[libvirt] [RFC] Adding a "none" model for video cards

Hi, I'm trying to run vexpress-a9 virtual machines (armv7l architecture) using libvirt. The vexpress-a9 doesn't support vga cards, but libvirt adds a video "cirrus" video card when there is a "graphics" element in the domain xml. This prevents the domain from booting. At the moment, the only workaround I found is to add the following to my xml file : <qemu:commandline> <qemu:arg value='-vga'/> <qemu:arg value='none'/> </qemu:commandline> What do you think about adding a "none" model type for video cards ? I know it is a bit strange to declare a video card in order to have none... I don't know the QEMU driver well, but maybe this could be handled by the driver ? i.e. "no VGA card when the machine is vexpress-a9". Regards, Jean-Baptiste -- Jean-Baptiste ROUAULT R&D Engineer - diateam : Architectes de l'information Phone : +33 (0)2 98 050 050 Fax : +33 (0)2 98 050 051

On 07/30/2014 03:51 AM, Jean-Baptiste Rouault wrote:
Hi,
I'm trying to run vexpress-a9 virtual machines (armv7l architecture) using libvirt.
The vexpress-a9 doesn't support vga cards, but libvirt adds a video "cirrus" video card when there is a "graphics" element in the domain xml. This prevents the domain from booting.
At the moment, the only workaround I found is to add the following to my xml file : <qemu:commandline> <qemu:arg value='-vga'/> <qemu:arg value='none'/> </qemu:commandline>
What do you think about adding a "none" model type for video cards ? I know it is a bit strange to declare a video card in order to have none... I don't know the QEMU driver well, but maybe this could be handled by the driver ? i.e. "no VGA card when the machine is vexpress-a9".
Not the first time we've had to do that (for example, we support 'none' for the memballoon device, as well as for a <seclabel> tag when there is no labeling to be done), all to maintain back-compat where the defaults auto-populate devices. So your idea sounds reasonable on that front. However, I also wonder if this is a case where our code for post-parse touchups could be made smarter to not provide a cirrus default on architectures that don't support it, so that you would get the correct behavior without having to use an explicit none. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
participants (2)
-
Eric Blake
-
Jean-Baptiste Rouault