Re: [PATCH 0/12] virtio_net perf patches

[Mark McLoughlin <markmc@redhat.com>]

On Mon, 2008-08-11 at 15:30 -0500, Anthony Liguori wrote:
> Mark McLoughlin wrote:
> > Hi Avi,
> >         Here's the set of patches that I think make sense to apply.
> >   
> 
> We probably need to disable checksum offload on the RX side until we 
> figure out what to do about the broken dhclient.  That's going to hit a 
> lot of users otherwise.

If I could reproduce this, I'd get right on it ... but I'm not seeing
the issue here.

Wait, wait, wait. Bells are going off all of a sudden :-)

Yes, I've been through this before. See:

  https://bugzilla.redhat.com/231444

So, we've had this long-standing dhclient patch in Fedora:

  http://cvs.fedoraproject.org/viewcvs/*checkout*/rpms/dhcp/devel/dhcp-4.0.0-xen-checksum.patch

Herbert - any clue why this isn't upstream? That's quite surprising ...

Ah, I see Rusty moved this to linux-netdev without cc-ing:

  http://marc.info/?l=linux-netdev&m=121844837826895

Herbert wrote:

  "One easy way of doing this is to hook up the rx checksum offload
   option in the guest with the tx offload option in the host.  In
   other words, when rx offload is enabled in the guest we enable
   tx offload in the host, and disable it vice versa."

Are you basically just saying that guests with a "broken" dhclient
should manually disable rx checksum offload with ethtool? And that the
host should react by disabling tx offload on the tap interfacE?

Cheers,
Mark.