xref: /freebsd/sys/dev/kvm_clock/kvm_clock.c (revision f49fd63a6a130ae464cdc7756e6f7d0d747c82c4)
16c69c6bbSAdam Fenn /*-
26c69c6bbSAdam Fenn  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
36c69c6bbSAdam Fenn  *
46c69c6bbSAdam Fenn  * Copyright (c) 2014 Bryan Venteicher <bryanv@FreeBSD.org>
56c69c6bbSAdam Fenn  * Copyright (c) 2021 Mathieu Chouquet-Stringer
66c69c6bbSAdam Fenn  * Copyright (c) 2021 Juniper Networks, Inc.
76c69c6bbSAdam Fenn  * Copyright (c) 2021 Klara, Inc.
86c69c6bbSAdam Fenn  *
96c69c6bbSAdam Fenn  * Redistribution and use in source and binary forms, with or without
106c69c6bbSAdam Fenn  * modification, are permitted provided that the following conditions
116c69c6bbSAdam Fenn  * are met:
126c69c6bbSAdam Fenn  * 1. Redistributions of source code must retain the above copyright
136c69c6bbSAdam Fenn  *    notice, this list of conditions and the following disclaimer.
146c69c6bbSAdam Fenn  * 2. Redistributions in binary form must reproduce the above copyright
156c69c6bbSAdam Fenn  *    notice, this list of conditions and the following disclaimer in the
166c69c6bbSAdam Fenn  *    documentation and/or other materials provided with the distribution.
176c69c6bbSAdam Fenn  *
186c69c6bbSAdam Fenn  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
196c69c6bbSAdam Fenn  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
206c69c6bbSAdam Fenn  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
216c69c6bbSAdam Fenn  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
226c69c6bbSAdam Fenn  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
236c69c6bbSAdam Fenn  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
246c69c6bbSAdam Fenn  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
256c69c6bbSAdam Fenn  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
266c69c6bbSAdam Fenn  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
276c69c6bbSAdam Fenn  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
286c69c6bbSAdam Fenn  * SUCH DAMAGE.
296c69c6bbSAdam Fenn  */
306c69c6bbSAdam Fenn 
316c69c6bbSAdam Fenn /*
326c69c6bbSAdam Fenn  * Linux KVM paravirtual clock support
336c69c6bbSAdam Fenn  *
346c69c6bbSAdam Fenn  * References:
356c69c6bbSAdam Fenn  *     - [1] https://www.kernel.org/doc/html/latest/virt/kvm/cpuid.html
366c69c6bbSAdam Fenn  *     - [2] https://www.kernel.org/doc/html/latest/virt/kvm/msr.html
376c69c6bbSAdam Fenn  */
386c69c6bbSAdam Fenn 
396c69c6bbSAdam Fenn #include <sys/cdefs.h>
406c69c6bbSAdam Fenn __FBSDID("$FreeBSD$");
416c69c6bbSAdam Fenn 
426c69c6bbSAdam Fenn #include <sys/param.h>
436c69c6bbSAdam Fenn #include <sys/bus.h>
446c69c6bbSAdam Fenn #include <sys/domainset.h>
456c69c6bbSAdam Fenn #include <sys/kernel.h>
466c69c6bbSAdam Fenn #include <sys/malloc.h>
476c69c6bbSAdam Fenn #include <sys/module.h>
486c69c6bbSAdam Fenn #include <sys/smp.h>
4957a8fa6fSColin Percival #include <sys/sysctl.h>
506c69c6bbSAdam Fenn 
516c69c6bbSAdam Fenn #include <vm/vm.h>
526c69c6bbSAdam Fenn #include <vm/pmap.h>
536c69c6bbSAdam Fenn #include <vm/vm_extern.h>
546c69c6bbSAdam Fenn 
556c69c6bbSAdam Fenn #include <machine/pvclock.h>
566c69c6bbSAdam Fenn #include <x86/kvm.h>
576c69c6bbSAdam Fenn 
586c69c6bbSAdam Fenn #include "clock_if.h"
596c69c6bbSAdam Fenn 
606c69c6bbSAdam Fenn #define	KVM_CLOCK_DEVNAME		"kvmclock"
616c69c6bbSAdam Fenn /*
626c69c6bbSAdam Fenn  * Note: Chosen to be (1) above HPET's value (always 950), (2) above the TSC's
636c69c6bbSAdam Fenn  * default value of 800, and (3) below the TSC's value when it supports the
646c69c6bbSAdam Fenn  * "Invariant TSC" feature and is believed to be synchronized across all CPUs.
