xref: /freebsd/sys/x86/x86/tsc.c (revision ecaab0fb)

/*-
 * SPDX-License-Identifier: BSD-2-Clause
 *
 * Copyright (c) 1998-2003 Poul-Henning Kamp
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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.
 */

#include <sys/cdefs.h>
#include "opt_clock.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bus.h>
#include <sys/cpu.h>
#include <sys/eventhandler.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/sched.h>
#include <sys/sysctl.h>
#include <sys/time.h>
#include <sys/timetc.h>
#include <sys/kernel.h>
#include <sys/smp.h>
#include <sys/vdso.h>
#include <machine/clock.h>
#include <machine/cputypes.h>
#include <machine/fpu.h>
#include <machine/md_var.h>
#include <machine/specialreg.h>
#include <x86/vmware.h>
#include <dev/acpica/acpi_hpet.h>
#include <contrib/dev/acpica/include/acpi.h>

#include "cpufreq_if.h"

uint64_t	tsc_freq;
int		tsc_is_invariant;
int		tsc_perf_stat;
static int	tsc_early_calib_exact;

static eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag;

SYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN,
    &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant");

#ifdef SMP
int	smp_tsc;
SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0,
    "Indicates whether the TSC is safe to use in SMP mode");

int	smp_tsc_adjust = 0;
SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc_adjust, CTLFLAG_RDTUN,
    &smp_tsc_adjust, 0, "Try to adjust TSC on APs to match BSP");
#endif

static int	tsc_shift = 1;
SYSCTL_INT(_kern_timecounter, OID_AUTO, tsc_shift, CTLFLAG_RDTUN,
    &tsc_shift, 0, "Shift to pre-apply for the maximum TSC frequency");

static int	tsc_disabled;
SYSCTL_INT(_machdep, OID_AUTO, disable_tsc, CTLFLAG_RDTUN, &tsc_disabled, 0,
    "Disable x86 Time Stamp Counter");

static int	tsc_skip_calibration;
SYSCTL_INT(_machdep, OID_AUTO, disable_tsc_calibration, CTLFLAG_RDTUN,
    &tsc_skip_calibration, 0,
    "Disable early TSC frequency calibration");

static void tsc_freq_changed(void *arg, const struct cf_level *level,
    int status);
static void tsc_freq_changing(void *arg, const struct cf_level *level,
    int *status);
static u_int tsc_get_timecount(struct timecounter *tc);
static inline u_int tsc_get_timecount_low(struct timecounter *tc);
static u_int tsc_get_timecount_lfence(struct timecounter *tc);
static u_int tsc_get_timecount_low_lfence(struct timecounter *tc);
static u_int tsc_get_timecount_mfence(struct timecounter *tc);
static u_int tsc_get_timecount_low_mfence(struct timecounter *tc);
static u_int tscp_get_timecount(struct timecounter *tc);
static u_int tscp_get_timecount_low(struct timecounter *tc);
static void tsc_levels_changed(void *arg, int unit);
static uint32_t x86_tsc_vdso_timehands(struct vdso_timehands *vdso_th,
    struct timecounter *tc);
#ifdef COMPAT_FREEBSD32
static uint32_t x86_tsc_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
    struct timecounter *tc);
#endif

static struct timecounter tsc_timecounter = {
	.tc_get_timecount =		tsc_get_timecount,
	.tc_counter_mask =		~0u,
	.tc_name =			"TSC",
	.tc_quality =			800,	/* adjusted in code */
	.tc_fill_vdso_timehands = 	x86_tsc_vdso_timehands,
#ifdef COMPAT_FREEBSD32
	.tc_fill_vdso_timehands32 = 	x86_tsc_vdso_timehands32,
#endif
};

static int
tsc_freq_cpuid_vm(void)
{
	u_int regs[4];

	if (vm_guest == VM_GUEST_NO)
		return (false);
	if (hv_high < 0x40000010)
		return (false);
	do_cpuid(0x40000010, regs);
	tsc_freq = (uint64_t)(regs[0]) * 1000;
	tsc_early_calib_exact = 1;
	return (true);
}

static void
tsc_freq_vmware(void)
{
	u_int regs[4];

	vmware_hvcall(0, VMW_HVCMD_GETHZ, VMW_HVCMD_DEFAULT_PARAM, regs);
	if (regs[1] != UINT_MAX)
		tsc_freq = regs[0] | ((uint64_t)regs[1] << 32);
	tsc_early_calib_exact = 1;
}

static void
tsc_freq_xen(void)
{
	u_int regs[4];

