3com NIC problems with interface configuration

[Andrew Morton]

Guys,

I'm posting here way too much.  I was #3 poster on the kerneltraffic
hall of shame last week :(

In future I'll be pushing most of the 3com NIC discussion onto
linux-vortex@beowulf.gsfc.nasa.gov

Interested parties will find subscription info at
http://www.scyld.com/network/vortex.html#mail

"Allen K. Smith" wrote:
> 
> I fixed the autonegotiation stuff by just adding |HAS_NWAY to the 905C
> definition in 3c59x.c

That's the one.  Hopefully it will fix Igor's negotiation problems.  My
latest&greatest 2.2 driver is at
http://www.uow.edu.au/~andrewm/linux/2.2.16-pre2-3c59x.c.gz . It
includes the HAS_NWAY fix.

Please try it out - I'd be interested in seeing if it fixes Lee's Tx
stalling problems.  These are usually Tx timeouts and a couple of causes
of this were nailed in this patch.  One cause still remains; in the 2.3
driver I'm using a TxReset to recover from 16 successive collisions. I
didn't backport this into the 2.2 driver because it's damn rare and the
fix is contrary to the published documentation...  

> I'm not seeing any stalls here, although I do get the:
> 
> eth0: 3Com 3c905C Tornado at 0xe880,  ***INVALID CHECKSUM 005f***
> 00:c0:4f:04:f0:28, IRQ 10
> 
> checksum message.

You and half the rest of the world :)

I think there may be a logic problem in the reportage of this checksum. 
If you run 'vortex-diag -ee' (http://www.scyld.com/diag/#pci-diags)
you'll see that it displays the results of three different checksum
algorithms.  If they're all wrong then yeah, your EEPROM has been
corrupted.  It's usually benign though.

-- 
-akpm-
-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-vortex-request@beowulf.org