[Spice-devel] [PATCH spice] gl: fix client mouse mode

[Marc-André Lureau]

----- Original Message -----
> > 
> > From: Marc-André Lureau <marcandre.lureau at redhat.com>
> > 
> > Since 2.8, QEMU now longer creates QXL primary surfaces when using
> > GL. This change broke client-side mouse mode, because Spice server
> > relies on primary surface conditions.
> > 
> > When GL is enabled, use GL scanout informations, similar to what
> > red_qxl_get_allow_client_mouse() is doing. The main difference is that
> > the GL surface doesn't have mouse_mode attached (what for?).
> > 
> > NB: this code looks like it handles only first QXL instance only, and
> > uses the whole scanout/surface size for the input table size, perhaps
> > monitor-configs would be a better choice.
> > 
> > Signed-off-by: Marc-André Lureau <marcandre.lureau at redhat.com>
> > ---
> >  server/red-qxl.c | 2 ++
> >  server/reds.c    | 8 +++++++-
> >  2 files changed, 9 insertions(+), 1 deletion(-)
> > 
> > diff --git a/server/red-qxl.c b/server/red-qxl.c
> > index b6b3770b..ba7b641a 100644
> > --- a/server/red-qxl.c
> > +++ b/server/red-qxl.c
> > @@ -897,6 +897,8 @@ void spice_qxl_gl_scanout(QXLInstance *qxl,
> >      /* FIXME: find a way to coallesce all pending SCANOUTs */
> >      dispatcher_send_message(qxl_state->dispatcher,
> >                              RED_WORKER_MESSAGE_GL_SCANOUT, &payload);
> > +
> > +    reds_update_client_mouse_allowed(qxl_state->reds);
> >  }
> >  
> >  SPICE_GNUC_VISIBLE
> > diff --git a/server/reds.c b/server/reds.c
> > index 40c94851..a89c5e90 100644
> > --- a/server/reds.c
> > +++ b/server/reds.c
> > @@ -4256,10 +4256,16 @@ void reds_update_client_mouse_allowed(RedsState
> > *reds)
> >      if (num_active_workers > 0) {
> >          GListIter it;
> >          QXLInstance *qxl;
> > +        SpiceMsgDisplayGlScanoutUnix *gl;
> >  
> >          allow_now = TRUE;
> >          FOREACH_QXL_INSTANCE(reds, it, qxl) {
> > -            if (red_qxl_get_primary_active(qxl)) {
> > +            if ((gl = red_qxl_get_gl_scanout(qxl))) {
> > +                x_res = gl->width;
> > +                y_res = gl->height;
> > +                red_qxl_put_gl_scanout(qxl, gl);
> > +                break;
> > +            } else if (red_qxl_get_primary_active(qxl)) {
> >                  allow_now = red_qxl_get_allow_client_mouse(qxl, &x_res,
> >                  &y_res);
> >                  break;
> >              }
> 
> First though when I first read this patch was:
> 
> A: Is your cat red?
> R: Yes, it's red. I don't have a cat.
> 
> Why Qemu don't allocate the primary surface in the first place?

Because it no longer uses QXL 2d drawing commands.

> The issue is that multiple code (for instance monitor configs)
> rely on having a primary surface.

We would need to fix that then, I haven't checked multiple monitors with 2.8, I know it used to work not long ago.

> Maybe a lazy allocation of the canvas could help?

What for?

> This patch looks like a workaround for a bug introduced by Qemu.

It depends on spice-server behaviour, which is not enforced at this point. Imho, there is no need to create a 2d primary surface if all you do is GL scanouts.

> I'm sure it solves the client mouse but I'd like to know what
> other people knowing better client and multi monitor code/protocol
> think about not having the primary surface.
> 
> Reviewed-by: Frediano Ziglio <fziglio at redhat.com>
> 
> Frediano
>