	/*
	 * Must run *after* generic tsc_freq_cpuid_vm, so that when Xen is
	 * emulating Viridian support the Viridian leaf is used instead.
	 */
	KASSERT(hv_high >= 0x40000003, ("Invalid max hypervisor leaf on Xen"));
	cpuid_count(0x40000003, 0, regs);
	tsc_freq = (uint64_t)(regs[2]) * 1000;
	tsc_early_calib_exact = 1;
}

/*
 * Calculate TSC frequency using information from the CPUID leaf 0x15 'Time
 * Stamp Counter and Nominal Core Crystal Clock'.  If leaf 0x15 is not
 * functional, as it is on Skylake/Kabylake, try 0x16 'Processor Frequency
 * Information'.  Leaf 0x16 is described in the SDM as informational only, but
 * we can use this value until late calibration is complete.
 */
static bool
tsc_freq_cpuid(uint64_t *res)
{
	u_int regs[4];

	if (cpu_high < 0x15)
		return (false);
	do_cpuid(0x15, regs);
	if (regs[0] != 0 && regs[1] != 0 && regs[2] != 0) {
		*res = (uint64_t)regs[2] * regs[1] / regs[0];
		return (true);
	}

	if (cpu_high < 0x16)
		return (false);
	do_cpuid(0x16, regs);
	if (regs[0] != 0) {
		*res = (uint64_t)regs[0] * 1000000;
		return (true);
	}

	return (false);
}

static bool
tsc_freq_intel_brand(uint64_t *res)
{
	char brand[48];
	u_int regs[4];
	uint64_t freq;
	char *p;
	u_int i;

	/*
	 * Intel Processor Identification and the CPUID Instruction
	 * Application Note 485.
	 * http://www.intel.com/assets/pdf/appnote/241618.pdf
	 */
	if (cpu_exthigh >= 0x80000004) {
		p = brand;
		for (i = 0x80000002; i < 0x80000005; i++) {
			do_cpuid(i, regs);
			memcpy(p, regs, sizeof(regs));
			p += sizeof(regs);
		}
		p = NULL;
		for (i = 0; i < sizeof(brand) - 1; i++)
			if (brand[i] == 'H' && brand[i + 1] == 'z')
				p = brand + i;
		if (p != NULL) {
			p -= 5;
			switch (p[4]) {
			case 'M':
				i = 1;
				break;
			case 'G':
				i = 1000;
				break;
			case 'T':
				i = 1000000;
				break;
			default:
				return (false);
			}
#define	C2D(c)	((c) - '0')
			if (p[1] == '.') {
				freq = C2D(p[0]) * 1000;
				freq += C2D(p[2]) * 100;
				freq += C2D(p[3]) * 10;
				freq *= i * 1000;
			} else {
				freq = C2D(p[0]) * 1000;
				freq += C2D(p[1]) * 100;
				freq += C2D(p[2]) * 10;
				freq += C2D(p[3]);
				freq *= i * 1000000;
			}
#undef C2D
			*res = freq;
			return (true);
		}
	}
	return (false);
}

static void
tsc_freq_tc(uint64_t *res)
{
	uint64_t tsc1, tsc2;
	int64_t overhead;
	int count, i;

	overhead = 0;
	for (i = 0, count = 8; i < count; i++) {
		tsc1 = rdtsc_ordered();
		DELAY(0);
		tsc2 = rdtsc_ordered();
		if (i > 0)
			overhead += tsc2 - tsc1;
	}
	overhead /= count;

	tsc1 = rdtsc_ordered();
	DELAY(100000);
	tsc2 = rdtsc_ordered();
	tsc_freq = (tsc2 - tsc1 - overhead) * 10;
}

