On Fri, 24 Nov 2017 16:30:24 +0100 Halil Pasic <pa...@linux.vnet.ibm.com> wrote:
> On 11/24/2017 03:58 PM, Christian Borntraeger wrote: > > > > > > On 11/24/2017 02:27 PM, Cornelia Huck wrote: > >> On Fri, 24 Nov 2017 14:01:20 +0100 > >> Christian Borntraeger <borntrae...@de.ibm.com> wrote: > >> > >>> I first liked the idea to have it as a property of the css, but > >>> this is all pretty unclear how to do right. I start to think that going > >>> with > >>> Halils first patch (a property per virtio device) is going to be the most > >>> simple solution without causing any harm. After all as of today we only > >>> want > >>> to have a way to tell libvirt that devices can be everywhere. Specifying > >>> the > >>> default css might be something that we want to have in the future, but > >>> here > >>> future might even mean never. > >> > >> I still don't like the idea of a per-device property, but I agree that > >> adding a css property would need too much discussion to get to a > >> solution in the near future. > >> > >> Is there anything that speaks against a machine property, though? While > >> not ideal, I like it better than the per-device one. > > > > In theory this should work. > > > > In reality it seems more complicated. A per-device property is easy and can > > be > > inspected on the command line (e.g. -device virtio-blk-ccw,help), while a > > new > > machine property would require to change the qemu help output and > > qemu-options > > file (which makes it visible for all architectures). > > And then we have the fun of describing, that this property is weird, and can > not be set, and it's value does not matter. Well, that's the case for both, no? (Unless we simply make this a "default cssid" prop after all - then it would be more than just a simple indication for libvirt...) > > BTW one can do -machine s390-ccw-virtio-2.11,help and get a list of > properties, > so the command line introspection would work similar. > > I've already brought some other arguments against the machine property. > Won't repeat them here. I have not read them yet. > > > Not sure if there are > > easier ways to do it. (e.g. QOM-only things, but then we have no command > > line > > way of doing it) > > > > > > If we don't care about if it can be introspected on the command line my > trait type approach is pretty minimal. But the least surprise is probably > still the device property (IMHO). I think it's the other way around.