On 05/11/2017 15:07, Liran Alon wrote: > + /* > + * If we reinjected a previous event, > + * don't consider a new pending event > + */ > + if (kvm_event_needs_reinjection(vcpu)) > + return 0; > + Could you end up with WARN_ON(vcpu->arch.exception.pending); in vcpu_enter_guest after returning 0 here? Maybe it would be safer to return a non-zero value so that the caller sets req_immediate_exit = true. But I haven't really thought through the consequences. Thanks, Paolo