On Wed, 2014-01-08 at 12:20 +0100, Stefan Bader wrote:
On 08.01.2014 11:42, Ian Campbell wrote:
> On Wed, 2014-01-08 at 11:39 +0100, Stefan Bader wrote:
>> This basically reverts commit ba64b97134a6129a48684f22f31be92c3b6eef96
>> "libxl: Allow libxl to set NIC devid". However assigning devid's
>> before calling libxlMakeNic does not work as that is calling
>> libxl_device_nic_init which sets it back to -1.
>> Right now auto-assignment only works in the hotplug case. But even if
>> that would be fixed at some point (if that is possible at all),
>
> I think it should be -- any chance you could prepare a patch to libxl as
> well?
Not sure that is that simple. At least not in libxl_device_nic_init. The way I
understand this is done does not seem to allow a good solution. Creation looks
to allow initializing NIC definitions in some array that is not connected to a
specific domain, yet.
Right, I think I previously concluded that for the domain create case
this could be done in initiate_domain_create, with the hotplug case
handled via the existing code in libxl_device_nic_add.
I suppose what I could come up with is the kind of modification I was
proposing
in one of those emails last year (sounds like an awful long time :)). So at the
point where the device-model is about to get started and the code already loops
over the NIC definitions. That won't give the libvirt path default devid's but
any other user that has not done yet.
-Stefan