Fernando F. ManceraMar 9

The kernel patch series removing IPV6=m is out! I am quite happy about it and I hope I did not mess up :)

A lot of people were CC'ed, sorry about the noise ^^

https://lore.kernel.org/netdev/2026030[email protected]/

[PATCH 00/10 net-next] Convert CONFIG_IPV6 to built-in and remove stubs - Fernando Fernandez Mancera

Show threadFernando F. ManceraMar 9

It seems I messed up. fs_initcall() is not good enough as under certain situations there might be a race condition when booting up.

Replacing it with device_initcall() seems good enough.

One thing I do not understand is why this doesn't happen on my environment but it does happen on virtme-ng..

Show threadMarcos Paulo de Souza

@ffmancera doesnt 9pfs setups network early, before calling init, making it fail this way?

That would also depend on how you invoke virtme-ng, I'm just guessing here :)

Mar 9 at 3:34pmWeb
Show threadMarcos Paulo de SouzaMar 10
@ffmancera please tell us your findings :)
Show threadFernando F. ManceraMar 10

@mpdesouza So in essence, this seems a race condition as after booting several times on my libvirt environment I managed to catch it.

I think it is related to addrconf checking on loopback or some other uninitialized device..

After looking at the code using device_initcall() makes more sense.