/*
 * Try to determine the TSC frequency using CPUID or hypercalls.  If successful,
 * this lets use the TSC for early DELAY() calls instead of the 8254 timer,
 * which may be unreliable or entirely absent on contemporary systems.  However,
 * avoid calibrating using the 8254 here so as to give hypervisors a chance to
 * register a timecounter that can be used instead.
 */
static void
probe_tsc_freq_early(void)
{
#ifdef __i386__
	/* The TSC is known to be broken on certain CPUs. */
	switch (cpu_vendor_id) {
	case CPU_VENDOR_AMD:
		switch (cpu_id & 0xFF0) {
		case 0x500:
			/* K5 Model 0 */
			tsc_disabled = 1;
			return;
		}
		break;
	case CPU_VENDOR_CENTAUR:
		switch (cpu_id & 0xff0) {
		case 0x540:
			/*
			 * http://www.centtech.com/c6_data_sheet.pdf
			 *
			 * I-12 RDTSC may return incoherent values in EDX:EAX
			 * I-13 RDTSC hangs when certain event counters are used
			 */
			tsc_disabled = 1;
			return;
		}
		break;
	case CPU_VENDOR_NSC:
		switch (cpu_id & 0xff0) {
		case 0x540:
			if ((cpu_id & CPUID_STEPPING) == 0) {
				tsc_disabled = 1;
				return;
			}
			break;
		}
		break;
	}
#endif

	switch (cpu_vendor_id) {
	case CPU_VENDOR_AMD:
	case CPU_VENDOR_HYGON:
		if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 ||
		    (vm_guest == VM_GUEST_NO &&
		    CPUID_TO_FAMILY(cpu_id) >= 0x10))
			tsc_is_invariant = 1;
		if (cpu_feature & CPUID_SSE2) {
			tsc_timecounter.tc_get_timecount =
			    tsc_get_timecount_mfence;
		}
		break;
	case CPU_VENDOR_INTEL:
		if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 ||
		    (vm_guest == VM_GUEST_NO &&
		    ((CPUID_TO_FAMILY(cpu_id) == 0x6 &&
		    CPUID_TO_MODEL(cpu_id) >= 0xe) ||
		    (CPUID_TO_FAMILY(cpu_id) == 0xf &&
		    CPUID_TO_MODEL(cpu_id) >= 0x3))))
			tsc_is_invariant = 1;
		if (cpu_feature & CPUID_SSE2) {
			tsc_timecounter.tc_get_timecount =
			    tsc_get_timecount_lfence;
		}
		break;
	case CPU_VENDOR_CENTAUR:
		if (vm_guest == VM_GUEST_NO &&
		    CPUID_TO_FAMILY(cpu_id) == 0x6 &&
		    CPUID_TO_MODEL(cpu_id) >= 0xf &&
		    (rdmsr(0x1203) & 0x100000000ULL) == 0)
			tsc_is_invariant = 1;
		if (cpu_feature & CPUID_SSE2) {
			tsc_timecounter.tc_get_timecount =
			    tsc_get_timecount_lfence;
		}
		break;
	}

	if (tsc_freq_cpuid_vm()) {
		if (bootverbose)
			printf(
		    "Early TSC frequency %juHz derived from hypervisor CPUID\n",
			    (uintmax_t)tsc_freq);
	} else if (vm_guest == VM_GUEST_VMWARE) {
		tsc_freq_vmware();
		if (bootverbose)
			printf(
		    "Early TSC frequency %juHz derived from VMWare hypercall\n",
			    (uintmax_t)tsc_freq);
	} else if (vm_guest == VM_GUEST_XEN) {
		tsc_freq_xen();
		if (bootverbose)
			printf(
			"Early TSC frequency %juHz derived from Xen CPUID\n",
			    (uintmax_t)tsc_freq);
	} else if (tsc_freq_cpuid(&tsc_freq)) {
		/*
		 * If possible, use the value obtained from CPUID as the initial
		 * frequency.  This will be refined later during boot but is
		 * good enough for now.  The 8254 PIT is not functional on some
		 * newer platforms anyway, so don't delay our boot for what
		 * might be a garbage result.  Late calibration is required if
		 * the initial frequency was obtained from CPUID.16H, as the
		 * derived value may be off by as much as 1%.
		 */
		if (bootverbose)
			printf("Early TSC frequency %juHz derived from CPUID\n",
			    (uintmax_t)tsc_freq);
	}
}

