Hi Stefan, On Thu, Mar 15, 2012 at 07:08:47PM +0100, Stefan Bader wrote: > Package: libvirt > Version: 0.9.8-2 > Severity: normal > Tags: patch > User: [email protected] > Usertags: origin-ubuntu precise ubuntu-patch > > Dear Maintainer, > > libvirt uses xend (xm stack) to manage Xen instances. Even when using > any emulated NIC, there will always be a paravirt NIC in parallel if the > pci-platform device is present. Newer guests will unplug the emulated > NIC on boot. However there is a bug in the stack that will cause the > paravirt NIC to have no MAC address when "type=ioemu" is used in the > vif definition. On the other hand, even without this keyword, the > emulated NIC will be provided. > > Current libvirt already has a quirk to drop this when using no specific > NIC model (defaulting to rtl8139). But the problem exists as well when > using any other emulated NIC. > > So currently, when booting a 3.0+ kernel after setting up the instance > with the default NIC model, everything works. Changing the model to > something like e1000, then the guest cannot use the paravirt interface > after unplugging the emulated one. Ending up with no network. > > *** /tmp/tmpay_5Gn/bug_body > In Ubuntu, the attached patch was applied to achieve the following: > > The parch replicates the quirk used to prevent the usage of "type=ioemu" > to be used for the case of using a NIC model name explicitely. > I think it should be useful/working for Debian as well. Sorry if this > is not the ideal justification. Since this is the first time I use > submittodebian, I am not sure how this exactly should be done. > > > [ Stefan Bader ] > * Never use type=ioemu for NIC definitions. It is not needed > and actually breaks the paravirt interface which always gets > created in parallel.
What's the upstream status of this issue? Please push this upstream directly - this doesn't look like anything debian specific. -- Guido -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
