct bintime bt;

	getbinboottime(&bt);
	bintime2timespec(&bt, tsp);
}

void
getmicroboottime(struct timeval *tvp)
{
	struct bintime bt;

	getbinboottime(&bt);
	bintime2timeval(&bt, tvp);
}

void
getbinboottime(struct bintime *basep)
{
	struct bintime base;
	unsigned gen;

	do {
		/* Spin until the timebase isn't changing.  */
		while ((gen = atomic_load_relaxed(&timebase.gen)) & 1)
			SPINLOCK_BACKOFF_HOOK;

		/* Read out a snapshot of the timebase.  */
		membar_consumer();
		base = timebase.bin;
		membar_consumer();

		/* Restart if it changed while we were reading.  */
	} while (gen != atomic_load_relaxed(&timebase.gen));

	*basep = base;
}

/*
 * Initialize a new timecounter and possibly use it.
 */
void
tc_init(struct timecounter *tc)
{
	u_int u;

	KASSERTMSG(tc->tc_next == NULL, "timecounter %s already initialised",
	    tc->tc_name);

	u = tc->tc_frequency / tc->tc_counter_mask;
	/* XXX: We need some margin here, 10% is a guess */
	u *= 11;
	u /= 10;
	if (u > hz && tc->tc_quality >= 0) {
		tc->tc_quality = -2000;
		aprint_verbose(
		    "timecounter: Timecounter \"%s\" frequency %ju Hz",
			    tc->tc_name, (uintmax_t)tc->tc_frequency);
		aprint_verbose(" -- Insufficient hz, needs at least %u\n", u);
	} else if (tc->tc_quality >= 0 || bootverbose) {
		aprint_verbose(
		    "timecounter: Timecounter \"%s\" frequency %ju Hz "
		    "quality %d\n", tc->tc_name, (uintmax_t)tc->tc_frequency,
		    tc->tc_quality);
	}

	mutex_spin_enter(&timecounter_lock);
	tc->tc_next = timecounters;
	timecounters = tc;
	timecounter_mods++;
	/*
	 * Never automatically use a timecounter with negative quality.
	 * Even though we run on the dummy counter, switching here may be
	 * worse since this timecounter may not be monotonous.
	 */
	if (tc->tc_quality >= 0 && (tc->tc_quality > timecounter->tc_quality ||
	    (tc->tc_quality == timecounter->tc_quality &&
	    tc->tc_frequency > timecounter->tc_frequency))) {
		(void)tc->tc_get_timecount(tc);
		(void)tc->tc_get_timecount(tc);
		timecounter = tc;
		tc_windup();
	}
	mutex_spin_exit(&timecounter_lock);
}

/*
 * Pick a new timecounter due to the existing counter going bad.
 */
static void
tc_pick(void)
{
	struct timecounter *best, *tc;

	KASSERT(mutex_owned(&timecounter_lock));

	for (best = tc = timecounters; tc != NULL; tc = tc->tc_next) {
		if (tc->tc_quality > best->tc_quality)
			best = tc;
		else if (tc->tc_quality < best->tc_quality)
			continue;
		else if (tc->tc_frequency > best->tc_frequency)
			best = tc;
	}
	(void)best->tc_get_timecount(best);
	(void)best->tc_get_timecount(best);
	timecounter = best;
}

/*
 * A timecounter has gone bad, arrange to pick a new one at the next
 * clock tick.
 */
void
tc_gonebad(struct timecounter *tc)
{

	tc->tc_quality = -100;
	membar_producer();
	atomic_inc_uint(&timecounter_bad);
}

/*
 * Stop using a timecounter and remove it from the timecounters list.
 */
int
tc_detach(struct timecounter *target)
{
	struct timecounter *tc;
	struct timecounter **tcp = NULL;
	int removals;
	lwp_t *l;

	/* First, find the timecounter. */
	mutex_spin_enter(&timecounter_lock);
	for (tcp = &timecounters, tc = timecounters;
	     tc != NULL;
	     tcp = &tc->tc_next, tc = tc->tc_next) {
		if (tc == target)
			break;
	}
	if (tc == NULL) {
		mutex_spin_exit(&timecounter_lock);
		return ESRCH;
	}