/*
 * If we were unable to determine the TSC frequency via CPU registers, try
 * to calibrate against a known clock.
 */
static void
probe_tsc_freq_late(void)
{
	if (tsc_freq != 0)
		return;

	if (tsc_skip_calibration) {
		/*
		 * Try to parse the brand string to obtain the nominal TSC
		 * frequency.
		 */
		if (cpu_vendor_id == CPU_VENDOR_INTEL &&
		    tsc_freq_intel_brand(&tsc_freq)) {
			if (bootverbose)
				printf(
		    "Early TSC frequency %juHz derived from brand string\n",
				    (uintmax_t)tsc_freq);
		} else {
			tsc_disabled = 1;
		}
	} else {
		/*
		 * Calibrate against a timecounter or the 8254 PIT.  This
		 * estimate will be refined later in tsc_calib().
		 */
		tsc_freq_tc(&tsc_freq);
		if (bootverbose)
			printf(
		    "Early TSC frequency %juHz calibrated from 8254 PIT\n",
			    (uintmax_t)tsc_freq);
	}
}

void
start_TSC(void)
{
	if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled)
		return;

	probe_tsc_freq_late();

	if (cpu_power_ecx & CPUID_PERF_STAT) {
		/*
		 * XXX Some emulators expose host CPUID without actual support
		 * for these MSRs.  We must test whether they really work.
		 */
		wrmsr(MSR_MPERF, 0);
		wrmsr(MSR_APERF, 0);
		DELAY(10);
		if (rdmsr(MSR_MPERF) > 0 && rdmsr(MSR_APERF) > 0)
			tsc_perf_stat = 1;
	}

	/*
	 * Inform CPU accounting about our boot-time clock rate.  This will
	 * be updated if someone loads a cpufreq driver after boot that
	 * discovers a new max frequency.
	 *
	 * The frequency may also be updated after late calibration is complete;
	 * however, we register the TSC as the ticker now to avoid switching
	 * counters after much of the kernel has already booted and potentially
	 * sampled the CPU clock.
	 */
	if (tsc_freq != 0)
		set_cputicker(rdtsc, tsc_freq, !tsc_is_invariant);

	if (tsc_is_invariant)
		return;

	/* Register to find out about changes in CPU frequency. */
	tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change,
	    tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST);
	tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change,
	    tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST);
	tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed,
	    tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY);
}

#ifdef SMP

/*
 * RDTSC is not a serializing instruction, and does not drain
 * instruction stream, so we need to drain the stream before executing
 * it.  It could be fixed by use of RDTSCP, except the instruction is
 * not available everywhere.
 *
 * Use CPUID for draining in the boot-time SMP constistency test.  The
 * timecounters use MFENCE for AMD CPUs, and LFENCE for others (Intel
 * and VIA) when SSE2 is present, and nothing on older machines which
 * also do not issue RDTSC prematurely.  There, testing for SSE2 and
 * vendor is too cumbersome, and we learn about TSC presence from CPUID.
 *
 * Do not use do_cpuid(), since we do not need CPUID results, which
 * have to be written into memory with do_cpuid().
 */
#define	TSC_READ(x)							\
static void								\
tsc_read_##x(void *arg)							\
{									\
	uint64_t *tsc = arg;						\
	u_int cpu = PCPU_GET(cpuid);					\
									\
	__asm __volatile("cpuid" : : : "eax", "ebx", "ecx", "edx");	\
	tsc[cpu * 3 + x] = rdtsc();					\
}
TSC_READ(0)
TSC_READ(1)
TSC_READ(2)
#undef TSC_READ

