[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZrY5h746smS4j5ak@google.com>
Date: Fri, 9 Aug 2024 08:45:11 -0700
From: Sean Christopherson <seanjc@...gle.com>
To: Pratik Rajesh Sampat <pratikrajesh.sampat@....com>
Cc: Tom Lendacky <thomas.lendacky@....com>, kvm@...r.kernel.org, shuah@...nel.org,
michael.roth@....com, pbonzini@...hat.com, pgonda@...gle.com,
linux-kselftest@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [RFC 3/5] selftests: KVM: SEV IOCTL test
On Thu, Jul 11, 2024, Pratik Rajesh Sampat wrote:
> >> +static void sev_guest_status_assert(struct kvm_vm *vm, uint32_t type)
> >> +{
> >> + struct kvm_sev_guest_status status;
> >> + bool cond;
> >> + int ret;
> >> +
> >> + ret = __vm_sev_ioctl(vm, KVM_SEV_GUEST_STATUS, &status);
> >> + cond = type == KVM_X86_SEV_VM ? !ret : ret;
> >> + TEST_ASSERT(cond,
> >> + "KVM_SEV_GUEST_STATUS should fail, invalid VM Type.");
> >> +}
> >> +
> >> +static void test_sev_launch(void *guest_code, uint32_t type, uint64_t policy)
> >> +{
> >> + struct kvm_vcpu *vcpu;
> >> + struct kvm_vm *vm;
> >> + struct ucall uc;
> >> + bool cond;
> >> + int ret;
> >> +
> >
> > Maybe a block comment here indicating what you're actually doing would
> > be good, because I'm a bit confused.
> >
> > A policy value of 0 is valid for SEV, so you expect each call to
> > succeed, right? And, actually, for SEV-ES the launch start will succeed,
> > too, but the launch update will fail because LAUNCH_UPDATE_VMSA is not
> > valid for SEV, but then the launch measure should succeed. Is that
> > right? What about the other calls?
> >
>
> Sure, I can do that.
> Yes for SEV, the policy value of 0 succeeds for everything except when
> we try to run and we see a KVM_EXIT_IO.
>
> For SEV-ES, with the policy value of 0 - we don't see launch_start
> succeed. It fails with EIO in this case. Post that all the calls for
> SEV-ES also fail subsequent to that. I guess the core idea behind this
> test is to ensure that once the first bad case of launch_start fails, we
> should see a cascading list of failures.
>
> >> + vm = vm_sev_create_with_one_vcpu(type, guest_code, &vcpu);
> >> + ret = sev_vm_launch_start(vm, 0);
> >> + cond = type == KVM_X86_SEV_VM ? !ret : ret;
> >> + TEST_ASSERT(cond,
Don't bury the result in a local boolean. It's confusing, and _worse_ for debug
as it makes it impossible to see what actually failed (the assert message will
simply print "cond", which is useless).
> >> + "KVM_SEV_LAUNCH_START should fail, invalid policy.");
This is a blatant lie, because the KVM_X86_SEV_VM case apparently expects success.
Similar to Tom's comments about explaing what this code is doing, these assert
messages need to explain what the actually expected result it, provide a hint as
to _why_ that result is expected, and print the result. As is, this will be
unnecessarily difficult to debug if/when it fails.
Powered by blists - more mailing lists