Re: Transmit timeouts on 440GX Ocotea on 10baseT-HD network

On Wed, Dec 07, 2005 at 08:15:28AM -0700, Wade Farnsworth wrote:

Hi Eugene,

I'm seeing some "NETDEV WATCHDOG: eth0: transmit timed out" messages on
the Ocotea when it's connected to a 10baseT hub, and it's put under
heavy load.  I'm using the most current 2.6 git tree.

This can be reproduced by ssh'ing into the Ocotea and running the
command "ping <host-machine> -A -s 1200", then also doing the same ping
command from the host to the Ocotea.  The pings will be successful for a
short time, then all transmits on the Ocotea will stop for a few seconds
(usually preceded by a few duplicate packets).  Transmits begin again
once the timeout occurs. /proc/net/dev doesn't report any errors, just a
few dropped packets.

Do you know what might be causing the EMAC to stop transmitting in this
situation?

Hmm, I think this should have been fixed in the latest 2.6. Check 
that you tree has this patch:

	[PATCH] ibm_emac: fix graceful stop timeout handling

It went in on 01 Dec. Although this fix assumed FDX operation for 
timeouts, without collisions. Maybe I was too optimistic thinking that 
nobody uses 10/HDX :).

Try making STOP_TIMEOUT_10 bigger, say twice as big.

If this doesn't help, I'll send you patch with enables some additional 
debugging, so I can check that stop you are experiencing is the same 
problem I had last month.

-- 
Eugene