Re: [PATCH] libxl: use qemu-xen (upstream QEMU) as device model by default

["Roger Pau Monné" <roger.pau@citrix.com>]

On 05/12/12 16:58, Ian Jackson wrote:
> Roger Pau Monne writes ("Re: [Xen-devel] [PATCH] libxl: use qemu-xen (upstream QEMU) as device model by default"):
>> On 27/11/12 16:17, Stefano Stabellini wrote:
>>> -                LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN_TRADITIONAL;
>>> +                LIBXL_DEVICE_MODEL_VERSION_QEMU_XEN;
>>
>> Is there anyway we may keep qemu-traditional as default for NetBSD?
>> Upstream Qemu is not working on NetBSD, and I'm afraid it needs some
>> heavy patching.
> 
> Right.  OK, that's something we need to take care of then.

I have a pending patch for NetBSD that fixes a problem with the privcmd
device, and the way NetBSD handles IOCTL_PRIVCMD_MMAPBATCH. It is here:
http://mail-index.netbsd.org/port-xen/2012/06/27/msg007464.html

With this patch at least we are able to launch Qemu-upstream without
crashing, but the next problem is with network interfaces. There's no
way in NetBSD to change the name of a cloned tap interface, and in libxl
we pass the desired name of the tap interface to be created to Qemu, and
then we launch hotplug scripts according to that name. This doesn't work
in NetBSD, but I see several possible solutions:

1. Implement interface renaming in NetBSD

2. Implement a QMP interface in Qemu to query information about network
devices, so we can get the actual name of the interface that Qemu has
created. There was a partial implementation of this as part of a Qemu
GSoC, but it never got commited, see
http://wiki.qemu.org/Google_Summer_of_Code_2010/QMP#query-netdev

3. Create the tap interface in libxl and use pass fd to pass that file
descriptor to Qemu "-net tap,fd=XXX[,...]"

More suggestions?

>> Could a helper function be added to libxl_{netbsd/linux}.c to decide
>> which device model to use?
>
> Yes, I think that would be fine.
> 
> Ian.
>