Hello! > I liked the idea of adding all this > if user supplies e.g. <controller type='pci' model='pci-root'/> IMHO there would be a usability problem with this: a) Too much excessive typing. b) It is not intuitively understood that you have to define the controller explicitly. Especially because if you run qemu by hands, this is not true. For example, when i first started to cope with PCI devices, i quickly found out about <address type='pci'> in google. But having to add something else is not that obvious and easy to discover. c) AFAIK virt-manager even cannot add PCI controller explicitly, it relies on libvirt's automagic in question. I think it would be a perfect consensus if we add "PCI devices are actually present" to the condition. In this case, if there is at least one PCI device in the input XML, controller description is automatically added. If there is none, it's OK to remove it. IMHO it's not difficult to implement because it's post-parse stage, and everything that the user entered has already been parsed. But we still want the capability check because old qemu versions simply cannot accommodate this. Therefore it doesn't remove the need to fix up tests. Kind regards, Pavel Fedin Expert Engineer Samsung Electronics Research center Russia -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list