Regression with txcsum/rxcsum on vge(4) drivers on 8.0-Release

Pyun YongHyeon pyunyh at gmail.com
Fri Jan 15 17:45:41 UTC 2010


On Fri, Jan 15, 2010 at 10:05:09AM -0700, Michael Loftis wrote:
> 
> 
> --On Friday, January 15, 2010 10:32 AM +0100 Olivier Cochard-Labb?? 
> <olivier at cochard.me> wrote:
> 
> >Hi,
> >
> >I've just upgraded on of my server from 7.2 to 8.0-Release and meet a
> >problem with the vge(4) drivers:
> >All my SCP transferts didn't works since this upgrade: they close, after a
> >random time, with "Corrupted MAC on input" message.
> >And Putty SSH tunnel closed with "Incorrect MAC received on packet".
> >
> >I need to disable txcsum and rxcsum on the vge network card for solving
> >this problem.
> 
> nfe(4) has complete deadlocks with checksums enabled under high  transmit 
> (and possibly receive) loads.  It appears to be a regression from 7.2 as 
> well.  We think it might have something to do with error recovery, but 
> aren't sure if it's related to *just* nfe(4) or to HW CSums in general.
> 

Please open new thread for this issue. I'm not aware of this and
this is the first time I heard for the checksum offloading issue of
nfe(4). Include dmesg output for your controller and let me know
whether it's really regression from 7.2-RELEASE. If you have
reliable way to reproduce the issue, please also let me know.
There is no nfe(4) source code differences between 7.2-RELEASE and
8.0-RELEASE.


More information about the freebsd-stable mailing list