On Fri, Jan 14, 2022 at 10:17:34AM +0200, Maxim Levitsky wrote: >On Fri, 2022-01-14 at 10:58 +0800, Chao Gao wrote: >> On Thu, Jan 13, 2022 at 10:19:21PM +0000, Sean Christopherson wrote: >> > On Tue, Jan 11, 2022, Maxim Levitsky wrote: >> > > Both Intel and AMD's PRM also state that changing APIC ID is implementation >> > > dependent. >> > > >> > > I vote to forbid changing apic id, at least in the case any APIC acceleration >> > > is used, be that APICv or AVIC. >> > >> > That has my vote as well. For IPIv in particular there's not much concern with >> > backwards compability, i.e. we can tie the behavior to enable_ipiv. >Great! >> >> Hi Sean and Levitsky, >> >> Let's align on the implementation. >> >> To disable changes for xAPIC ID when IPIv/AVIC is enabled: >> >> 1. introduce a variable (forbid_apicid_change) for this behavior in kvm.ko >> and export it so that kvm-intel, kvm-amd can set it when IPIv/AVIC is >> enabled. To reduce complexity, this variable is a module level setting. >> >> 2. when guest attempts to change xAPIC ID but it is forbidden, KVM prints >> a warning on host and injects a #GP to guest. >> >> 3. remove AVIC code that deals with changes to xAPIC ID. >> > >I have a patch for both, I attached them. Looks good to me. We will drop this patch and rely on the first attached patch to forbid guest from changing xAPIC ID. >I haven't tested either of these patches that much other than a smoke test, >but I did test all of the guests I have and none broke in regard to boot. > >I will send those patches as part of larger patch series that implements >nesting for AVIC. I hope to do this next week. Thanks.