#define	N	1000

static void
comp_smp_tsc(void *arg)
{
	uint64_t *tsc;
	int64_t d1, d2;
	u_int cpu = PCPU_GET(cpuid);
	u_int i, j, size;

	size = (mp_maxid + 1) * 3;
	for (i = 0, tsc = arg; i < N; i++, tsc += size)
		CPU_FOREACH(j) {
			if (j == cpu)
				continue;
			d1 = tsc[cpu * 3 + 1] - tsc[j * 3];
			d2 = tsc[cpu * 3 + 2] - tsc[j * 3 + 1];
			if (d1 <= 0 || d2 <= 0) {
				smp_tsc = 0;
				return;
			}
		}
}

static void
adj_smp_tsc(void *arg)
{
	uint64_t *tsc;
	int64_t d, min, max;
	u_int cpu = PCPU_GET(cpuid);
	u_int first, i, size;

	first = CPU_FIRST();
	if (cpu == first)
		return;
	min = INT64_MIN;
	max = INT64_MAX;
	size = (mp_maxid + 1) * 3;
	for (i = 0, tsc = arg; i < N; i++, tsc += size) {
		d = tsc[first * 3] - tsc[cpu * 3 + 1];
		if (d > min)
			min = d;
		d = tsc[first * 3 + 1] - tsc[cpu * 3 + 2];
		if (d > min)
			min = d;
		d = tsc[first * 3 + 1] - tsc[cpu * 3];
		if (d < max)
			max = d;
		d = tsc[first * 3 + 2] - tsc[cpu * 3 + 1];
		if (d < max)
			max = d;
	}
	if (min > max)
		return;
	d = min / 2 + max / 2;
	__asm __volatile (
		"movl $0x10, %%ecx\n\t"
		"rdmsr\n\t"
		"addl %%edi, %%eax\n\t"
		"adcl %%esi, %%edx\n\t"
		"wrmsr\n"
		: /* No output */
		: "D" ((uint32_t)d), "S" ((uint32_t)(d >> 32))
		: "ax", "cx", "dx", "cc"
	);
}

static int
test_tsc(int adj_max_count)
{
	uint64_t *data, *tsc;
	u_int i, size, adj;

	if ((!smp_tsc && !tsc_is_invariant))
		return (-100);
	/*
	 * Misbehavior of TSC under VirtualBox has been observed.  In
	 * particular, threads doing small (~1 second) sleeps may miss their
	 * wakeup and hang around in sleep state, causing hangs on shutdown.
	 */
	if (vm_guest == VM_GUEST_VBOX)
		return (0);

	TSENTER();
	size = (mp_maxid + 1) * 3;
	data = malloc(sizeof(*data) * size * N, M_TEMP, M_WAITOK);
	adj = 0;
retry:
	for (i = 0, tsc = data; i < N; i++, tsc += size)
		smp_rendezvous(tsc_read_0, tsc_read_1, tsc_read_2, tsc);
	smp_tsc = 1;	/* XXX */
	smp_rendezvous(smp_no_rendezvous_barrier, comp_smp_tsc,
	    smp_no_rendezvous_barrier, data);
	if (!smp_tsc && adj < adj_max_count) {
		adj++;
		smp_rendezvous(smp_no_rendezvous_barrier, adj_smp_tsc,
		    smp_no_rendezvous_barrier, data);
		goto retry;
	}
	free(data, M_TEMP);
	if (bootverbose)
		printf("SMP: %sed TSC synchronization test%s\n",
		    smp_tsc ? "pass" : "fail",
		    adj > 0 ? " after adjustment" : "");
	TSEXIT();
	if (smp_tsc && tsc_is_invariant) {
		switch (cpu_vendor_id) {
		case CPU_VENDOR_AMD:
		case CPU_VENDOR_HYGON:
			/*
			 * Processor Programming Reference (PPR) for AMD
			 * Family 17h states that the TSC uses a common
			 * reference for all sockets, cores and threads.
			 */
			if (CPUID_TO_FAMILY(cpu_id) >= 0x17)
				return (1000);
			/*
			 * Starting with Family 15h processors, TSC clock
			 * source is in the north bridge.  Check whether
			 * we have a single-socket/multi-core platform.
			 * XXX Need more work for complex cases.
			 */
			if (CPUID_TO_FAMILY(cpu_id) < 0x15 ||
			    (amd_feature2 & AMDID2_CMP) == 0 ||
			    smp_cpus > (cpu_procinfo2 & AMDID_CMP_CORES) + 1)
				break;
			return (1000);
		case CPU_VENDOR_INTEL:
			/*
			 * XXX Assume Intel platforms have synchronized TSCs.
			 */
			return (1000);
		}
		return (800);
	}
	return (-100);
}

