Re: [PATCH v2 00/16] virtio-gpu: split into two devices.

[Gerd Hoffmann]

On Fri, Apr 30, 2021 at 07:32:58PM +0400, Marc-André Lureau wrote:
> Hi
> 
> On Fri, Apr 30, 2021 at 4:23 PM Gerd Hoffmann <kraxel@redhat.com> wrote:
> 
> > Currently we have one virtio-gpu device.  Problem with this approach is
> > that if you compile a full-featured qemu you'll get a virtio-gpu device
> > which depends on opengl and virgl, so these dependencies must be
> > installed and the libraries will be loaded into memory even if you don't
> > use virgl.  Also the code is cluttered with #ifdefs and a bit messy.
> >
> > This patch series splits the virtio-gpu device into two:
> >
> >  (1) virtio-gpu-device becomes the non-virgl device, same as
> >      virtio-gpu-device,virgl=off today.
> >  (2) virtio-gpu-gl-device is the new virgl device, same as
> >      virtio-gpu-device,virgl=on today.
> >
> > When compiling qemu without virglrenderer support virtio-gpu-device
> > behavior doesn't change.
> >
> > v2:
> >  - rebase to latest master.
> >  - move pci and vga wrappers to separate modules.
> >  - fix ci failures.
> >
> 
> The patch series looks good.
> 
> But isn't that a breaking change? Any existing user of
> virtio-gpu/vga,virgl=on will no longer get a working setup. Right?

Correct.

> Imho, in
> this case (virgl VM being not very common) the benefit is worth it.
> However, I suggest to keep the 'virgl=' property, and print a deprecation /
> replaced-by warning, falling back to non-virgl/2d mode.

Problem with that is that libvirt uses the virgl property to figure
whenever virtio-gpu-pci / virtio-vga supports virgl or not.  So a dummy
virgl property just for printing a warning message doesn't look like a
good idea to me.

> Or perhaps we could
> have more cleverness to have virgl=on aliasing to the new devices.

If that is doable without ugly hacks I'm open to it.
Suggestions how to do that anyone?

take care,
  Gerd