On Wed, 2012-04-04 at 09:55 +0200, Josip Rodin wrote: > On Mon, Apr 02, 2012 at 05:22:37AM +0100, Ben Hutchings wrote: > > On Sun, 2012-04-01 at 12:40 +0200, Josip Rodin wrote: > > > On Sun, Apr 01, 2012 at 03:09:56AM +0100, Ben Hutchings wrote: > > > > I bet this is due to the combination of LRO plus bridging. We try to > > > > turn off LRO in devices under a bridge, but that won't work if there's > > > > an intermediate bonding device. > > > > > > > > If you run: > > > > > > > > # ethtool -K eth0 lro off > > > > # ethtool -K eth2 lro off > > > > > > > > does the bridge start working? > > > > > > Err... > > > > > > % sudo ethtool -K eth0 lro off > > > Cannot set large receive offload settings: Operation not supported > > > % sudo ethtool -K eth2 lro off > > > Cannot set large receive offload settings: Operation not supported > > > > Hmm. Well it shouldn't be a problem but you could try also turning off > > GRO (similar commands). > > Ah, there we go. Once I ran sudo ethtool -K eth0 gro off, > sudo ifenslave bond54 eth0 produced a still-working bond54.
OK, this is quite unexpected. At least you have a workaround now (/usr/share/doc/ethtool/README.Debian.gz explains how to make this setting persistent). > > > That's with eth0 removed from bonding, and eth2 inside. > > > > So the bonding device has only one slave now? > > Yes, it was like that. > > > What if you take the bonding device out completely and add eth2 directly > > to the bridge? > > I think I had already tested that and everything was fine, too. > Do you want me to test that or is the GRO removal conclusive? No need to test that. I would like to take this upstream now, but first I need to check whether it has already been fixed after 2.6.32. Please can you test the current kernel package from testing, unstable or squeeze-backports (linux-image-3.2.0-2-amd64 or linux-image-3.2.0-0.bpo.2-amd64)? Ben. -- Ben Hutchings Larkinson's Law: All laws are basically false.
signature.asc
Description: This is a digitally signed message part