656c69c6bbSAdam Fenn  */
666c69c6bbSAdam Fenn #define	KVM_CLOCK_TC_QUALITY		975
676c69c6bbSAdam Fenn 
686c69c6bbSAdam Fenn struct kvm_clock_softc {
696c69c6bbSAdam Fenn 	struct pvclock			 pvc;
706c69c6bbSAdam Fenn 	struct pvclock_wall_clock	 wc;
716c69c6bbSAdam Fenn 	struct pvclock_vcpu_time_info	*timeinfos;
726c69c6bbSAdam Fenn 	u_int				 msr_tc;
736c69c6bbSAdam Fenn 	u_int				 msr_wc;
746c69c6bbSAdam Fenn };
756c69c6bbSAdam Fenn 
766c69c6bbSAdam Fenn static struct pvclock_wall_clock *kvm_clock_get_wallclock(void *arg);
776c69c6bbSAdam Fenn static void	kvm_clock_system_time_enable(struct kvm_clock_softc *sc);
786c69c6bbSAdam Fenn static void	kvm_clock_system_time_enable_pcpu(void *arg);
7957a8fa6fSColin Percival static void	kvm_clock_setup_sysctl(device_t);
806c69c6bbSAdam Fenn 
816c69c6bbSAdam Fenn static struct pvclock_wall_clock *
826c69c6bbSAdam Fenn kvm_clock_get_wallclock(void *arg)
836c69c6bbSAdam Fenn {
846c69c6bbSAdam Fenn 	struct kvm_clock_softc *sc = arg;
856c69c6bbSAdam Fenn 
866c69c6bbSAdam Fenn 	wrmsr(sc->msr_wc, vtophys(&sc->wc));
876c69c6bbSAdam Fenn 	return (&sc->wc);
886c69c6bbSAdam Fenn }
896c69c6bbSAdam Fenn 
906c69c6bbSAdam Fenn static void
916c69c6bbSAdam Fenn kvm_clock_system_time_enable(struct kvm_clock_softc *sc)
926c69c6bbSAdam Fenn {
936c69c6bbSAdam Fenn 	smp_rendezvous(NULL, kvm_clock_system_time_enable_pcpu, NULL, sc);
946c69c6bbSAdam Fenn }
956c69c6bbSAdam Fenn 
966c69c6bbSAdam Fenn static void
976c69c6bbSAdam Fenn kvm_clock_system_time_enable_pcpu(void *arg)
986c69c6bbSAdam Fenn {
996c69c6bbSAdam Fenn 	struct kvm_clock_softc *sc = arg;
1006c69c6bbSAdam Fenn 
1016c69c6bbSAdam Fenn 	/*
1026c69c6bbSAdam Fenn 	 * See [2]; the lsb of this MSR is the system time enable bit.