	/* And now, remove it. */
	*tcp = tc->tc_next;
	if (timecounter == target) {
		tc_pick();
		tc_windup();
	}
	timecounter_mods++;
	removals = timecounter_removals++;
	mutex_spin_exit(&timecounter_lock);

	/*
	 * We now have to determine if any threads in the system are still
	 * making use of this timecounter.
	 *
	 * We issue a broadcast cross call to elide memory ordering issues,
	 * then scan all LWPs in the system looking at each's timecounter
	 * generation number.  We need to see a value of zero (not actively
	 * using a timecounter) or a value greater than our removal value.
	 *
	 * We may race with threads that read `timecounter_removals' and
	 * and then get preempted before updating `l_tcgen'.  This is not
	 * a problem, since it means that these threads have not yet started
	 * accessing timecounter state.  All we do need is one clean
	 * snapshot of the system where every thread appears not to be using
	 * old timecounter state.
	 */
	for (;;) {
		xc_barrier(0);

		mutex_enter(&proc_lock);
		LIST_FOREACH(l, &alllwp, l_list) {
			if (l->l_tcgen == 0 || l->l_tcgen > removals) {
				/*
				 * Not using timecounter or old timecounter
				 * state at time of our xcall or later.
				 */
				continue;
			}
			break;
		}
		mutex_exit(&proc_lock);

		/*
		 * If the timecounter is still in use, wait at least 10ms
		 * before retrying.
		 */
		if (l == NULL) {
			break;
		}
		(void)kpause("tcdetach", false, mstohz(10), NULL);
	}

	tc->tc_next = NULL;
	return 0;
}

/* Report the frequency of the current timecounter. */
uint64_t
tc_getfrequency(void)
{

	return atomic_load_consume(&timehands)->th_counter->tc_frequency;
}

/*
 * Step our concept of UTC.  This is done by modifying our estimate of
 * when we booted.
 */
void
tc_setclock(const struct timespec *ts)
{
	struct timespec ts2;
	struct bintime bt, bt2;

	mutex_spin_enter(&timecounter_lock);
	TC_COUNT(nsetclock);
	binuptime(&bt2);
	timespec2bintime(ts, &bt);
	bintime_sub(&bt, &bt2);
	bintime_add(&bt2, &timebase.bin);
	timebase.gen |= 1;	/* change in progress */
	membar_producer();
	timebase.bin = bt;
	membar_producer();
	timebase.gen++;		/* commit change */
	tc_windup();
	mutex_spin_exit(&timecounter_lock);

	if (timestepwarnings) {
		bintime2timespec(&bt2, &ts2);
		log(LOG_INFO,
		    "Time stepped from %lld.%09ld to %lld.%09ld\n",
		    (long long)ts2.tv_sec, ts2.tv_nsec,
		    (long long)ts->tv_sec, ts->tv_nsec);
	}
}

/*
 * Initialize the next struct timehands in the ring and make
 * it the active timehands.  Along the way we might switch to a different
 * timecounter and/or do seconds processing in NTP.  Slightly magic.
 */
static void
tc_windup(void)
{
	struct bintime bt;
	struct timehands *th, *tho;
	uint64_t scale;
	u_int delta, ncount, ogen;
	int i, s_update;
	time_t t;

	KASSERT(mutex_owned(&timecounter_lock));

	s_update = 0;

	/*
	 * Make the next timehands a copy of the current one, but do not
	 * overwrite the generation or next pointer.  While we update
	 * the contents, the generation must be zero.  Ensure global
	 * visibility of the generation before proceeding.
	 */
	tho = timehands;
	th = tho->th_next;
	ogen = th->th_generation;
	th->th_generation = 0;
	membar_producer();
	bcopy(tho, th, offsetof(struct timehands, th_generation));

	/*
	 * Capture a timecounter delta on the current timecounter and if
	 * changing timecounters, a counter value from the new timecounter.
	 * Update the offset fields accordingly.
	 */
	delta = tc_delta(th);
	if (th->th_counter != timecounter)
		ncount = timecounter->tc_get_timecount(timecounter);
	else
		ncount = 0;
	th->th_offset_count += delta;
	bintime_addx(&th->th_offset, th->th_scale * delta);

