Re: [PATCH] KVM: svm: virtualize MSR reads from MC4_MISC1

[Paolo Bonzini <pbonzini@redhat.com>]

Il 26/06/2014 14:22, Matthias Lange ha scritto:
> Linux' AMD MCE code tries to read from the MC4_MISC1 (0xc0000408) MSR. Because
> this read is not virtualized within KVM, a GPE is injected into the guest.
> This patch handles guest reads from MC4_MISC and returns 0 to the guest.
>
> Signed-off-by: Matthias Lange <matthias.lange@kernkonzept.com>
> ---
>  arch/x86/kvm/x86.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 24d70d4..d5a305b 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2530,6 +2530,9 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata)
>  			return 1;
>  		data = vcpu->arch.osvw.status;
>  		break;
> +	case 0xc0000408: /* MC4_MISC1 */
> +		data = 0;
> +		break;
>  	default:
>  		if (kvm_pmu_msr(vcpu, msr))
>  			return kvm_pmu_get_msr(vcpu, msr, pdata);
>

Why don't you have to do the same with MC4_MISC0?

Is your kernel compiled with or without CONFIG_PARAVIRT?  Paravirt does

         .read_msr = native_read_msr_safe,
         .write_msr = native_write_msr_safe,

so you shouldn't get these #GP exceptions.

Paolo