1036c69c6bbSAdam Fenn 	 */
1046c69c6bbSAdam Fenn 	wrmsr(sc->msr_tc, vtophys(&(sc->timeinfos)[curcpu]) | 1);
1056c69c6bbSAdam Fenn }
1066c69c6bbSAdam Fenn 
1076c69c6bbSAdam Fenn static void
1086c69c6bbSAdam Fenn kvm_clock_identify(driver_t *driver, device_t parent)
1096c69c6bbSAdam Fenn {
1106c69c6bbSAdam Fenn 	u_int regs[4];
1116c69c6bbSAdam Fenn 
1126c69c6bbSAdam Fenn 	kvm_cpuid_get_features(regs);
1136c69c6bbSAdam Fenn 	if ((regs[0] &
1146c69c6bbSAdam Fenn 	    (KVM_FEATURE_CLOCKSOURCE2 | KVM_FEATURE_CLOCKSOURCE)) == 0)
1156c69c6bbSAdam Fenn 		return;
1166c69c6bbSAdam Fenn 	if (device_find_child(parent, KVM_CLOCK_DEVNAME, -1))
1176c69c6bbSAdam Fenn 		return;
1186c69c6bbSAdam Fenn 	BUS_ADD_CHILD(parent, 0, KVM_CLOCK_DEVNAME, 0);
1196c69c6bbSAdam Fenn }
1206c69c6bbSAdam Fenn 
1216c69c6bbSAdam Fenn static int
1226c69c6bbSAdam Fenn kvm_clock_probe(device_t dev)
1236c69c6bbSAdam Fenn {
1246c69c6bbSAdam Fenn 	device_set_desc(dev, "KVM paravirtual clock");
1256c69c6bbSAdam Fenn 	return (BUS_PROBE_DEFAULT);
1266c69c6bbSAdam Fenn }
1276c69c6bbSAdam Fenn 
1286c69c6bbSAdam Fenn static int
1296c69c6bbSAdam Fenn kvm_clock_attach(device_t dev)
1306c69c6bbSAdam Fenn {
1316c69c6bbSAdam Fenn 	u_int regs[4];
1326c69c6bbSAdam Fenn 	struct kvm_clock_softc *sc = device_get_softc(dev);
1336c69c6bbSAdam Fenn 	bool stable_flag_supported;
1346c69c6bbSAdam Fenn 
1356c69c6bbSAdam Fenn 	/* Process KVM "features" CPUID leaf content: */
1366c69c6bbSAdam Fenn 	kvm_cpuid_get_features(regs);
1376c69c6bbSAdam Fenn 	if ((regs[0] & KVM_FEATURE_CLOCKSOURCE2) != 0) {
1386c69c6bbSAdam Fenn 		sc->msr_tc = KVM_MSR_SYSTEM_TIME_NEW;
1396c69c6bbSAdam Fenn 		sc->msr_wc = KVM_MSR_WALL_CLOCK_NEW;
1406c69c6bbSAdam Fenn 	} else {
1416c69c6bbSAdam Fenn 		KASSERT((regs[0] & KVM_FEATURE_CLOCKSOURCE) != 0,
1426c69c6bbSAdam Fenn 		    ("Clocksource feature flags disappeared since "
1436c69c6bbSAdam Fenn 		    "kvm_clock_identify: regs[0] %#0x.", regs[0]));
1446c69c6bbSAdam Fenn 		sc->msr_tc = KVM_MSR_SYSTEM_TIME;
1456c69c6bbSAdam Fenn 		sc->msr_wc = KVM_MSR_WALL_CLOCK;
1466c69c6bbSAdam Fenn 	}
1476c69c6bbSAdam Fenn 	stable_flag_supported =
1486c69c6bbSAdam Fenn 	    (regs[0] & KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) != 0;
1496c69c6bbSAdam Fenn 
1506c69c6bbSAdam Fenn 	/* Set up 'struct pvclock_vcpu_time_info' page(s): */
151*f49fd63aSJohn Baldwin 	sc->timeinfos = kmem_malloc(mp_ncpus *
1526c69c6bbSAdam Fenn 	    sizeof(struct pvclock_vcpu_time_info), M_WAITOK | M_ZERO);
1536c69c6bbSAdam Fenn 	kvm_clock_system_time_enable(sc);
1546c69c6bbSAdam Fenn 
1556c69c6bbSAdam Fenn 	/*
1566c69c6bbSAdam Fenn 	 * Init pvclock; register KVM clock wall clock, register KVM clock
1576c69c6bbSAdam Fenn 	 * timecounter, and set up the requisite infrastructure for vDSO access
1586c69c6bbSAdam Fenn 	 * to this timecounter.
