> On 25 Nov 2018, at 19:59, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > On 13/11/18 17:37, Vitaly Kuznetsov wrote: >> I was wondering how evmcs selftest was passing before this but then I >> remembered the Hyper-V bug (which puts VMCS revision from >> IA32_VMX_BASIC_MSR - where we have VMCS12_REVISION and not >> KVM_EVMCS_VERSION) so 'evmcs->revision_id' in this context is actually >> VMCS12_REVISION and there's no change post-patch. >> >> We, however, will need to fix the selftest after we start allowing >> KVM_EVMCS_VERSION as it currently mocks erroneous Hyper-V behavior :-) > > Is this a Hyper-V bug, or a TLFS mistake? > > Queuing the patch, anyway. Thanks! > > Paolo We have later discussed this with Microsoft folks. They have confirmed this is a Hyper-V bug. They plan to fix this and told us that we can detect if guest is buggy by examining HV_X64_MSR_GUEST_OS_ID and check if guest is Windows NT (VendorId=1 and OsId=4) and BuildNumber<18000. -Liran
