On Wed, Oct 18, 2023, Ashish Kalra wrote:
>
> On 10/18/2023 3:38 PM, Sean Christopherson wrote:
> > On Wed, Oct 18, 2023, Ashish Kalra wrote:
> > > > static int snp_handle_ext_guest_request(struct vcpu_svm *svm)
> > > > {
> > > > struct kvm_vcpu *vcpu = &svm->vcpu;
> > > > struct kvm *kvm = vcpu->kvm;
> > > > struct kvm_sev_info *sev;
> > > > unsigned long exitcode;
> > > > u64 data_gpa;
> > > >
> > > > if (!sev_snp_guest(vcpu->kvm)) {
> > > > ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, SEV_RET_INVALID_GUEST);
> > > > return 1;
> > > > }
> > > >
> > > > data_gpa = vcpu->arch.regs[VCPU_REGS_RAX];
> > > > if (!IS_ALIGNED(data_gpa, PAGE_SIZE)) {
> > > > ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, SEV_RET_INVALID_ADDRESS);
> > > > return 1;
> > > > }
> > > >
Doh, I forget to set
vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
> > > > vcpu->run->hypercall.nr = KVM_HC_SNP_GET_CERTS;
> > > > vcpu->run->hypercall.args[0] = data_gpa;
> > > > vcpu->run->hypercall.args[1] = vcpu->arch.regs[VCPU_REGS_RBX];
> > > > vcpu->run->hypercall.flags = KVM_EXIT_HYPERCALL_LONG_MODE;
> > > > vcpu->arch.complete_userspace_io = snp_complete_ext_guest_request;
> > > > return 0;
> > > > }
> > > >
> > >
> > > IIRC, the important consideration here is to ensure that getting the
> > > attestation report and retrieving the certificates appears atomic to the
> > > guest. When SNP live migration is supported we don't want a case where the
> > > guest could have migrated between the call to obtain the certificates and
> > > obtaining the attestation report, which can potentially cause failure of
> > > validation of the attestation report.
> >
> > Where does "obtaining the attestation report" happen? I see the guest request
> > and the certificate stuff, I don't see anything about attestation reports (though
> > I'm not looking very closely).
> >
>
> The guest requests that the firmware construct an attestation report via the
> SNP_GUEST_REQUEST command. The certificates are piggy-backed to the guest
> along with the attestation report (retrieved from the FW via the
> SNP_GUEST_REQUEST command) as part of the SNP Extended Guest Request NAE
> handling.
Ah, thanks!
In that case, my proposal should more or less Just Work™, we simply need to define
KVM's ABI to be that userspace is responsible for doing KVM_RUN with
vcpu->run->immediate_exit set before migrating if the previous exit was
KVM_EXIT_HYPERCALL with KVM_HC_SNP_GET_CERTS. This is standard operating procedure
for userspace exits where KVM needs to "complete" the VM-Exit, e.g. for MMIO, I/O,
etc. that are punted to userspace.
