[Pkg-libvirt-maintainers] Bug#583111: Bug#583111: device

[Guido Günther]

Hi Ian,
On Fri, Jul 30, 2010 at 03:04:06AM +0800, Windows Live wrote:
> 
> Guido, 
>  
> I could have elaborated a bit.  libvirt has a gap re video devices.  What I did was manually added a video device in the xml file.  I then defined the vm with that xml file.  Having defined and started the vm, I do an xml dump of the machine, to find that the video device I have added has been discarded by libvirt.
>  
>  
> from my description of this to the virt-manager developer of redhat;
>  
> > The other is the making of video devices. Devices like serial screen or watchdog, virt-manager is quick to indicate if they are not supported for the hypervisor environment you are in. Video devices seem to have no validation checking. It seems that if a cirrus video emulation is selected in a full virt install, it isn't included in the xml config file. I tried manually including one, defined the vm with a cirrus video device, then once defined , it was seemingly discarded by libvirt, but without any feedback of error state. i.e., I defined the vm, then straight away did dumpxml of the vm, and no video device. Leaves me wondering.
> > 
> > If you have an established virt-manager vnc console, manually add a video device, it doesn't get added to the hardware. Select something else, and it draws error if it can't be added, or does appear if it is added.
> > 
> > For a para-virt, it makes sense if one isn't stipulated or used, but with full virt, it's a required element, is it not?
> > Does this make sense? I might not have made a clear description.
> >
>  
> Looks like libvirt doesn't support specifying video devices for xen.
> This is a result of the funkiness of how libvirt handles XML it doesn't
> support, it just throws it away.
>  
> We could prevent users from attaching a video device in the 'Add
> Hardware' wizard, but if libvirt ever catches up and allows specifying
> video devices, we will have to update virt-manager to handle this.
Yes, it'd be great if libvirt could communicate the capabilities of the
underlying hypervisor upfront, somewhat like "virsh capabilities" with
more details.

> Libvirt really needs some progammatic way to tell us about what it
> supports for each driver so virt-manager doesn't need to play these
> guessing games. There isn't really anything I think we should change at
> the virt-manager level for now though.
I see. Thanks for your feedback.
 -- Guido