	/*
	 * Hardware latching timecounters may not generate interrupts on
	 * PPS events, so instead we poll them.  There is a finite risk that
	 * the hardware might capture a count which is later than the one we
	 * got above, and therefore possibly in the next NTP second which might
	 * have a different rate than the current NTP second.  It doesn't
	 * matter in practice.
	 */
	if (tho->th_counter->tc_poll_pps)
		tho->th_counter->tc_poll_pps(tho->th_counter);

	/*
	 * Deal with NTP second processing.  The for loop normally
	 * iterates at most once, but in extreme situations it might
	 * keep NTP sane if timeouts are not run for several seconds.
	 * At boot, the time step can be large when the TOD hardware
	 * has been read, so on really large steps, we call
	 * ntp_update_second only twice.  We need to call it twice in
	 * case we missed a leap second.
	 * If NTP is not compiled in ntp_update_second still calculates
	 * the adjustment resulting from adjtime() calls.
	 */
	bt = th->th_offset;
	bintime_add(&bt, &timebase.bin);
	i = bt.sec - tho->th_microtime.tv_sec;
	if (i > LARGE_STEP)
		i = 2;
	for (; i > 0; i--) {
		t = bt.sec;
		ntp_update_second(&th->th_adjustment, &bt.sec);
		s_update = 1;
		if (bt.sec != t) {
			timebase.gen |= 1;	/* change in progress */
			membar_producer();
			timebase.bin.sec += bt.sec - t;
			membar_producer();
			timebase.gen++;		/* commit change */
		}
	}

	/* Update the UTC timestamps used by the get*() functions. */
	/* XXX shouldn't do this here.  Should force non-`get' versions. */
	bintime2timeval(&bt, &th->th_microtime);
	bintime2timespec(&bt, &th->th_nanotime);
	/* Now is a good time to change timecounters. */
	if (th->th_counter != timecounter) {
		th->th_counter = timecounter;
		th->th_offset_count = ncount;
		s_update = 1;
	}

	/*-
	 * Recalculate 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, but that
	 * leaves 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.
	 *
	 */
	if (s_update) {
		scale = (uint64_t)1 << 63;
		scale += (th->th_adjustment / 1024) * 2199;
		scale /= th->th_counter->tc_frequency;
		th->th_scale = scale * 2;
	}
	/*
	 * Now that the struct timehands is again consistent, set the new
	 * generation number, making sure to not make it zero.  Ensure
	 * changes are globally visible before changing.
	 */
	if (++ogen == 0)
		ogen = 1;
	membar_producer();
	th->th_generation = ogen;

	/*
	 * Go live with the new struct timehands.  Ensure changes are
	 * globally visible before changing.
	 */
	setrealuptime(th->th_microtime.tv_sec, th->th_offset.sec);
	atomic_store_release(&timehands, th);

	/*
	 * Force users of the old timehand to move on.  This is
	 * necessary for MP systems; we need to ensure that the
	 * consumers will move away from the old timehand before
	 * we begin updating it again when we eventually wrap
	 * around.
	 */
	if (++tho->th_generation == 0)
		tho->th_generation = 1;
}

/*
 * RFC 2783 PPS-API implementation.
 */

int
pps_ioctl(u_long cmd, void *data, struct pps_state *pps)
{
	pps_params_t *app;
	pps_info_t *pipi;
#ifdef PPS_SYNC
	int *epi;
#endif

	KASSERT(mutex_owned(&timecounter_lock));

	KASSERT(pps != NULL);

	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:
		pipi = (pps_info_t *)data;
		pps->ppsinfo.current_mode = pps->ppsparam.mode;
		*pipi = pps->ppsinfo;
		return 0;
	case PPS_IOC_KCBIND:
#ifdef PPS_SYNC
		epi = (int *)data;
		/* XXX Only root should be able to do this */
		if (*epi & ~pps->ppscap)
			return EINVAL;
		pps->kcmode = *epi;
		return 0;
#else
		return EOPNOTSUPP;
#endif
	default:
		return EPASSTHROUGH;
	}
}

void
pps_init(struct pps_state *pps)
{

	KASSERT(mutex_owned(&timecounter_lock));

	pps->ppscap |= PPS_TSFMT_TSPEC;
	if (pps->ppscap & PPS_CAPTUREASSERT)
		pps->ppscap |= PPS_OFFSETASSERT;
	if (pps->ppscap & PPS_CAPTURECLEAR)
		pps->ppscap |= PPS_OFFSETCLEAR;
}

