Re: [Qemu-devel] [PATCH 2/9] virtio-gpu/2d: add virtio gpu core code

[Gerd Hoffmann]

  Hi,

> > +    for (i = 0; i < g->conf.max_outputs; i++) {
> > +        g->scanout[i].con =
> > +            graphic_console_init(DEVICE(g), i, &virtio_gpu_ops, g);
> > +        if (i > 0) {
> > +            dpy_gfx_replace_surface(g->scanout[i].con, NULL);
> 
> GTK can't cope very well with NULL surfaces (read: after patch 7 (when 
> using -vga virtio finally works) "x86_64-softmmu/qemu-system-x86_64 -vga 
> virtio -display gtk" segfaults). With SDL (SDL2), it works fine.

Yes, I know.  I have a gtk patch for that in the pipeline.

But that alone doesn't get multihead work sanely as there is at least
one more problem to tackle:  Mouse coordinates are not correct.  Our
options are:

 (a) Have guest notify host about screen arrangement, then transform
     coordinates in qemu.  Dave did that initially.
 (b) Have a guest agent (spice style).
 (c) Have one tablet per head and link them.  That is very simliar to
     physical hardware (think touchscreen).

I think (c) will work best long-term, but it is also the most difficult
one as various projects are involved here to get the guest-side
configuration done.

I've changed the number of heads to be 1 by default, we better sort all
those issues before we flip the multihead default switch.

cheers,
  Gerd