Kvm/ARM, linux: how can the guest code be interrupted?

The following code is an excerpt of function int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) (every function that I mention is linked to a hyperlink, but this is not visible from the mobile version!) and it shows what happens when the main vcpu is going to run guest code (context: kvm, arm):

    int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run) {
    //other code 

    preempt_disable();
    kvm_timer_flush_hwstate(vcpu);
    kvm_vgic_flush_hwstate(vcpu);
    local_irq_disable();

    // other code

    kvm_arm_setup_debug(vcpu);

    /**************************************************************
     * Enter the guest
     */
    trace_kvm_entry(*vcpu_pc(vcpu));
    __kvm_guest_enter();
    vcpu->mode = IN_GUEST_MODE;

    ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);

    vcpu->mode = OUTSIDE_GUEST_MODE;
    /*
     * Back from guest
     ************************************************************
    */

    //other code
    local_irq_enable();
    //rest of the function
}

I would like to let you focus on function local_irq_disable(): while we are in hypervisor mode (so executing the guest), we don't want to take HOST interrupts, so local_irq_disable() function lets all the HOST interrupts PENDING while we execute the guest.

What I need to figure out is that how the GuestOS can come back to HostOS when an interrupt happens. I did track that GuestOS comes back to HostOS upon a __kvm_vcpu_return. __kvm_vcpu_return is in turn called by el1_trap or by el1_irq. What I can't figure out is how such interrupts are taken into account if we called local_irq_disable() function before entering in the guest (namely, before invoking __kvm_guest_enter()). Theoretically, EL1 interrupts shall be disabled, not?