found 506419 2.6.26-8 thanks Bastian Blank helpfully diagnosed the problem to be with the forcedeth driver. He pointed me at commit edcfe5f[0], which should be in 2.6.26.4, but apparently it's either not present in Debian's 2.6.26-10 package, or it does not fix the bug entirely. 0. http://repo.or.cz/w/linux-2.6.git?a=commitdiff;h=edcfe5f7e307846e578fb88d69fa27051fded0ab The issue doesn't seem new[1] and Bastian thinks it might have to do with bridging, which is in fact being used on my system. 1. http://kerneltrap.org/mailarchive/linux-netdev/2008/10/16/3674714 khyber:~# brctl show bridge name bridge id STP enabled interfaces br0 8000.003048c2f5e8 no eth0 He suggests it's to do with checksum offloading and asked me to experiment with disabling the offloading: khyber:~# ethtool --show-offload eth0 Offload parameters for eth0: rx-checksumming: on tx-checksumming: on scatter-gather: on tcp segmentation offload: on udp fragmentation offload: off generic segmentation offload: on khyber:~# ethtool --show-offload br0 Offload parameters for br0: Cannot get device rx csum settings: Operation not supported rx-checksumming: off tx-checksumming: on scatter-gather: on tcp segmentation offload: on udp fragmentation offload: off generic segmentation offload: off The problem goes away if I turn all features off, on either one interface, or on both together, e.g. khyber:~# i=eth0; set -x; for s in rx tx sgo tso ufo gso; do ethtool -K $i $s off; done; set +x + for s in rx tx sgo tso ufo gso + ethtool -K eth0 rx off + for s in rx tx sgo tso ufo gso + ethtool -K eth0 tx off + for s in rx tx sgo tso ufo gso + ethtool -K eth0 sgo off no offload settings changed + for s in rx tx sgo tso ufo gso + ethtool -K eth0 tso off + for s in rx tx sgo tso ufo gso + ethtool -K eth0 ufo off Cannot set device udp large send offload settings: Operation not supported + for s in rx tx sgo tso ufo gso + ethtool -K eth0 gso off + set +x The problem can indeed be fixed with ethtool -K eth0 tx off or so it seems. Unfortunately, this reduces the throughput rate by a noticeable amount. The problem also exists with 2.6.26-8, although I have not noticed it before last night's upgrade to -10. Thanks, Bastian, -- .''`. martin f. krafft <madduck@debian.org> : :' : proud Debian developer, author, administrator, and user `. `'` http://people.debian.org/~madduck - http://debiansystem.info `- Debian - when you have better things to do than fixing systems
Attachment:
digital_signature_gpg.asc
Description: Digital signature (see http://martin-krafft.net/gpg/)