@@ -7231,6 +7231,7 @@ static Property x86_cpu_properties[] = {
false),
DEFINE_PROP_BOOL("x-intel-pt-auto-level", X86CPU, intel_pt_auto_level,
true),
+ DEFINE_PROP_BOOL("x-precise-tsc", X86CPU, precise_tsc, false),
DEFINE_PROP_END_OF_LIST()
};
@@ -1460,6 +1460,7 @@ typedef struct CPUX86State {
uint64_t tsc_adjust;
uint64_t tsc_deadline;
uint64_t tsc_aux;
+ uint64_t tsc_ns_timestamp;
uint64_t xcr0;
@@ -1743,6 +1744,9 @@ struct X86CPU {
/* Number of physical address bits supported */
uint32_t phys_bits;
+ /* Use KVM_GET_TSC_PRECISE/KVM_SET_TSC_PRECISE to access IA32_TSC */
+ bool precise_tsc;
+
/* in order to simplify APIC support, we leave this pointer to the
user */
struct DeviceState *apic_state;
@@ -121,6 +121,7 @@ static int has_xsave;
static int has_xcrs;
static int has_pit_state2;
static int has_exception_payload;
+static int has_precise_tsc;
static bool has_msr_mcg_ext_ctl;
@@ -196,31 +197,109 @@ static int kvm_get_tsc(CPUState *cs)
{
X86CPU *cpu = X86_CPU(cs);
CPUX86State *env = &cpu->env;
- struct {
- struct kvm_msrs info;
- struct kvm_msr_entry entries[1];
- } msr_data = {};
int ret;
if (env->tsc_valid) {
return 0;
}
- memset(&msr_data, 0, sizeof(msr_data));
- msr_data.info.nmsrs = 1;
- msr_data.entries[0].index = MSR_IA32_TSC;
- env->tsc_valid = !runstate_is_running();
+ if (cpu->precise_tsc) {
+ struct kvm_tsc_state tsc_state;
- ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
- if (ret < 0) {
- return ret;
+ ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_TSC_STATE, &tsc_state);
+ if (ret < 0) {
+ return ret;
+ }
+
+ env->tsc = tsc_state.tsc;
+ if (tsc_state.flags & KVM_TSC_STATE_TSC_ADJUST_VALID) {
+ env->tsc_adjust = tsc_state.tsc_adjust;
+ }
+ env->tsc_ns_timestamp = tsc_state.nsec;
+
+ } else {
+ struct {
+ struct kvm_msrs info;
+ struct kvm_msr_entry entries[2];
+ } msr_data = {};
+
+ memset(&msr_data, 0, sizeof(msr_data));
+ msr_data.info.nmsrs = 1;
+ msr_data.entries[0].index = MSR_IA32_TSC;
+
+ if (has_msr_tsc_adjust) {
+ msr_data.info.nmsrs++;
+ msr_data.entries[1].index = MSR_TSC_ADJUST;
+ }
+
+ ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MSRS, &msr_data);
+ if (ret < 0) {
+ return ret;
+ }
+
+ assert(ret == msr_data.info.nmsrs);
+
+ env->tsc = msr_data.entries[0].data;
+ if (has_msr_tsc_adjust) {
+ env->tsc_adjust = msr_data.entries[1].data;
+ }
}
- assert(ret == 1);
- env->tsc = msr_data.entries[0].data;
+ env->tsc_valid = !runstate_is_running();
return 0;
}
+static int kvm_set_tsc(CPUState *cs)
+{
+ int ret;
+ X86CPU *cpu = X86_CPU(cs);
+ CPUX86State *env = &cpu->env;
+
+ if (cpu->precise_tsc) {
+ struct kvm_tsc_state tsc_state;
+
+ memset(&tsc_state, 0, sizeof(tsc_state));
+
+ tsc_state.tsc = env->tsc;
+ tsc_state.nsec = env->tsc_ns_timestamp;
+
+ if (has_msr_tsc_adjust) {
+ tsc_state.tsc_adjust = env->tsc_adjust;
+ tsc_state.flags |= KVM_TSC_STATE_TSC_ADJUST_VALID;
+ }
+
+ ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_TSC_STATE, &tsc_state);
+ if (ret < 0) {
+ return ret;
+ }
+
+ } else {
+ struct {
+ struct kvm_msrs info;
+ struct kvm_msr_entry entries[2];
+ } msr_data = {};
+
+ memset(&msr_data, 0, sizeof(msr_data));
+ msr_data.info.nmsrs = 1;
+ msr_data.entries[0].index = MSR_IA32_TSC;
+ msr_data.entries[0].data = env->tsc;
+
+ if (has_msr_tsc_adjust) {
+ msr_data.info.nmsrs++;
+ msr_data.entries[1].index = MSR_TSC_ADJUST;
+ msr_data.entries[1].data = env->tsc_adjust;
+ }
+
+ ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MSRS, &msr_data);
+ if (ret < 0) {
+ return ret;
+ }
+
+ assert(ret == msr_data.info.nmsrs);
+ }
+ return ret;
+}
+
static inline void do_kvm_synchronize_tsc(CPUState *cpu, run_on_cpu_data arg)
{
kvm_get_tsc(cpu);
@@ -1780,6 +1859,13 @@ int kvm_arch_init_vcpu(CPUState *cs)
}
}
+ if (cpu->precise_tsc) {
+ if (!kvm_check_extension(cs->kvm_state, KVM_CAP_PRECISE_TSC)) {
+ error_report("kvm: Precise TSC is not supported by the host's KVM");
+ return -ENOTSUP;
+ }
+ }
+
if (cpu->vmware_cpuid_freq
/* Guests depend on 0x40000000 to detect this feature, so only expose
* it if KVM exposes leaf 0x40000000. (Conflicts with Hyper-V) */
@@ -2196,6 +2282,8 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
int disable_exits = kvm_check_extension(s, KVM_CAP_X86_DISABLE_EXITS);
int ret;
+
+
/* Work around for kernel header with a typo. TODO: fix header and drop. */
#if defined(KVM_X86_DISABLE_EXITS_HTL) && !defined(KVM_X86_DISABLE_EXITS_HLT)
#define KVM_X86_DISABLE_EXITS_HLT KVM_X86_DISABLE_EXITS_HTL
@@ -2215,6 +2303,8 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
}
}
+ has_precise_tsc = kvm_check_extension(s, KVM_CAP_PRECISE_TSC);
+
return 0;
}
@@ -2756,9 +2846,6 @@ static int kvm_put_msrs(X86CPU *cpu, int level)
if (has_msr_tsc_aux) {
kvm_msr_entry_add(cpu, MSR_TSC_AUX, env->tsc_aux);
}
- if (has_msr_tsc_adjust) {
- kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, env->tsc_adjust);
- }
if (has_msr_misc_enable) {
kvm_msr_entry_add(cpu, MSR_IA32_MISC_ENABLE,
env->msr_ia32_misc_enable);
@@ -2802,7 +2889,6 @@ static int kvm_put_msrs(X86CPU *cpu, int level)
* for normal writeback. Limit them to reset or full state updates.