#undef N

#endif /* SMP */

static void
init_TSC_tc(void)
{
	uint64_t max_freq;
	int shift;

	if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled)
		return;

	/*
	 * Limit timecounter frequency to fit in an int and prevent it from
	 * overflowing too fast.
	 */
	max_freq = UINT_MAX;

	/*
	 * Intel CPUs without a C-state invariant TSC can stop the TSC
	 * in either C2 or C3.  Disable use of C2 and C3 while using
	 * the TSC as the timecounter.  The timecounter can be changed
	 * to enable C2 and C3.
	 *
	 * Note that the TSC is used as the cputicker for computing
	 * thread runtime regardless of the timecounter setting, so
	 * using an alternate timecounter and enabling C2 or C3 can
	 * result incorrect runtimes for kernel idle threads (but not
	 * for any non-idle threads).
	 */
	if (cpu_vendor_id == CPU_VENDOR_INTEL &&
	    (amd_pminfo & AMDPM_TSC_INVARIANT) == 0) {
		tsc_timecounter.tc_flags |= TC_FLAGS_C2STOP;
		if (bootverbose)
			printf("TSC timecounter disables C2 and C3.\n");
	}

	/*
	 * We can not use the TSC in SMP mode unless the TSCs on all CPUs
	 * are synchronized.  If the user is sure that the system has
	 * synchronized TSCs, set kern.timecounter.smp_tsc tunable to a
	 * non-zero value.  The TSC seems unreliable in virtualized SMP
	 * environments, so it is set to a negative quality in those cases.
	 */
#ifdef SMP
	if (mp_ncpus > 1)
		tsc_timecounter.tc_quality = test_tsc(smp_tsc_adjust);
	else
#endif /* SMP */
	if (tsc_is_invariant)
		tsc_timecounter.tc_quality = 1000;
	max_freq >>= tsc_shift;

	for (shift = 0; shift <= 31 && (tsc_freq >> shift) > max_freq; shift++)
		;

	/*
	 * Timecounter implementation selection, top to bottom:
	 * - If RDTSCP is available, use RDTSCP.
	 * - If fence instructions are provided (SSE2), use LFENCE;RDTSC
	 *   on Intel, and MFENCE;RDTSC on AMD.
	 * - For really old CPUs, just use RDTSC.
	 */
	if ((amd_feature & AMDID_RDTSCP) != 0) {
		tsc_timecounter.tc_get_timecount = shift > 0 ?
		    tscp_get_timecount_low : tscp_get_timecount;
	} else if ((cpu_feature & CPUID_SSE2) != 0 && mp_ncpus > 1) {
		if (cpu_vendor_id == CPU_VENDOR_AMD ||
		    cpu_vendor_id == CPU_VENDOR_HYGON) {
			tsc_timecounter.tc_get_timecount = shift > 0 ?
			    tsc_get_timecount_low_mfence :
			    tsc_get_timecount_mfence;
		} else {
			tsc_timecounter.tc_get_timecount = shift > 0 ?
			    tsc_get_timecount_low_lfence :
			    tsc_get_timecount_lfence;
		}
	} else {
		tsc_timecounter.tc_get_timecount = shift > 0 ?
		    tsc_get_timecount_low : tsc_get_timecount;
	}
	if (shift > 0) {
		tsc_timecounter.tc_name = "TSC-low";
		if (bootverbose)
			printf("TSC timecounter discards lower %d bit(s)\n",
			    shift);
	}
	if (tsc_freq != 0) {
		tsc_timecounter.tc_frequency = tsc_freq >> shift;
		tsc_timecounter.tc_priv = (void *)(intptr_t)shift;

