[Engine-devel] http://www.ovirt.org/wiki/Features/Design/StablePCIAddresses

[Daniel P. Berrange]

On Tue, Dec 13, 2011 at 10:02:24AM +0200, Dor Laor wrote:
> Hi,
> 
> I read the design here and I like to make sure that the future road map 
> will expand beyond the current scope.
> 
> The current design totally rely on libvirt and does not parse the 
> content of the PCI addressing. That's really really basic. The user 
> should be able to specify pci slot allocation of his devices through the 
> gui. I guess you won't be able to do that w/ the current scheme.
> 
> Also, what about devices that can't be hot plug (like qxl)? You need to 
> reveal this info to the user. Currently we have ability in the kvm bios 
> (seabios) to automatically disable the host plug of some critical 
> devices like the vga driver (qxl) and others. The user should be allowed 
> to hot plug/unplug only allowed devices.
> 
> You have to make your design work w/ pci bridges since we'll add it to 
> qemu and once there is such VM (management should enable the bridge) 
> there will be more pci devices available to it.

The general principle from libvirt is that there are two elements to
be tracked, the actual per-device <address> element, and at the VM
level one or more <controller> device elements. Currently we don't
expose any <controller> devices for PCI, since we only have a single
PCI root complex and single PCI bus. We do, however, use <controller>
for SCSI, CCID, etc, etc.  In the future when we add support for
PCI bridges, or PCI root complexes, then we'll add new <controller>
devices to represent them.

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|