1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2014 Bryan Venteicher <bryanv@FreeBSD.org> 5 * Copyright (c) 2021 Mathieu Chouquet-Stringer 6 * Copyright (c) 2021 Juniper Networks, Inc. 7 * Copyright (c) 2021 Klara, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 */ 30 31 /* 32 * Linux KVM paravirtual clock support 33 * 34 * References: 35 * - [1] https://www.kernel.org/doc/html/latest/virt/kvm/cpuid.html 36 * - [2] https://www.kernel.org/doc/html/latest/virt/kvm/msr.html 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include <sys/param.h> 43 #include <sys/bus.h> 44 #include <sys/domainset.h> 45 #include <sys/kernel.h> 46 #include <sys/malloc.h> 47 #include <sys/module.h> 48 #include <sys/smp.h> 49 #include <sys/sysctl.h> 50 51 #include <vm/vm.h> 52 #include <vm/pmap.h> 53 #include <vm/vm_extern.h> 54 55 #include <machine/pvclock.h> 56 #include <x86/kvm.h> 57 58 #include "clock_if.h" 59 60 #define KVM_CLOCK_DEVNAME "kvmclock" 61 /* 62 * Note: Chosen to be (1) above HPET's value (always 950), (2) above the TSC's 63 * default value of 800, and (3) below the TSC's value when it supports the 64 * "Invariant TSC" feature and is believed to be synchronized across all CPUs. 65 */ 66 #define KVM_CLOCK_TC_QUALITY 975 67 68 struct kvm_clock_softc { 69 struct pvclock pvc; 70 struct pvclock_wall_clock wc; 71 struct pvclock_vcpu_time_info *timeinfos; 72 u_int msr_tc; 73 u_int msr_wc; 74 #ifndef EARLY_AP_STARTUP 75 int firstcpu; 76 #endif 77 }; 78 79 static struct pvclock_wall_clock *kvm_clock_get_wallclock(void *arg); 80 static void kvm_clock_system_time_enable(struct kvm_clock_softc *sc, 81 const cpuset_t *cpus); 82 static void kvm_clock_system_time_enable_pcpu(void *arg); 83 static void kvm_clock_setup_sysctl(device_t); 84 85 static struct pvclock_wall_clock * 86 kvm_clock_get_wallclock(void *arg) 87 { 88 struct kvm_clock_softc *sc = arg; 89 90 wrmsr(sc->msr_wc, vtophys(&sc->wc)); 91 return (&sc->wc); 92 } 93 94 static void 95 kvm_clock_system_time_enable(struct kvm_clock_softc *sc, const cpuset_t *cpus) 96 { 97 smp_rendezvous_cpus(*cpus, NULL, kvm_clock_system_time_enable_pcpu, 98 NULL, sc); 99 } 100 101 static void 102 kvm_clock_system_time_enable_pcpu(void *arg) 103 { 104 struct kvm_clock_softc *sc = arg; 105 106 /* 107 * See [2]; the lsb of this MSR is the system time enable bit. 108 */ 109 wrmsr(sc->msr_tc, vtophys(&(sc->timeinfos)[curcpu]) | 1); 110 } 111 112 #ifndef EARLY_AP_STARTUP 113 static void 114 kvm_clock_init_smp(void *arg __unused) 115 { 116 devclass_t kvm_clock_devclass; 117 cpuset_t cpus; 118 struct kvm_clock_softc *sc; 119 120 kvm_clock_devclass = devclass_find(KVM_CLOCK_DEVNAME); 121 sc = devclass_get_softc(kvm_clock_devclass, 0); 122 if (sc == NULL || mp_ncpus == 1) 123 return; 124 125 /* 126 * Register with the hypervisor on all CPUs except the one that 127 * registered in kvm_clock_attach(). 128 */ 129 cpus = all_cpus; 130 KASSERT(CPU_ISSET(sc->firstcpu, &cpus), 131 ("%s: invalid first CPU %d", __func__, sc->firstcpu)); 132 CPU_CLR(sc->firstcpu, &cpus); 133 kvm_clock_system_time_enable(sc, &cpus); 134 } 135 SYSINIT(kvm_clock, SI_SUB_SMP, SI_ORDER_ANY, kvm_clock_init_smp, NULL); 136 #endif 137 138 static void 139 kvm_clock_identify(driver_t *driver, device_t parent) 140 { 141 u_int regs[4]; 142 143 kvm_cpuid_get_features(regs); 144 if ((regs[0] & 145 (KVM_FEATURE_CLOCKSOURCE2 | KVM_FEATURE_CLOCKSOURCE)) == 0) 146 return; 147 if (device_find_child(parent, KVM_CLOCK_DEVNAME, -1)) 148 return; 149 BUS_ADD_CHILD(parent, 0, KVM_CLOCK_DEVNAME, 0); 150 } 151 152 static int 153 kvm_clock_probe(device_t dev) 154 { 155 device_set_desc(dev, "KVM paravirtual clock"); 156 return (BUS_PROBE_DEFAULT); 157 } 158 159 static int 160 kvm_clock_attach(device_t dev) 161 { 162 u_int regs[4]; 163 struct kvm_clock_softc *sc = device_get_softc(dev); 164 bool stable_flag_supported; 165 166 /* Process KVM "features" CPUID leaf content: */ 167 kvm_cpuid_get_features(regs); 168 if ((regs[0] & KVM_FEATURE_CLOCKSOURCE2) != 0) { 169 sc->msr_tc = KVM_MSR_SYSTEM_TIME_NEW; 170 sc->msr_wc = KVM_MSR_WALL_CLOCK_NEW; 171 } else { 172 KASSERT((regs[0] & KVM_FEATURE_CLOCKSOURCE) != 0, 173 ("Clocksource feature flags disappeared since " 174 "kvm_clock_identify: regs[0] %#0x.", regs[0])); 175 sc->msr_tc = KVM_MSR_SYSTEM_TIME; 176 sc->msr_wc = KVM_MSR_WALL_CLOCK; 177 } 178 stable_flag_supported = 179 (regs[0] & KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) != 0; 180 181 /* Set up 'struct pvclock_vcpu_time_info' page(s): */ 182 sc->timeinfos = kmem_malloc(mp_ncpus * 183 sizeof(struct pvclock_vcpu_time_info), M_WAITOK | M_ZERO); 184 #ifdef EARLY_AP_STARTUP 185 kvm_clock_system_time_enable(sc, &all_cpus); 186 #else 187 sc->firstcpu = curcpu; 188 kvm_clock_system_time_enable_pcpu(sc); 189 #endif 190 191 /* 192 * Init pvclock; register KVM clock wall clock, register KVM clock 193 * timecounter, and set up the requisite infrastructure for vDSO access 194 * to this timecounter. 195 * Regarding 'tc_flags': Since the KVM MSR documentation does not 196 * specifically discuss suspend/resume scenarios, conservatively 197 * leave 'TC_FLAGS_SUSPEND_SAFE' cleared and assume that the system 198 * time must be re-inited in such cases. 199 */ 200 sc->pvc.get_wallclock = kvm_clock_get_wallclock; 201 sc->pvc.get_wallclock_arg = sc; 202 sc->pvc.timeinfos = sc->timeinfos; 203 sc->pvc.stable_flag_supported = stable_flag_supported; 204 pvclock_init(&sc->pvc, dev, KVM_CLOCK_DEVNAME, KVM_CLOCK_TC_QUALITY, 0); 205 kvm_clock_setup_sysctl(dev); 206 return (0); 207 } 208 209 static int 210 kvm_clock_detach(device_t dev) 211 { 212 struct kvm_clock_softc *sc = device_get_softc(dev); 213 214 return (pvclock_destroy(&sc->pvc)); 215 } 216 217 static int 218 kvm_clock_suspend(device_t dev) 219 { 220 return (0); 221 } 222 223 static int 224 kvm_clock_resume(device_t dev) 225 { 226 /* 227 * See note in 'kvm_clock_attach()' regarding 'TC_FLAGS_SUSPEND_SAFE'; 228 * conservatively assume that the system time must be re-inited in 229 * suspend/resume scenarios. 230 */ 231 kvm_clock_system_time_enable(device_get_softc(dev), &all_cpus); 232 pvclock_resume(); 233 inittodr(time_second); 234 return (0); 235 } 236 237 static int 238 kvm_clock_gettime(device_t dev, struct timespec *ts) 239 { 240 struct kvm_clock_softc *sc = device_get_softc(dev); 241 242 pvclock_gettime(&sc->pvc, ts); 243 return (0); 244 } 245 246 static int 247 kvm_clock_settime(device_t dev, struct timespec *ts) 248 { 249 /* 250 * Even though it is not possible to set the KVM clock's wall clock, to 251 * avoid the possibility of periodic benign error messages from 252 * 'settime_task_func()', report success rather than, e.g., 'ENODEV'. 253 */ 254 return (0); 255 } 256 257 static int 258 kvm_clock_tsc_freq_sysctl(SYSCTL_HANDLER_ARGS) 259 { 260 struct kvm_clock_softc *sc = oidp->oid_arg1; 261 uint64_t freq = pvclock_tsc_freq(sc->timeinfos); 262 263 return (sysctl_handle_64(oidp, &freq, 0, req)); 264 } 265 266 static void 267 kvm_clock_setup_sysctl(device_t dev) 268 { 269 struct kvm_clock_softc *sc = device_get_softc(dev); 270 struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev); 271 struct sysctl_oid *tree = device_get_sysctl_tree(dev); 272 struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree); 273 274 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tsc_freq", 275 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0, 276 kvm_clock_tsc_freq_sysctl, "QU", 277 "Time Stamp Counter frequency"); 278 } 279 280 static device_method_t kvm_clock_methods[] = { 281 DEVMETHOD(device_identify, kvm_clock_identify), 282 DEVMETHOD(device_probe, kvm_clock_probe), 283 DEVMETHOD(device_attach, kvm_clock_attach), 284 DEVMETHOD(device_detach, kvm_clock_detach), 285 DEVMETHOD(device_suspend, kvm_clock_suspend), 286 DEVMETHOD(device_resume, kvm_clock_resume), 287 /* clock interface */ 288 DEVMETHOD(clock_gettime, kvm_clock_gettime), 289 DEVMETHOD(clock_settime, kvm_clock_settime), 290 291 DEVMETHOD_END 292 }; 293 294 static driver_t kvm_clock_driver = { 295 KVM_CLOCK_DEVNAME, 296 kvm_clock_methods, 297 sizeof(struct kvm_clock_softc), 298 }; 299 300 DRIVER_MODULE(kvm_clock, nexus, kvm_clock_driver, 0, 0); 301