2019-01-28 17:52:51

by Anton Kuchin

[permalink] [raw]
Subject: [PATCH] KVM: Ignore LBR MSRs with no effect

Win10 attempts to save these registers during KiSaveDebugRegisterState
if LBR or BTF bits are set in MSR_IA32_DEBUGCTLMSR. It uses DR7 GE and LE
flags for per-thread switching of these these features so zero value that
is returned for MSR_IA32_DEBUGCTLMSR has no effect.

These registers are used for debugging and shouldn't cause #GP and
guest crash so just return zeroes just like we do for common x86 LBR
MSRs (DEBUGCTLMSR, LAST[BRANCH|INT][TO|FROM]IP).

Signed-off-by: Anton Kuchin <[email protected]>
---
arch/x86/kvm/vmx/vmx.c | 7 +++++++
1 file changed, 7 insertions(+)

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index f6915f10e584..8bc56cf027ed 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -1769,6 +1769,13 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
else
msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
break;
+ case MSR_LBR_TOS:
+ case MSR_LBR_NHM_FROM ... MSR_LBR_NHM_FROM + 31:
+ case MSR_LBR_NHM_TO ... MSR_LBR_NHM_TO + 31:
+ case MSR_LBR_CORE_FROM ... MSR_LBR_CORE_FROM + 7:
+ case MSR_LBR_CORE_TO ... MSR_LBR_CORE_TO + 7:
+ msr_info->data = 0;
+ break;
case MSR_TSC_AUX:
if (!msr_info->host_initiated &&
!guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
--
2.19.1



2019-02-19 06:41:00

by Like Xu

[permalink] [raw]
Subject: Re: [PATCH] KVM: Ignore LBR MSRs with no effect

On 2019/1/29 1:43, Anton Kuchin wrote:
> Win10 attempts to save these registers during KiSaveDebugRegisterState
> if LBR or BTF bits are set in MSR_IA32_DEBUGCTLMSR. It uses DR7 GE and LE
> flags for per-thread switching of these these features so zero value that
> is returned for MSR_IA32_DEBUGCTLMSR has no effect.
>
> These registers are used for debugging and shouldn't cause #GP and
> guest crash so just return zeroes just like we do for common x86 LBR
> MSRs (DEBUGCTLMSR, LAST[BRANCH|INT][TO|FROM]IP).
>
> Signed-off-by: Anton Kuchin <[email protected]>
> ---
> arch/x86/kvm/vmx/vmx.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index f6915f10e584..8bc56cf027ed 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -1769,6 +1769,13 @@ static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> else
> msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
> break;
> + case MSR_LBR_TOS:
> + case MSR_LBR_NHM_FROM ... MSR_LBR_NHM_FROM + 31:
> + case MSR_LBR_NHM_TO ... MSR_LBR_NHM_TO + 31:
> + case MSR_LBR_CORE_FROM ... MSR_LBR_CORE_FROM + 7:
> + case MSR_LBR_CORE_TO ... MSR_LBR_CORE_TO + 7:
> + msr_info->data = 0;
> + break;

It's better to move LBR stack MSRs set-up to
vmx/pmu_intel.c:intel_pmu_get_msr and using specified sizes (31 or 7) is
not a good practice.

vLBR is still in an unenabled state and
please check https://lkml.org/lkml/2018/12/26/82 for dependency check.

> case MSR_TSC_AUX:
> if (!msr_info->host_initiated &&
> !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
>