*/
if (level >= KVM_PUT_RESET_STATE) {
- kvm_msr_entry_add(cpu, MSR_IA32_TSC, env->tsc);
kvm_msr_entry_add(cpu, MSR_KVM_SYSTEM_TIME, env->system_time_msr);
kvm_msr_entry_add(cpu, MSR_KVM_WALL_CLOCK, env->wall_clock_msr);
if (env->features[FEAT_KVM] & (1 << KVM_FEATURE_ASYNC_PF_INT)) {
@@ -3142,9 +3228,6 @@ static int kvm_get_msrs(X86CPU *cpu)
if (has_msr_tsc_aux) {
kvm_msr_entry_add(cpu, MSR_TSC_AUX, 0);
}
- if (has_msr_tsc_adjust) {
- kvm_msr_entry_add(cpu, MSR_TSC_ADJUST, 0);
- }
if (has_msr_tsc_deadline) {
kvm_msr_entry_add(cpu, MSR_IA32_TSCDEADLINE, 0);
}
@@ -3178,10 +3261,6 @@ static int kvm_get_msrs(X86CPU *cpu)
if (has_msr_virt_ssbd) {
kvm_msr_entry_add(cpu, MSR_VIRT_SSBD, 0);
}
- if (!env->tsc_valid) {
- kvm_msr_entry_add(cpu, MSR_IA32_TSC, 0);
- env->tsc_valid = !runstate_is_running();
- }
#ifdef TARGET_X86_64
if (lm_capable_kernel) {
@@ -3385,9 +3464,6 @@ static int kvm_get_msrs(X86CPU *cpu)
case MSR_TSC_AUX:
env->tsc_aux = msrs[i].data;
break;
- case MSR_TSC_ADJUST:
- env->tsc_adjust = msrs[i].data;
- break;
case MSR_IA32_TSCDEADLINE:
env->tsc_deadline = msrs[i].data;
break;
@@ -3995,6 +4071,11 @@ int kvm_arch_put_registers(CPUState *cpu, int level)
if (ret < 0) {
return ret;
}
+
+ ret = kvm_set_tsc(cpu);
+ if (ret < 0) {
+ return ret;
+ }
}
ret = kvm_put_tscdeadline_msr(x86_cpu);
@@ -4064,6 +4145,12 @@ int kvm_arch_get_registers(CPUState *cs)
if (ret < 0) {
goto out;
}
+
+ ret = kvm_get_tsc(cs);
+ if (ret < 0) {
+ goto out;
+ }
+
ret = 0;
out:
cpu_sync_bndcs_hflags(&cpu->env);
@@ -1359,6 +1359,25 @@ static const VMStateDescription vmstate_msr_tsx_ctrl = {
}
};
+
+static bool tsc_info_needed(void *opaque)
+{
+ X86CPU *cpu = opaque;
+ return cpu->precise_tsc;
+}
+
+static const VMStateDescription vmstate_tsc_info = {
+ .name = "cpu/tsc_nsec_info",
+ .version_id = 1,
+ .minimum_version_id = 1,
+ .needed = tsc_info_needed,
+ .fields = (VMStateField[]) {
+ VMSTATE_UINT64(env.tsc_ns_timestamp, X86CPU),
+ VMSTATE_END_OF_LIST()
+ }
+};
+
+
VMStateDescription vmstate_x86_cpu = {
.name = "cpu",
.version_id = 12,
@@ -1493,6 +1512,7 @@ VMStateDescription vmstate_x86_cpu = {
#endif
#ifdef CONFIG_KVM
&vmstate_nested_state,
+ &vmstate_tsc_info,
#endif
&vmstate_msr_tsx_ctrl,
NULL
Currently to enable it, you need to set x-precise-tsc=on for each vcpu. Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com> --- target/i386/cpu.c | 1 + target/i386/cpu.h | 4 ++ target/i386/kvm.c | 141 ++++++++++++++++++++++++++++++++++-------- target/i386/machine.c | 20 ++++++ 4 files changed, 139 insertions(+), 27 deletions(-)