[libvirt] [PATCH v1 00/11] Enable vfio-pci 'property' for mediated device

[Erik Skultety]

On Wed, Jun 27, 2018 at 02:53:28PM +0100, Daniel P. Berrangé wrote:
> On Wed, Jun 27, 2018 at 03:34:37PM +0200, Erik Skultety wrote:
> > See the RFC here:
> > https://www.redhat.com/archives/libvir-list/2018-May/msg02218.html
> >
> > Since RFC:
> > - split graphics 'gl' to a standalone structure since it's now SPICE, SDL,
> >   *and* VNC that will support it
> >
> > - egl-headless support for VNC, since VNC doesn't support OpenGL natively like
> > SPICE does
> >
> > - added a new attribute 'native' for spice which will instruct libvirt to use
> > '-display egl-headless' instead of libvirt trying to figure this on out by
> > itself, since egl-headless might have other uses besides mdev with VNC
>
> Can you elaborate on this part, as it feels a bit odd to me.
>
> Currently -display accepts vnc or spice or sdl or gtk or none
>
> Each of these -display options corresponds to a different <graphics type=XXX/>
> element in libvirt XML, except 'none' which corresponds to no <graphics> at
> all.
>
> We already allow multiple <graphics> elements to turn on say both SPICE and
> VNC at once.
>
> So why would we want to treat egl-headless differently as an attribute under
> the SPICE graphics, as opposed to just having  <graphics type=egl-headless>
> as a thing in its own right.

Egl-headless doesn't make sense on its own, neither does it make sense with
anything else but SPICE and VNC. And since we're talking about dri render
nodes, I figured it would be better to design this as an attribute to the gl
element rather than a standalone element.

Erik