/*
 * capture a timetamp in the pps structure
 */
void
pps_capture(struct pps_state *pps)
{
	struct timehands *th;

	KASSERT(mutex_owned(&timecounter_lock));
	KASSERT(pps != NULL);

	th = timehands;
	pps->capgen = th->th_generation;
	pps->capth = th;
	pps->capcount = (uint64_t)tc_delta(th) + th->th_offset_count;
	if (pps->capgen != th->th_generation)
		pps->capgen = 0;
}

#ifdef PPS_DEBUG
int ppsdebug = 0;
#endif

/*
 * process a pps_capture()ed event
 */
void
pps_event(struct pps_state *pps, int event)
{
	pps_ref_event(pps, event, NULL, PPS_REFEVNT_PPS|PPS_REFEVNT_CAPTURE);
}

/*
 * extended pps api /  kernel pll/fll entry point
 *
 * feed reference time stamps to PPS engine
 *
 * will simulate a PPS event and feed
 * the NTP PLL/FLL if requested.
 *
 * the ref time stamps should be roughly once
 * a second but do not need to be exactly in phase
 * with the UTC second but should be close to it.
 * this relaxation of requirements allows callout
 * driven timestamping mechanisms to feed to pps
 * capture/kernel pll logic.
 *
 * calling pattern is:
 *  pps_capture() (for PPS_REFEVNT_{CAPTURE|CAPCUR})
 *  read timestamp from reference source
 *  pps_ref_event()
 *
 * supported refmodes:
 *  PPS_REFEVNT_CAPTURE
 *    use system timestamp of pps_capture()
 *  PPS_REFEVNT_CURRENT
 *    use system timestamp of this call
 *  PPS_REFEVNT_CAPCUR
 *    use average of read capture and current system time stamp
 *  PPS_REFEVNT_PPS
 *    assume timestamp on second mark - ref_ts is ignored
 *
 */

void
pps_ref_event(struct pps_state *pps,
	      int event,
	      struct bintime *ref_ts,
	      int refmode
	)
{
	struct bintime bt;	/* current time */
	struct bintime btd;	/* time difference */
	struct bintime bt_ref;	/* reference time */
	struct timespec ts, *tsp, *osp;
	struct timehands *th;
	uint64_t tcount, acount, dcount, *pcount;
	int foff, gen;
#ifdef PPS_SYNC
	int fhard;
#endif
	pps_seq_t *pseq;

	KASSERT(mutex_owned(&timecounter_lock));

	KASSERT(pps != NULL);

        /* pick up current time stamp if needed */
	if (refmode & (PPS_REFEVNT_CURRENT|PPS_REFEVNT_CAPCUR)) {
		/* pick up current time stamp */
		th = timehands;
		gen = th->th_generation;
		tcount = (uint64_t)tc_delta(th) + th->th_offset_count;
		if (gen != th->th_generation)
			gen = 0;

		/* If the timecounter was wound up underneath us, bail out. */
		if (pps->capgen == 0 ||
		    pps->capgen != pps->capth->th_generation ||
		    gen == 0 ||
		    gen != pps->capgen) {
#ifdef PPS_DEBUG
			if (ppsdebug & 0x1) {
				log(LOG_DEBUG,
				    "pps_ref_event(pps=%p, event=%d, ...): DROP (wind-up)\n",
				    pps, event);
			}
#endif
			return;
		}
	} else {
		tcount = 0;	/* keep GCC happy */
	}

#ifdef PPS_DEBUG
	if (ppsdebug & 0x1) {
		struct timespec tmsp;

		if (ref_ts == NULL) {
			tmsp.tv_sec = 0;
			tmsp.tv_nsec = 0;
		} else {
			bintime2timespec(ref_ts, &tmsp);
		}

		log(LOG_DEBUG,
		    "pps_ref_event(pps=%p, event=%d, ref_ts=%"PRIi64
		    ".%09"PRIi32", refmode=0x%1x)\n",
		    pps, event, tmsp.tv_sec, (int32_t)tmsp.tv_nsec, refmode);
	}
#endif