1596c69c6bbSAdam Fenn 	 *     Regarding 'tc_flags': Since the KVM MSR documentation does not
1606c69c6bbSAdam Fenn 	 *     specifically discuss suspend/resume scenarios, conservatively
1616c69c6bbSAdam Fenn 	 *     leave 'TC_FLAGS_SUSPEND_SAFE' cleared and assume that the system
1626c69c6bbSAdam Fenn 	 *     time must be re-inited in such cases.
1636c69c6bbSAdam Fenn 	 */
1646c69c6bbSAdam Fenn 	sc->pvc.get_wallclock = kvm_clock_get_wallclock;
1656c69c6bbSAdam Fenn 	sc->pvc.get_wallclock_arg = sc;
1666c69c6bbSAdam Fenn 	sc->pvc.timeinfos = sc->timeinfos;
1676c69c6bbSAdam Fenn 	sc->pvc.stable_flag_supported = stable_flag_supported;
1686c69c6bbSAdam Fenn 	pvclock_init(&sc->pvc, dev, KVM_CLOCK_DEVNAME, KVM_CLOCK_TC_QUALITY, 0);
16957a8fa6fSColin Percival 	kvm_clock_setup_sysctl(dev);
1706c69c6bbSAdam Fenn 	return (0);
1716c69c6bbSAdam Fenn }
1726c69c6bbSAdam Fenn 
1736c69c6bbSAdam Fenn static int
1746c69c6bbSAdam Fenn kvm_clock_detach(device_t dev)
1756c69c6bbSAdam Fenn {
1766c69c6bbSAdam Fenn 	struct kvm_clock_softc *sc = device_get_softc(dev);
1776c69c6bbSAdam Fenn 
1786c69c6bbSAdam Fenn 	return (pvclock_destroy(&sc->pvc));
1796c69c6bbSAdam Fenn }
1806c69c6bbSAdam Fenn 
1816c69c6bbSAdam Fenn static int
1826c69c6bbSAdam Fenn kvm_clock_suspend(device_t dev)
1836c69c6bbSAdam Fenn {
1846c69c6bbSAdam Fenn 	return (0);
1856c69c6bbSAdam Fenn }
1866c69c6bbSAdam Fenn 
1876c69c6bbSAdam Fenn static int
1886c69c6bbSAdam Fenn kvm_clock_resume(device_t dev)
1896c69c6bbSAdam Fenn {
1906c69c6bbSAdam Fenn 	/*
1916c69c6bbSAdam Fenn 	 * See note in 'kvm_clock_attach()' regarding 'TC_FLAGS_SUSPEND_SAFE';
1926c69c6bbSAdam Fenn 	 * conservatively assume that the system time must be re-inited in
1936c69c6bbSAdam Fenn 	 * suspend/resume scenarios.
1946c69c6bbSAdam Fenn 	 */
1956c69c6bbSAdam Fenn 	kvm_clock_system_time_enable(device_get_softc(dev));
1966c69c6bbSAdam Fenn 	pvclock_resume();
1976c69c6bbSAdam Fenn 	inittodr(time_second);
1986c69c6bbSAdam Fenn 	return (0);
1996c69c6bbSAdam Fenn }
2006c69c6bbSAdam Fenn 
2016c69c6bbSAdam Fenn static int
2026c69c6bbSAdam Fenn kvm_clock_gettime(device_t dev, struct timespec *ts)
2036c69c6bbSAdam Fenn {
2046c69c6bbSAdam Fenn 	struct kvm_clock_softc *sc = device_get_softc(dev);
2056c69c6bbSAdam Fenn 
2066c69c6bbSAdam Fenn 	pvclock_gettime(&sc->pvc, ts);
2076c69c6bbSAdam Fenn 	return (0);
2086c69c6bbSAdam Fenn }
2096c69c6bbSAdam Fenn 
2106c69c6bbSAdam Fenn static int
2116c69c6bbSAdam Fenn kvm_clock_settime(device_t dev, struct timespec *ts)
2126c69c6bbSAdam Fenn {
2136c69c6bbSAdam Fenn 	/*
2146c69c6bbSAdam Fenn 	 * Even though it is not possible to set the KVM clock's wall clock, to
2156c69c6bbSAdam Fenn 	 * avoid the possibility of periodic benign error messages from
2166c69c6bbSAdam Fenn 	 * 'settime_task_func()', report success rather than, e.g., 'ENODEV'.
