2000-11-01 23:37:23

by Markus Fischer

[permalink] [raw]
Subject: [2.2.17] 3c59x and Transmit list error

Hello,

For some time now we're observing a very bad [tm] network
problem with our 3Com 3c905B Cyclone card. After booting the
network works just fine, but after a few days ( ranging from five
to 10) really hard network problems occur, rendering the network
accessabilty to zero; only manually restarting the network solves
this problem (from local, of course).

I appended the relevant lines of the logfiles for a more
informative description. Altough the date/time stamps are a bit
old (a few feks) the problem still exists recurrence every few
days now :/

I allready searched the mailing list and I'm aware of scylds [1]
website, but unfortunatly none of this information helped us.

kind regards,
Markus

[1] http://www.scyld.com/network/vortex.html

--
Markus Fischer, http://josefine.ben.tuwien.ac.at/~mfischer/
EMail: [email protected]
PGP Public Key: http://josefine.ben.tuwien.ac.at/~mfischer/C2272BD0.asc
PGP Fingerprint: D3B0 DD4F E12B F911 3CE1 C2B5 D674 B445 C227 2BD0


Attachments:
(No filename) (0.99 kB)
kernel.msg (4.70 kB)
Download all attachments

2000-11-02 08:28:50

by Andrew Morton

[permalink] [raw]
Subject: Re: [2.2.17] 3c59x and Transmit list error

Markus Fischer wrote:
>
> Hello,
>
> For some time now we're observing a very bad [tm] network
> problem with our 3Com 3c905B Cyclone card. After booting the
> network works just fine, but after a few days ( ranging from five
> to 10) really hard network problems occur, rendering the network
> accessabilty to zero; only manually restarting the network solves
> this problem (from local, of course).

eth0: Transmit error, Tx status register 82.

This is an "out of window collision", otherwise known
as a "Transmit reclaim error".

It is almost certainly caused by another machine on
your network entering full-duplex mode. It stomps
on the 3c905's Tx packets and the collision is
detected too late for the 905 to be able to
retransmit - some of the packet contents have
been discarded from the on-chip FIFO.

Next time it happens, you need to run around the
building and find out who just did something such
as turning their computer on, or mucking around
with dangerous Linux netdriver module parameters.

If this description doesn't match your situation
please let me know and we'll work on it.