	/* setup correct event references */
	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;
	}

	/* determine system time stamp according to refmode */
	dcount = 0;		/* keep GCC happy */
	switch (refmode & PPS_REFEVNT_RMASK) {
	case PPS_REFEVNT_CAPTURE:
		acount = pps->capcount;	/* use capture timestamp */
		break;

	case PPS_REFEVNT_CURRENT:
		acount = tcount; /* use current timestamp */
		break;

	case PPS_REFEVNT_CAPCUR:
		/*
		 * calculate counter value between pps_capture() and
		 * pps_ref_event()
		 */
		dcount = tcount - pps->capcount;
		acount = (dcount / 2) + pps->capcount;
		break;

	default:		/* ignore call error silently */
		return;
	}

	/*
	 * If the timecounter changed, we cannot compare the count values, so
	 * we have to drop the rest of the PPS-stuff until the next event.
	 */
	if (pps->ppstc != pps->capth->th_counter) {
		pps->ppstc = pps->capth->th_counter;
		pps->capcount = acount;
		*pcount = acount;
		pps->ppscount[2] = acount;
#ifdef PPS_DEBUG
		if (ppsdebug & 0x1) {
			log(LOG_DEBUG,
			    "pps_ref_event(pps=%p, event=%d, ...): DROP (time-counter change)\n",
			    pps, event);
		}
#endif
		return;
	}

	pps->capcount = acount;

	/* Convert the count to a bintime. */
	bt = pps->capth->th_offset;
	bintime_addx(&bt, pps->capth->th_scale * (acount - pps->capth->th_offset_count));
	bintime_add(&bt, &timebase.bin);

	if ((refmode & PPS_REFEVNT_PPS) == 0) {
		/* determine difference to reference time stamp */
		bt_ref = *ref_ts;

		btd = bt;
		bintime_sub(&btd, &bt_ref);

		/*
		 * simulate a PPS timestamp by dropping the fraction
		 * and applying the offset
		 */
		if (bt.frac >= (uint64_t)1<<63)	/* skip to nearest second */
			bt.sec++;
		bt.frac = 0;
		bintime_add(&bt, &btd);
	} else {
		/*
		 * create ref_ts from current time -
		 * we are supposed to be called on
		 * the second mark
		 */
		bt_ref = bt;
		if (bt_ref.frac >= (uint64_t)1<<63)	/* skip to nearest second */
			bt_ref.sec++;
		bt_ref.frac = 0;
	}

	/* convert bintime to timestamp */
	bintime2timespec(&bt, &ts);

	/* If the timecounter was wound up underneath us, bail out. */
	if (pps->capgen != pps->capth->th_generation)
		return;

	/* store time stamp */
	*pcount = pps->capcount;
	(*pseq)++;
	*tsp = ts;

	/* add offset correction */
	if (foff) {
		timespecadd(tsp, osp, tsp);
		if (tsp->tv_nsec < 0) {
			tsp->tv_nsec += 1000000000;
			tsp->tv_sec -= 1;
		}
	}

#ifdef PPS_DEBUG
	if (ppsdebug & 0x2) {
		struct timespec ts2;
		struct timespec ts3;

		bintime2timespec(&bt_ref, &ts2);

		bt.sec = 0;
		bt.frac = 0;

		if (refmode & PPS_REFEVNT_CAPCUR) {
			    bintime_addx(&bt, pps->capth->th_scale * dcount);
		}
		bintime2timespec(&bt, &ts3);

		log(LOG_DEBUG, "ref_ts=%"PRIi64".%09"PRIi32
		    ", ts=%"PRIi64".%09"PRIi32", read latency=%"PRIi64" ns\n",
		    ts2.tv_sec, (int32_t)ts2.tv_nsec,
		    tsp->tv_sec, (int32_t)tsp->tv_nsec,
		    timespec2ns(&ts3));
	}
#endif

#ifdef PPS_SYNC
	if (fhard) {
		uint64_t scale;
		uint64_t div;

