[3897] in linux-net channel archive
Re: linux-2.0 and eexpress net card
daemon@ATHENA.MIT.EDU (Philip Blundell)
Mon Jul 29 12:18:17 1996
Date: Mon, 29 Jul 1996 17:10:57 +0100 (BST)
From: Philip Blundell <pjb27@cam.ac.uk>
To: "Darrin R. Smith" <drsmith@eznet.net>
cc: linux-net@vger.rutgers.edu
In-Reply-To: <31F80783.59F75D8C@eznet.net>
On Thu, 25 Jul 1996, Darrin R. Smith wrote:
> transfer will appear to hang and I get a message indicating, "CU wedged
> - status xx4x xxxx - resetting..."(note that the 4 is the only character
> in the status that is always the same - the other x's represent
> changing numbers). Sooner or later, the machine locks solid and I
> have to power cycle it.
>
> This is on linux-2.0, and it appears irregardless of what
> options are chosen in the configure script(well, lets just say I've
> recompiled this kernel more than 12 times with different options :( ).
> I looked in the eexpress.c file and it indicates there that this is
> a problem with the driver, but it also says this should be a rare
> occurrance -- not every single time. telnet/rlogin sessions work fine
> most of the time(Hmm -- maybe we should put a short duration sleep
> in the send code and see what it does since tcpdump doesn't give me
> any problems recieving packets? -- ideas? suggestions? )
It's a known problem. Unfortunately it's also a "your mileage may vary"
problem -- some people are using those cards with no problems, some
people get occasional crashes under high load, and some people find them
totally broken. The receive side seems fine, but I suspect there is a
race condition or a subtle bug in the transmit code which is upsetting
the 82586 from time to time. The secondary problem is that the recovery
code we use when a problem turns up is fairly broken, and often just
stiffs the machine altogether rather than getting things back to a
working state (mostly because it tries to reset the card, which is itself
a rather hazardous undertaking for mysterious reasons).
As yet I haven't managed to work out what goes wrong, which has been
mostly due to the fact that I haven't had access to any
EtherExpress-equipped machines that I can conveniently hack on, and also
partly because I've been far too busy with other things recently. At this
moment, though, I have an EtherExpress card sitting on my desk at home,
and so I hope to get somewhere with this in the next few days (work and
stuff permitting). Watch this space.
Incidentally, John Sullivan has now left Cambridge, and I believe his
js10039@cam.ac.uk account has been cancelled (so mail to it will
bounce). I am looking after the eexpress driver to all intents and
purposes these days, so it would be best to send messages about it to me
(and optionally linux-net) rather than him and linux-kernel. :)
phil
