[libvirt] [PATCH RFC 0/8] qemu: allow disabling certain virtio revisions

Laine Stump laine at laine.org
Thu Aug 4 15:24:43 UTC 2016

On 08/04/2016 11:11 AM, Andrea Bolognani wrote:
> On Mon, 2016-08-01 at 10:02 -0400, Laine Stump wrote:
>> 2) To assure that the actual device presented to the guest doesn't
>> change in the future when defaults are changed, we may want to autofill
>> "version" even when none is specified. (We would of course be stuck with
>> the unfortunate problem of what to do with existing configs)
> I don't think we should worry too much about this: since we
> lock on a versioned machine type on guest creation, and we can
> assume the default virtio version will only change along with
> a bump in machine type version[1], we'll always present the
> same hardware to the guest.

Hmm, and anyway, lack of  a version means "be compatible with *all* the 
versions", so there isn't any single value of version that could be 
filled in.


It does somehow bother me that a single attribute is being used to 
control two separate options in qemu. I suppose, though, that this is 
okay, because those two options have a possibility of 4 different 
combinations, with one of those combinations being nonsensical 
(disable_modern and disable_legacy both set to yes), so the remaining 
three combinations map to version = (nothing), "0.9", and "1.0".

> [1] What would be the point in versioned machine types
>      otherwise? :)

Oh, there are plenty of other things aside from device default option 
settings that are preserved by versioned machinetypes.

More information about the libvir-list mailing list