Re: [Qemu-devel] Virtualbox svga card in KVM

[Stefan Hajnoczi]

On Tue, Apr 09, 2013 at 11:25:05AM -0700, Sriram Murthy wrote:
> 
> 
> Hi,
> 
> 
> >Personally, I think that the connection between VirtualBox and QEMU is
> >very unidirectional if there is any... So code-wise our focus should
> >rather be to avoid code copies/divergence within our tree and to share
> >code with existing in-tree devices, especially if you are not paid to
> >continuously take care of this device once accepted into QEMU - that's
> >how I interpret PMM's question below.
> 
> >There is nothing generally wrong with using KVM for guest driver
> >development or to make existing stripped-down guest images work at all
> >by adding such a special device.
> 
> >However, proposing to adopt a random vendor's paravirtual graphics card
> >just because it has a few more resolutions and drivers on a particular
> >platform does not strike me as a big advantage over SPICE, VMware VGA or
> >past virtio-vga/-fb standardization attempts.
> 
> Well, this idea came to me after I tried to use an OpenGL program that I 
> wrote in Win7 running on KVM with a local SDL display interface. 
> The experience was so bad, and to give it a shot, I ran it on VirtualBox, 
> with the same guest image (but with VirtualBox WDDM drivers) and there was a 
> night and day difference in experience.
> Also, frankly, IMHO, adding a new card to KVM (regardless of the features it 
> brings to the table) will only provide more choices for the end KVM user!.

It's even better to improve the existing stuff because adding new stuff
often creates partially overlapping but different feature sets.  Then
users need to learn about and decide which option is right for them.

Given your mention of OpenGL I guess the problem isn't just that a few
display modes are missing.

Please share what you find as you dig more into the vbox vga code so we
can learn what specific pieces are missing in QEMU.

Stefan