[tulip] TOO MUCH WORK...

Donald Becker becker@scyld.com
Sat Dec 28 18:11:01 2002

On Sat, 28 Dec 2002, Homer Wilson Smith wrote:

>      Running Linux 2.0.38 as a router, Kingston 21140 cards, with tulip
> v0.95.

2.0.38 doesn't usually have this problem -- it usually does a
significantly better job than later kernels.

>      During nightly 'maintenance' runs involving disk copying and
> other jobs, we get this error message.
>      Is this serious?

It's the driver reporting that something is going wrong with some other
part of the system.

> Dec 28 15:00:54 fw1 kernel: eth0: Too much work during an interrupt, csr5=0xfc6980c0.

This is reporting that the kernel is out of network buffers.  This can
occur because
  The system is short of memory
  The kernel is not keeping enough available memory for itself (typical
    with later kernels, not common with 2.0.*)
  Some other driver (usually a SCSI driver) is taking too much time
    during an interrupt.

> Dec 28 15:00:59 fw1 kernel: eth0: Restarted Rx at 1268329 / 1268329.

As long as the NIC doesn't stop, this is mostly harmless.  The system
drops a few packets, which are later retransmitted by the protocol layer. 

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