cvs commit: src/sys/dev/em if_em.c

[Pyun YongHyeon]

On Wed, Aug 23, 2006 at 03:09:11PM +0200, Stefan Bethke wrote:
 > Am 22.08.2006 um 21:02 schrieb Patrick M. Hausen:
 > 
 > >em interface, FTPing a large file while putting disk and cpu/memory
 > >load on the box -> occasional but easily reproducable complete hangs
 > >of network communication (ssh, ping - no reaction). Every time a
 > >
 > >em0: Missing Tx completion interrupt!
 > >
 > >is logged.
 > 
 > I have no idea whether the effect I'm seeing is related at all, but  
 > the description fits.  If I load the CPU (make world -j4 or gzip -9 </ 
 > dev/random >/dev/null), traffic over re0 stops and I get
 > 
 > Aug 22 09:30:20 diesel kernel: re0: watchdog timeout
 > Aug 22 09:30:20 diesel kernel: re0: link state changed to DOWN
 > Aug 22 09:30:23 diesel kernel: re0: link state changed to UP
 > Aug 22 09:30:35 diesel kernel: re0: watchdog timeout
 > Aug 22 09:30:35 diesel kernel: re0: link state changed to DOWN
 > Aug 22 09:30:39 diesel kernel: re0: link state changed to UP
 > 
 > until I stop the CPU hog; then, it appears to be working fine again.   
 > I've got two rl's in there as well, and they appear to be completely  
 > unaffected.
 > 

Hmm... I guess it's not related with above fake watchdog timeout
error. Because re(4) does not make use of Tx interrupt moderation
by default, it might be other issue.(You need to define
RE_TX_MODERATION to enable Tx interrupt moderation.)

There is a WIP version at the following URL.
http://people.freebsd.org/~yongari/re.link.patch

-- 
Regards,
Pyun YongHyeon