2176c69c6bbSAdam Fenn 	 */
2186c69c6bbSAdam Fenn 	return (0);
2196c69c6bbSAdam Fenn }
2206c69c6bbSAdam Fenn 
22157a8fa6fSColin Percival static int
22257a8fa6fSColin Percival kvm_clock_tsc_freq_sysctl(SYSCTL_HANDLER_ARGS)
22357a8fa6fSColin Percival {
22457a8fa6fSColin Percival 	struct kvm_clock_softc *sc = oidp->oid_arg1;
22557a8fa6fSColin Percival         uint64_t freq = pvclock_tsc_freq(sc->timeinfos);
22657a8fa6fSColin Percival 
22757a8fa6fSColin Percival         return (sysctl_handle_64(oidp, &freq, 0, req));
22857a8fa6fSColin Percival }
22957a8fa6fSColin Percival 
23057a8fa6fSColin Percival static void
23157a8fa6fSColin Percival kvm_clock_setup_sysctl(device_t dev)
23257a8fa6fSColin Percival {
23357a8fa6fSColin Percival 	struct kvm_clock_softc *sc = device_get_softc(dev);
23457a8fa6fSColin Percival         struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev);
23557a8fa6fSColin Percival         struct sysctl_oid *tree = device_get_sysctl_tree(dev);
23657a8fa6fSColin Percival         struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree);
23757a8fa6fSColin Percival 
23857a8fa6fSColin Percival         SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tsc_freq",
23957a8fa6fSColin Percival             CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
24057a8fa6fSColin Percival             kvm_clock_tsc_freq_sysctl, "QU",
24157a8fa6fSColin Percival             "Time Stamp Counter frequency");
24257a8fa6fSColin Percival }
24357a8fa6fSColin Percival 
2446c69c6bbSAdam Fenn static device_method_t kvm_clock_methods[] = {
2456c69c6bbSAdam Fenn 	DEVMETHOD(device_identify,	kvm_clock_identify),
2466c69c6bbSAdam Fenn 	DEVMETHOD(device_probe,		kvm_clock_probe),
2476c69c6bbSAdam Fenn 	DEVMETHOD(device_attach,	kvm_clock_attach),
2486c69c6bbSAdam Fenn 	DEVMETHOD(device_detach,	kvm_clock_detach),
2496c69c6bbSAdam Fenn 	DEVMETHOD(device_suspend,	kvm_clock_suspend),
2506c69c6bbSAdam Fenn 	DEVMETHOD(device_resume,	kvm_clock_resume),
2516c69c6bbSAdam Fenn 	/* clock interface */
2526c69c6bbSAdam Fenn 	DEVMETHOD(clock_gettime,	kvm_clock_gettime),
2536c69c6bbSAdam Fenn 	DEVMETHOD(clock_settime,	kvm_clock_settime),
2546c69c6bbSAdam Fenn 
2556c69c6bbSAdam Fenn 	DEVMETHOD_END
2566c69c6bbSAdam Fenn };
2576c69c6bbSAdam Fenn 
2586c69c6bbSAdam Fenn static driver_t kvm_clock_driver = {
2596c69c6bbSAdam Fenn 	KVM_CLOCK_DEVNAME,
2606c69c6bbSAdam Fenn 	kvm_clock_methods,
2616c69c6bbSAdam Fenn 	sizeof(struct kvm_clock_softc),
2626c69c6bbSAdam Fenn };
2636c69c6bbSAdam Fenn 
26448b6e01dSJohn Baldwin DRIVER_MODULE(kvm_clock, nexus, kvm_clock_driver, 0, 0);
265