em regression, UDP LOR followed by ssh stall

John Baldwin jhb at freebsd.org
Fri Apr 16 12:22:30 UTC 2010


On Friday 16 April 2010 5:37:59 am Harald Schmalzbauer wrote:
> Jeremy Chadwick schrieb am 16.04.2010 11:28 (localtime):
> > On Fri, Apr 16, 2010 at 11:18:56AM +0200, Harald Schmalzbauer wrote:
> >> with RELENG_8 from 6 weeks ago I never ran into the problem that my
> >> ssh connection stalled.
> >> With today's RELENG_8 it reproducably hangs at first login. After
> >> some time I can open another ssh session which seems to stay without
> >> problems, but the first sessions is always dying a few seconds after
> >> login.
> >> here's a LOR:
> >> {snip}
> > 
> > The e1000/em driver was recently modified (heavily).  I saw the large
> > number of commits come across in a csup a few weeks ago, and there's
> > even more coming across in recent days (~38 hours ago):
> > 
> > http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/dev/e1000/
> > 
> > This is the 2nd problem report to come in about the recent e1000/em
> > changes.  Because of this, I've avoided building kernel on all of my
> > systems, and would recommend others avoid doing the same until these
> > problems can get worked out.
> > 
> > Jack, can you shed some light on what's going on here?
> 
> Som more info:
> 
> it seems only one of my em interfaces is affected. The other interface 
> doesn't show any odd behaviour:
> em1: <Intel(R) PRO/1000 Legacy Network Connection 1.0.0> port 
> 0x1800-0x183f mem 0xe1920000-0xe193ffff,0xe1900000-0xe191ffff irq 18 at 
> device 2.0 on pci3
> em1: [FILTER]
> em1: Ethernet address: 00:15:17:8d:aa:d4
> 
> I only have one
> em1: Watchdog timeout -- resetting

Try pulling in the latest version of if_lem.c from HEAD.  There was a bug in 
if_lem.c that was missing an assignment that caused spurious watchdog resets.

-- 
John Baldwin


More information about the freebsd-stable mailing list