		/*
		 * Feed the NTP PLL/FLL.
		 * The FLL wants to know how many (hardware) nanoseconds
		 * elapsed since the previous event (mod 1 second) thus
		 * we are actually looking at the frequency difference scaled
		 * in nsec.
		 * As the counter time stamps are not truly at 1Hz
		 * we need to scale the count by the elapsed
		 * reference time.
		 * valid sampling interval: [0.5..2[ sec
		 */

		/* calculate elapsed raw count */
		tcount = pps->capcount - pps->ppscount[2];
		pps->ppscount[2] = pps->capcount;
		tcount &= pps->capth->th_counter->tc_counter_mask;

		/* calculate elapsed ref time */
		btd = bt_ref;
		bintime_sub(&btd, &pps->ref_time);
		pps->ref_time = bt_ref;

		/* check that we stay below 2 sec */
		if (btd.sec < 0 || btd.sec > 1)
			return;

		/* we want at least 0.5 sec between samples */
		if (btd.sec == 0 && btd.frac < (uint64_t)1<<63)
			return;

		/*
		 * calculate cycles per period by multiplying
		 * the frequency with the elapsed period
		 * we pick a fraction of 30 bits
		 * ~1ns resolution for elapsed time
		 */
		div   = (uint64_t)btd.sec << 30;
		div  |= (btd.frac >> 34) & (((uint64_t)1 << 30) - 1);
		div  *= pps->capth->th_counter->tc_frequency;
		div >>= 30;

		if (div == 0)	/* safeguard */
			return;

		scale = (uint64_t)1 << 63;
		scale /= div;
		scale *= 2;

		bt.sec = 0;
		bt.frac = 0;
		bintime_addx(&bt, scale * tcount);
		bintime2timespec(&bt, &ts);

#ifdef PPS_DEBUG
		if (ppsdebug & 0x4) {
			struct timespec ts2;
			int64_t df;

			bintime2timespec(&bt_ref, &ts2);
			df = timespec2ns(&ts);
			if (df > 500000000)
				df -= 1000000000;
			log(LOG_DEBUG, "hardpps: ref_ts=%"PRIi64
			    ".%09"PRIi32", ts=%"PRIi64".%09"PRIi32
			    ", freqdiff=%"PRIi64" ns/s\n",
			    ts2.tv_sec, (int32_t)ts2.tv_nsec,
			    tsp->tv_sec, (int32_t)tsp->tv_nsec,
			    df);
		}
#endif

		hardpps(tsp, timespec2ns(&ts));
	}
#endif
}

/*
 * Timecounters need to be updated every so often to prevent the hardware
 * counter from overflowing.  Updating also recalculates the cached values
 * used by the get*() family of functions, so their precision depends on
 * the update frequency.
 */

static int tc_tick;

void
tc_ticktock(void)
{
	static int count;

	if (++count < tc_tick)
		return;
	count = 0;
	mutex_spin_enter(&timecounter_lock);
	if (__predict_false(timecounter_bad != 0)) {
		/* An existing timecounter has gone bad, pick a new one. */
		(void)atomic_swap_uint(&timecounter_bad, 0);
		if (timecounter->tc_quality < 0) {
			tc_pick();
		}
	}
	tc_windup();
	mutex_spin_exit(&timecounter_lock);
}

void
inittimecounter(void)
{
	u_int p;

	mutex_init(&timecounter_lock, MUTEX_DEFAULT, IPL_HIGH);

	/*
	 * Set the initial timeout to
	 * max(1, <approx. number of hardclock ticks in a millisecond>).
	 * People should probably not use the sysctl to set the timeout
	 * to smaller than its initial value, since that value is the
	 * smallest reasonable one.  If they want better timestamps they
	 * should use the non-"get"* functions.
	 */
	if (hz > 1000)
		tc_tick = (hz + 500) / 1000;
	else
		tc_tick = 1;
	p = (tc_tick * 1000000) / hz;
	aprint_verbose("timecounter: Timecounters tick every %d.%03u msec\n",
	    p / 1000, p % 1000);

	/* warm up new timecounter (again) and get rolling. */
	(void)timecounter->tc_get_timecount(timecounter);
	(void)timecounter->tc_get_timecount(timecounter);
}