		/*
		 * Timecounter registration is deferred until after late
		 * calibration is finished.
		 */
	}
}
SYSINIT(tsc_tc, SI_SUB_SMP, SI_ORDER_ANY, init_TSC_tc, NULL);

static void
tsc_update_freq(uint64_t new_freq)
{
	atomic_store_rel_64(&tsc_freq, new_freq);
	atomic_store_rel_64(&tsc_timecounter.tc_frequency,
	    new_freq >> (int)(intptr_t)tsc_timecounter.tc_priv);
}

void
tsc_init(void)
{
	if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled)
		return;

	probe_tsc_freq_early();
}

/*
 * Perform late calibration of the TSC frequency once ACPI-based timecounters
 * are available.  At this point timehands are not set up, so we read the
 * highest-quality timecounter directly rather than using (s)binuptime().
 */
void
tsc_calibrate(void)
{
	uint64_t freq;

	if (tsc_disabled)
		return;
	if (tsc_early_calib_exact)
		goto calibrated;

	fpu_kern_enter(curthread, NULL, FPU_KERN_NOCTX);
	freq = clockcalib(rdtsc_ordered, "TSC");
	fpu_kern_leave(curthread, NULL);
	tsc_update_freq(freq);

calibrated:
	tc_init(&tsc_timecounter);
	set_cputicker(rdtsc, tsc_freq, !tsc_is_invariant);
}

void
resume_TSC(void)
{
#ifdef SMP
	int quality;

	/* If TSC was not good on boot, it is unlikely to become good now. */
	if (tsc_timecounter.tc_quality < 0)
		return;
	/* Nothing to do with UP. */
	if (mp_ncpus < 2)
		return;

	/*
	 * If TSC was good, a single synchronization should be enough,
	 * but honour smp_tsc_adjust if it's set.
	 */
	quality = test_tsc(MAX(smp_tsc_adjust, 1));
	if (quality != tsc_timecounter.tc_quality) {
		printf("TSC timecounter quality changed: %d -> %d\n",
		    tsc_timecounter.tc_quality, quality);
		tsc_timecounter.tc_quality = quality;
	}
#endif /* SMP */
}

/*
 * When cpufreq levels change, find out about the (new) max frequency.  We
 * use this to update CPU accounting in case it got a lower estimate at boot.
 */
static void
tsc_levels_changed(void *arg, int unit)
{
	device_t cf_dev;
	struct cf_level *levels;
	int count, error;
	uint64_t max_freq;

	/* Only use values from the first CPU, assuming all are equal. */
	if (unit != 0)
		return;

	/* Find the appropriate cpufreq device instance. */
	cf_dev = devclass_get_device(devclass_find("cpufreq"), unit);
	if (cf_dev == NULL) {
		printf("tsc_levels_changed() called but no cpufreq device?\n");
		return;
	}

	/* Get settings from the device and find the max frequency. */
	count = 64;
	levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT);
	if (levels == NULL)
		return;
	error = CPUFREQ_LEVELS(cf_dev, levels, &count);
	if (error == 0 && count != 0) {
		max_freq = (uint64_t)levels[0].total_set.freq * 1000000;
		set_cputicker(rdtsc, max_freq, true);
	} else
		printf("tsc_levels_changed: no max freq found\n");
	free(levels, M_TEMP);
}

/*
 * If the TSC timecounter is in use, veto the pending change.  It may be
 * possible in the future to handle a dynamically-changing timecounter rate.
 */
static void
tsc_freq_changing(void *arg, const struct cf_level *level, int *status)
{

	if (*status != 0 || timecounter != &tsc_timecounter)
		return;

	printf("timecounter TSC must not be in use when "
	    "changing frequencies; change denied\n");
	*status = EBUSY;
}

