On Tue, Feb 02, 2021 at 08:54:30AM +0100, Paolo Bonzini wrote: > On 02/02/21 01:18, Eduardo Habkost wrote: > > On Tue, Feb 02, 2021 at 12:28:38AM +0100, Paolo Bonzini wrote: > > > Il mar 2 feb 2021, 00:05 Eduardo Habkost <[email protected]> ha scritto: > > > > > > > On Mon, Feb 01, 2021 at 11:59:48PM +0100, Paolo Bonzini wrote: > > > > > Il lun 1 feb 2021, 23:54 Eduardo Habkost <[email protected]> ha > > > > scritto: > > > > > > > > > > > Not having a feature name in feature_word_info breaks error > > > > > > reporting and query-cpu-model-expansion. Add the missing feature > > > > > > name to feature_word_info[FEAT_VMX_EPT_VPID_CAPS].feat_names[14]. > > > > > > > > > > > This is intentional, because there's no way that any hypervisor can > > > > > run > > > > if > > > > > this feature is disabled. > > > > > > > > If leaving the feature without name enables some desirable > > > > behavior, that's by accident and not by design. Which part of > > > > the existing behavior is intentional? > > > > > > > > > > Not being able to disable it. > > > > We can make it a hard dependency of vmx, then. We shouldn't > > leave it without a name, though. > > The feature is already added to the MSRs unconditionally in > kvm_msr_entry_add_vmx. I think we can just remove it from the models > instead.
Sounds even simpler, and better. I'll do it in v2. -- Eduardo
