21143 - Tx hung

[Brian D. Haymore]

I've seen the same effects so we do two things.  Because RH still uses .89
in their dist we install over the de4x5 driver and place a hub (can we say
poor mans hack?) between the box and the switch to install.  Then I use
version .91 (it isn't quite as fast as .91g and newer but it does crash)
and I have had very reliable performance.

--
Brian D. Haymore
University of Utah
Center for High Performance Computing
155 South 1452 East RM 405
Salt Lake City, Ut 84112-0190

Email: brian@chpc.utah.edu - Phone: (801) 585-1755 - Fax: (801) 585-5366

On Sat, 9 Oct 1999, Vojtech Pavlik wrote:

> Hi!
> 
> I've recently got two problematic Tulip type cards here. They
> have the Intel 21143-PD chip here (no longer the DEC mark on it),
> and a Davicom DM9101F MII transceiver. They are simple 100baseTX cards.
> 
> I've got them, because there were random crashes happening with them.
> So I did some testing:
> 
> de4x5.c v0.544 - Doesn't work when 100 mbit full duplex
> 	        - configures the MII wrong
> tulip.c v0.89H - Seems to do the same as above, plus the TX goes
>                  into an endless loop
> tulip.c v0.91g - Works, but can be made to crash
> 
> The latest tulip.c seems to work OK most of the time, but
> 
> tcpspray -v -e -b 10485760 -n 100 -f random test & ping -f test & ping -f test & ping -f test
> 
> takes it down (crash) in about 5-10 seconds. 'test' is a test machine with
> the 21143. The other end is a 3c905-B Cyclone 100baseTX card, because I have
> never had a crash with those. The 21143 card stops transmitting any data
> after that, and either a 
> 
> eth0: Tx hung, xxxxx vs xxxxx
> 
> message or nothing appears. Any ideas what should I try to find a solution
> for this problem? I do have quite some experience with kernel programming
> and am willing to do more tests ...
> 
> Vojtech
>