On Wed, 3 Feb 2021 at 15:11, Dr. David Alan Gilbert <dgilb...@redhat.com> wrote: > It's interesting that on x86 we've got a longterm request to *add* cpu > features to the stream to detect screwups caused by using mismatched > CPUs; so it's not necessarily a bad idea to include it once you realise > it's there.
I think we would want to do that by checking the ID registers, not the legacy ad-hoc feature-flags word. In fact I think for KVM at least we already perform the check on the ID registers, in that the kernel will return an error if we pass it a value for an ID register and it's not the value it should be. I forget whether we try to check this for TCG (the mechanism for handling sysreg migration there is different). As a side note I suspect that the first two issues Aaron ran into are TCG-only and don't affect KVM. thanks -- PMM