Re: [PATCH v2 14/27] KVM: x86: Reject memslot MOVE operations if KVMGT is attached

[Sean Christopherson <seanjc@google.com>]

On Wed, Mar 15, 2023, Yan Zhao wrote:
> On Fri, Mar 10, 2023 at 04:22:45PM -0800, Sean Christopherson wrote:
> > Disallow moving memslots if the VM has external page-track users, i.e. if
> > KVMGT is being used to expose a virtual GPU to the guest, as KVM doesn't
> > correctly handle moving memory regions.
> > 
> > Note, this is potential ABI breakage!  E.g. userspace could move regions
> > that aren't shadowed by KVMGT without harming the guest.  However, the
> > only known user of KVMGT is QEMU, and QEMU doesn't move generic memory
> > regions.  KVM's own support for moving memory regions was also broken for
> > multiple years (albeit for an edge case, but arguably moving RAM is
> > itself an edge case), e.g. see commit edd4fa37baa6 ("KVM: x86: Allocate
> > new rmap and large page tracking when moving memslot").
> > 
> > Signed-off-by: Sean Christopherson <seanjc@google.com>
> ...
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index 29dd6c97d145..47ac9291cd43 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -12484,6 +12484,13 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
> >  				   struct kvm_memory_slot *new,
> >  				   enum kvm_mr_change change)
> >  {
> > +	/*
> > +	 * KVM doesn't support moving memslots when there are external page
> > +	 * trackers attached to the VM, i.e. if KVMGT is in use.
> > +	 */
> > +	if (change == KVM_MR_MOVE && kvm_page_track_has_external_user(kvm))
> > +		return -EINVAL;
> Hmm, will page track work correctly on moving memslots when there's no
> external users?
> 
> in case of KVM_MR_MOVE,
> kvm_prepare_memory_region(kvm, old, new, change)
>   |->kvm_arch_prepare_memory_region(kvm, old, new, change)
>        |->kvm_alloc_memslot_metadata(kvm, new)
>             |->memset(&slot->arch, 0, sizeof(slot->arch));
>             |->kvm_page_track_create_memslot(kvm, slot, npages)
> The new->arch.arch.gfn_write_track will be fresh empty.
> 
> 
> kvm_arch_commit_memory_region(kvm, old, new, change);
>   |->kvm_arch_free_memslot(kvm, old);
>        |->kvm_page_track_free_memslot(slot);
> The old->arch.gfn_write_track is freed afterwards.
> 
> So, in theory, the new GFNs are not write tracked though the old ones are.
> 
> Is that acceptable for the internal page-track user?

It works because KVM zaps all SPTEs when a memslot is moved, i.e. the fact that
KVM loses the write-tracking counts is benign.  I suspect no VMM actually does
does KVM_MR_MOVE in conjunction with shadow paging, but the ongoing maintenance
cost of supporting KVM_MR_MOVE is quite low at this point, so trying to rip it
out isn't worth the pain of having to deal with potential ABI breakage.

Though in hindsight I wish I had tried disallowed moving memslots instead of
fixing the various bugs a few years back. :-(