[eepro100] eepro and nfs file handle

[Donald Becker]

On Thu, 7 Nov 2002, Traylen, SM (Steve)  wrote:

> >> This was after I removed the sleep mode bit as well
> >> # eepro100-diag -G 0 -w -w -f
> >Note that you must *cold* power cycle the machine for this to take
> >effect.
> >That means unplugging it, otherwise the board remains partially powered
> >so that Wake-On-LAN works.

> >Check for errors in /proc/net/dev  -- do not rely on the output of
> >'ifconfig' or 'netstat' which summarize only some of the error types.

> /proc/net/dev all looks fairly healthy, on eth0 there have been a number
> of bytes and packets recieved and transmitted. Everything else is zero
> though.

That is completely healthy.

> This is on a box with both /tmp/mnt and /tmp/mnt1 hung.

OK, there is one other problem source that is specific to certain
implementation and revs of the chip: some versions of the chip have a
firmware bug^H^H^H feature that snarfs UDP packets for IPMI management,
never passing them to the OS.  But NFS packets look a little too much
like IPMI packets, and...

The characteristic of this problem is
  - all other traffic is fine, even at high data rates
  - no errors are reported

I don't know the precise fix, but advice is available on the intel web site.

-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993