Re: [PATCH] spice: delay starting until display are initialized

[Marc-André Lureau]

On Thu, Jan 28, 2021 at 8:34 PM Gerd Hoffmann <kraxel@redhat.com> wrote:
>
>   Hi,
>
> > > So move the vmstate handler registration call too?
> > > I'd prefer to not add more state variables if we can avoid it ...
> >
> > Does that seem right to you?
>
> > @@ -626,7 +625,7 @@ static int add_channel(void *opaque, const char
> > *name, const char *value,
> >  static void vm_change_state_handler(void *opaque, int running,
> >                                      RunState state)
> >  {
> > -    if (running && spice_display_init_done) {
> > +    if (running) {
> >          qemu_spice_display_start();
> >      } else if (state != RUN_STATE_PAUSED) {
> >          qemu_spice_display_stop();
> > @@ -635,7 +634,7 @@ static void vm_change_state_handler(void *opaque,
> > int running,
> >
> >  void qemu_spice_display_init_done(void)
> >  {
> > -    spice_display_init_done = true;
> > +    qemu_add_vm_change_state_handler(vm_change_state_handler, NULL);
> >      vm_change_state_handler(NULL, runstate_is_running(), runstate_get());
>
> I'd just call qemu_spice_display_start() directly here, the need for
> runstate_get() goes away then.  Otherwise looks good to me.

Hmm, that could work, but it will behave differently as we will start
spice even if the VM is not running then.


-- 
Marc-André Lureau