/* Update TSC freq with the value indicated by the caller. */
static void
tsc_freq_changed(void *arg, const struct cf_level *level, int status)
{
	uint64_t freq;

	/* If there was an error during the transition, don't do anything. */
	if (tsc_disabled || status != 0)
		return;

	/* Total setting for this level gives the new frequency in MHz. */
	freq = (uint64_t)level->total_set.freq * 1000000;
	tsc_update_freq(freq);
}

static int
sysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS)
{
	int error;
	uint64_t freq;

	freq = atomic_load_acq_64(&tsc_freq);
	if (freq == 0)
		return (EOPNOTSUPP);
	error = sysctl_handle_64(oidp, &freq, 0, req);
	if (error == 0 && req->newptr != NULL)
		tsc_update_freq(freq);
	return (error);
}
SYSCTL_PROC(_machdep, OID_AUTO, tsc_freq,
    CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE,
    0, 0, sysctl_machdep_tsc_freq, "QU",
    "Time Stamp Counter frequency");

static u_int
tsc_get_timecount(struct timecounter *tc __unused)
{

	return (rdtsc32());
}

static u_int
tscp_get_timecount(struct timecounter *tc __unused)
{

	return (rdtscp32());
}

static inline u_int
tsc_get_timecount_low(struct timecounter *tc)
{
	uint32_t rv;

	__asm __volatile("rdtsc; shrd %%cl, %%edx, %0"
	    : "=a" (rv) : "c" ((int)(intptr_t)tc->tc_priv) : "edx");
	return (rv);
}

static u_int
tscp_get_timecount_low(struct timecounter *tc)
{
	uint32_t rv;

	__asm __volatile("rdtscp; movl %1, %%ecx; shrd %%cl, %%edx, %0"
	    : "=&a" (rv) : "m" (tc->tc_priv) : "ecx", "edx");
	return (rv);
}

static u_int
tsc_get_timecount_lfence(struct timecounter *tc __unused)
{

	lfence();
	return (rdtsc32());
}

static u_int
tsc_get_timecount_low_lfence(struct timecounter *tc)
{

	lfence();
	return (tsc_get_timecount_low(tc));
}

static u_int
tsc_get_timecount_mfence(struct timecounter *tc __unused)
{

	mfence();
	return (rdtsc32());
}

static u_int
tsc_get_timecount_low_mfence(struct timecounter *tc)
{

	mfence();
	return (tsc_get_timecount_low(tc));
}

static uint32_t
x86_tsc_vdso_timehands(struct vdso_timehands *vdso_th, struct timecounter *tc)
{

	vdso_th->th_algo = VDSO_TH_ALGO_X86_TSC;
	vdso_th->th_x86_shift = (int)(intptr_t)tc->tc_priv;
	vdso_th->th_x86_hpet_idx = 0xffffffff;
	vdso_th->th_x86_pvc_last_systime = 0;
	vdso_th->th_x86_pvc_stable_mask = 0;
	bzero(vdso_th->th_res, sizeof(vdso_th->th_res));
	return (1);
}

#ifdef COMPAT_FREEBSD32
static uint32_t
x86_tsc_vdso_timehands32(struct vdso_timehands32 *vdso_th32,
    struct timecounter *tc)
{

	vdso_th32->th_algo = VDSO_TH_ALGO_X86_TSC;
	vdso_th32->th_x86_shift = (int)(intptr_t)tc->tc_priv;
	vdso_th32->th_x86_hpet_idx = 0xffffffff;
	vdso_th32->th_x86_pvc_last_systime[0] = 0;
	vdso_th32->th_x86_pvc_last_systime[1] = 0;
	vdso_th32->th_x86_pvc_stable_mask = 0;
	bzero(vdso_th32->th_res, sizeof(vdso_th32->th_res));
	return (1);
}
#endif