Re: [PATCH v2 2/3] KVM: x86/mmu: Fully re-evaluate MMIO caching when SPTE masks change

From: Sean Christopherson
Date: Wed Aug 24 2022 - 17:08:57 EST


On Fri, Aug 19, 2022, Paolo Bonzini wrote:
> On 8/19/22 18:21, David Matlack wrote:
> > On Wed, Aug 3, 2022 at 3:50 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
> > >
> > > Fully re-evaluate whether or not MMIO caching can be enabled when SPTE
> > > masks change; simply clearing enable_mmio_caching when a configuration
> > > isn't compatible with caching fails to handle the scenario where the
> > > masks are updated, e.g. by VMX for EPT or by SVM to account for the C-bit
> > > location, and toggle compatibility from false=>true.
> > >
> > > Snapshot the original module param so that re-evaluating MMIO caching
> > > preserves userspace's desire to allow caching. Use a snapshot approach
> > > so that enable_mmio_caching still reflects KVM's actual behavior.
> >
> > Is updating module parameters to reflect the actual behavior (vs.
> > userspace desire) something we should do for all module parameters?
> >
> > I am doing an unrelated refactor to the tdp_mmu module parameter and
> > noticed it is not updated e.g. if userspace loads kvm_intel with
> > ept=N.
>
> If it is cheap/easy then yeah, updating the parameters is the right thing to
> do. Generally, however, this is only done for kvm_intel/kvm_amd modules
> that depend on hardware features, because they are more important for
> debugging user issues. (Or at least they were until vmx features were added
> to /proc/cpuinfo).

IMO, unless it's _really_ hard, KVM should keep the parameters up-to-date with
reality, e.g. so that userspace can assert that a feature is fully enabled, either
for testing purposes (unrestricted guest?) or to prevent running with a "bad" config.

We've had at least one internal OMG-level bug where KVM effectively ran with the
wrong MMU configuration, and IIRC one of the actions taken in response to that was
to assert on KVM's parameters post-load.