On Wed, Oct 20, 2021, Tom Lendacky wrote: > On 10/20/21 12:40 PM, Sean Christopherson wrote: > > On Wed, Oct 20, 2021, Joerg Roedel wrote: > > This can race with the SIPI and effectively corrupt svm->vmcb->control.ghcb_gpa. > > > > vCPU0 vCPU1 > > #VMGEXIT(RESET_HOLD) > > __kvm_vcpu_halt() > > INIT > > SIPI > > sev_vcpu_deliver_sipi_vector() > > ghcb_msr_ap_rst_resp(1); > > This isn't possible. vCPU0 doesn't set vCPU1's GHCB value. vCPU1's GHCB > value is set when vCPU1 handles events in vcpu_enter_guest(). Argh, I was thinking of injecting regular IPIs across vCPUs. In hindsight it makes sense that INIT and SIPI are handled on the current vCPU, stuffing all that state from a different vCPU would be needlessly complex.
