Re: [PATCH v11 13/21] KVM: selftests: Add helpers to init TDX memory and finalize VM

[Sean Christopherson <seanjc@google.com>]

On Fri, Oct 24, 2025, Sagi Shahar wrote:
> On Fri, Oct 24, 2025 at 11:02 AM Sean Christopherson <seanjc@google.com> wrote:
> >
> > On Thu, Oct 23, 2025, Sagi Shahar wrote:
> > > On Wed, Oct 15, 2025 at 11:25 AM Ira Weiny <ira.weiny@intel.com> wrote:
> > > >
> > > > Sagi Shahar wrote:
> > > > > From: Ackerley Tng <ackerleytng@google.com>
> > > > >
> > > >
> > > > [snip]
> > > >
> > > > > diff --git a/tools/testing/selftests/kvm/lib/x86/tdx/tdx_util.c b/tools/testing/selftests/kvm/lib/x86/tdx/tdx_util.c
> > > > > index 2551b3eac8f8..53cfadeff8de 100644
> > > > > --- a/tools/testing/selftests/kvm/lib/x86/tdx/tdx_util.c
> > > > > +++ b/tools/testing/selftests/kvm/lib/x86/tdx/tdx_util.c
> > > > > @@ -270,3 +270,61 @@ void vm_tdx_init_vm(struct kvm_vm *vm, uint64_t attributes)
> > > > >
> > > > >       free(init_vm);
> > > > >  }
> > > > > +
> > > >
> > > > [snip]
> > > >
> > > > > +
> > > > > +void vm_tdx_finalize(struct kvm_vm *vm)
> > > >
> > > > Why is this not a new kvm_arch_vm_finalize_vcpu() call?
> > >
> > > What do you mean?
> >
> > Ira is pointing out that upstream now has kvm_arch_vm_finalize_vcpus(), so you
> > can (and I agree, should) implement that for x86.c, and do vm_tdx_finalize() from
> > there (based on the VM shape) instead of requiring the caller to manually finalize
> > the TD.
> >
> > Unlike SEV, where userspace can manipulate guest state prior to LAUNCH, TDX guest
> > state is unreachable from time zero, i.e. there is unlikely to be many (any?) use
> > cases where a selftest wants to do something between creating vCPUs and finalizing
> > the TD.
> 
> There are actually a few use cases for calling vm_tdx_finalize
> seperately from the create phase. One such case is when a user wants
> to add additional memslots for testing shared memory conversions.

Why does creating memslots need to be done before finalizing the TD?  The extra
memslots shouldn't _need_ to be included in the initial image, and I don't recall
anything in the kernel that prevents adding a memslot after the TD is initialized.

> Another one is installing interrupt handlers. But these are probably
> rare enough that they can call __vm_create() and vm_vcpu_add()
> manually instead of using the wrapper

Hmm, for installing interrupt handlers, we could handle that by adding a guest
variant, e.g. so that the guest can install/remove exception handlers after
the initial image is finalized.  We probably want that anyways, otherwise tests
would be unable to manipulate handlers throughout a test.

In fact, it's tempting to kill off vm_install_exception_handler() in x86 entirely,
but that's probably not worth the churn (and some tests would be made more complex).