]> git.proxmox.com Git - mirror_qemu.git/blob - hw/i386/kvm/clock.c
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
[mirror_qemu.git] / hw / i386 / kvm / clock.c
1 /*
2 * QEMU KVM support, paravirtual clock device
3 *
4 * Copyright (C) 2011 Siemens AG
5 *
6 * Authors:
7 * Jan Kiszka <jan.kiszka@siemens.com>
8 *
9 * This work is licensed under the terms of the GNU GPL version 2.
10 * See the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu-common.h"
17 #include "qemu/host-utils.h"
18 #include "sysemu/sysemu.h"
19 #include "sysemu/kvm.h"
20 #include "sysemu/cpus.h"
21 #include "hw/sysbus.h"
22 #include "hw/kvm/clock.h"
23
24 #include <linux/kvm.h>
25 #include <linux/kvm_para.h>
26
27 #define TYPE_KVM_CLOCK "kvmclock"
28 #define KVM_CLOCK(obj) OBJECT_CHECK(KVMClockState, (obj), TYPE_KVM_CLOCK)
29
30 typedef struct KVMClockState {
31 /*< private >*/
32 SysBusDevice busdev;
33 /*< public >*/
34
35 uint64_t clock;
36 bool clock_valid;
37 } KVMClockState;
38
39 struct pvclock_vcpu_time_info {
40 uint32_t version;
41 uint32_t pad0;
42 uint64_t tsc_timestamp;
43 uint64_t system_time;
44 uint32_t tsc_to_system_mul;
45 int8_t tsc_shift;
46 uint8_t flags;
47 uint8_t pad[2];
48 } __attribute__((__packed__)); /* 32 bytes */
49
50 static uint64_t kvmclock_current_nsec(KVMClockState *s)
51 {
52 CPUState *cpu = first_cpu;
53 CPUX86State *env = cpu->env_ptr;
54 hwaddr kvmclock_struct_pa = env->system_time_msr & ~1ULL;
55 uint64_t migration_tsc = env->tsc;
56 struct pvclock_vcpu_time_info time;
57 uint64_t delta;
58 uint64_t nsec_lo;
59 uint64_t nsec_hi;
60 uint64_t nsec;
61
62 if (!(env->system_time_msr & 1ULL)) {
63 /* KVM clock not active */
64 return 0;
65 }
66
67 cpu_physical_memory_read(kvmclock_struct_pa, &time, sizeof(time));
68
69 assert(time.tsc_timestamp <= migration_tsc);
70 delta = migration_tsc - time.tsc_timestamp;
71 if (time.tsc_shift < 0) {
72 delta >>= -time.tsc_shift;
73 } else {
74 delta <<= time.tsc_shift;
75 }
76
77 mulu64(&nsec_lo, &nsec_hi, delta, time.tsc_to_system_mul);
78 nsec = (nsec_lo >> 32) | (nsec_hi << 32);
79 return nsec + time.system_time;
80 }
81
82 static void kvmclock_vm_state_change(void *opaque, int running,
83 RunState state)
84 {
85 KVMClockState *s = opaque;
86 CPUState *cpu;
87 int cap_clock_ctrl = kvm_check_extension(kvm_state, KVM_CAP_KVMCLOCK_CTRL);
88 int ret;
89
90 if (running) {
91 struct kvm_clock_data data = {};
92 uint64_t time_at_migration = kvmclock_current_nsec(s);
93
94 s->clock_valid = false;
95
96 /* We can't rely on the migrated clock value, just discard it */
97 if (time_at_migration) {
98 s->clock = time_at_migration;
99 }
100
101 data.clock = s->clock;
102 ret = kvm_vm_ioctl(kvm_state, KVM_SET_CLOCK, &data);
103 if (ret < 0) {
104 fprintf(stderr, "KVM_SET_CLOCK failed: %s\n", strerror(ret));
105 abort();
106 }
107
108 if (!cap_clock_ctrl) {
109 return;
110 }
111 CPU_FOREACH(cpu) {
112 ret = kvm_vcpu_ioctl(cpu, KVM_KVMCLOCK_CTRL, 0);
113 if (ret) {
114 if (ret != -EINVAL) {
115 fprintf(stderr, "%s: %s\n", __func__, strerror(-ret));
116 }
117 return;
118 }
119 }
120 } else {
121 struct kvm_clock_data data;
122 int ret;
123
124 if (s->clock_valid) {
125 return;
126 }
127
128 cpu_synchronize_all_states();
129 /* In theory, the cpu_synchronize_all_states() call above wouldn't
130 * affect the rest of the code, as the VCPU state inside CPUState
131 * is supposed to always match the VCPU state on the kernel side.
132 *
133 * In practice, calling cpu_synchronize_state() too soon will load the
134 * kernel-side APIC state into X86CPU.apic_state too early, APIC state
135 * won't be reloaded later because CPUState.vcpu_dirty==true, and
136 * outdated APIC state may be migrated to another host.
137 *
138 * The real fix would be to make sure outdated APIC state is read
139 * from the kernel again when necessary. While this is not fixed, we
140 * need the cpu_clean_all_dirty() call below.
141 */
142 cpu_clean_all_dirty();
143
144 ret = kvm_vm_ioctl(kvm_state, KVM_GET_CLOCK, &data);
145 if (ret < 0) {
146 fprintf(stderr, "KVM_GET_CLOCK failed: %s\n", strerror(ret));
147 abort();
148 }
149 s->clock = data.clock;
150
151 /*
152 * If the VM is stopped, declare the clock state valid to
153 * avoid re-reading it on next vmsave (which would return
154 * a different value). Will be reset when the VM is continued.
155 */
156 s->clock_valid = true;
157 }
158 }
159
160 static void kvmclock_realize(DeviceState *dev, Error **errp)
161 {
162 KVMClockState *s = KVM_CLOCK(dev);
163
164 qemu_add_vm_change_state_handler(kvmclock_vm_state_change, s);
165 }
166
167 static const VMStateDescription kvmclock_vmsd = {
168 .name = "kvmclock",
169 .version_id = 1,
170 .minimum_version_id = 1,
171 .fields = (VMStateField[]) {
172 VMSTATE_UINT64(clock, KVMClockState),
173 VMSTATE_END_OF_LIST()
174 }
175 };
176
177 static void kvmclock_class_init(ObjectClass *klass, void *data)
178 {
179 DeviceClass *dc = DEVICE_CLASS(klass);
180
181 dc->realize = kvmclock_realize;
182 dc->vmsd = &kvmclock_vmsd;
183 }
184
185 static const TypeInfo kvmclock_info = {
186 .name = TYPE_KVM_CLOCK,
187 .parent = TYPE_SYS_BUS_DEVICE,
188 .instance_size = sizeof(KVMClockState),
189 .class_init = kvmclock_class_init,
190 };
191
192 /* Note: Must be called after VCPU initialization. */
193 void kvmclock_create(void)
194 {
195 X86CPU *cpu = X86_CPU(first_cpu);
196
197 if (kvm_enabled() &&
198 cpu->env.features[FEAT_KVM] & ((1ULL << KVM_FEATURE_CLOCKSOURCE) |
199 (1ULL << KVM_FEATURE_CLOCKSOURCE2))) {
200 sysbus_create_simple(TYPE_KVM_CLOCK, -1, NULL);
201 }
202 }
203
204 static void kvmclock_register_types(void)
205 {
206 type_register_static(&kvmclock_info);
207 }
208
209 type_init(kvmclock_register_types)