On Wed, 30 Aug 2017 10:24:52 +0200 Daniel Reichelt <debian@nachtgeist.net> wrote: > On 08/30/2017 12:18 AM, Ben Hutchings wrote: > > I can't reproduce this. What driver is used for eth0 (ethtool -i shows > > this)? > > Ben, > > you're on to something: > > > # ethtool -i eth0 > driver: vif [...] > Indeed, this only happens in XEN guests running on a jessie host (didn't > get around yet to test whether a stretch host makes the difference). > > On any other hw I own (a couple of e1000e and a r8169), ethtool always > exits 0, with changes to offloading having been made or not. > > Sorry for not mentioning the fact about VMs earlier...just didn't think > of it. Returning to this at last… the problem is that the xen-netfront/vif driver always enables TX checksum offload for IPv4 and only allows it to be controlled for IPv6. That might be a bug in the driver or just a limitation of the Xen PV networking protocol. Initially the interface has TX checksum offload enabled for IPv4 only. Running "ethtool -K tx off" can't disable it for IPv4 (failure), but running "ethtool -K tx on" does enable it for IPv6 (success). Running "ethtool -K tx on" and then "ethtool -K tx off" succeeds, but still has no effect on TX checksum offload for IPv4. I don't see this as a bug in ethtool. Ben. -- Ben Hutchings For every complex problem there is a solution that is simple, neat, and wrong.
Attachment:
signature.asc
Description: This is a digitally signed message part