[realtek] Abnormal interrupt

[Donald Becker]

On Sun, 20 Aug 2000, Markus Westergren wrote:

> I have a computer, acting as a firewall, with two Realtek NIC's. The
> external (eth0) is connected to a 10 Mbit switch. The other (eth1) is
> connected to my internal network at 100 Mbps/full-duplex. Eth1 is working
> but I have problem with eth0. 
>  If i insert the rtl8139 driver(version 1.10) without options it will
> place the card in half-duplex which results in massive amounts of 
> collisions.

This is expected when the link partner is set to forced-full-duplex.
The only time full duplex should be used is with autonegotiation.

> I can then run 'mii-diag -F 10baseT eth0' and the collision problem goes
> away.

The v1.11 driver allows this from a module or LILO option with the setting
 0x10	  10baseT-FDX
 0x20	  100baseTx-HDX
 0x30	  100baseTx-FDX

The v1.12 driver will also allow the forced-full-duplex flag to be set by
doing 
  mii-diag -F 10baseT-FDX
Currently this sets the transceiver to 10baseT, but the full duplex bit is
ignored by the NIC.

> According to the system amin the switch is configured to 10 Mbps
> full-duplex so this explains the collisions.

Warn the sysadmin that forcing full duplex causes exactly these kinds of
headaches.  The real-world performance improvement of full duplex is minor
for workstations, so the switch ports should be left at half duplex.

> When running the card in 10
> Mbs full-duplex, the following is printed in the system log:  
> 
> Aug 20 15:11:50 220a kernel: eth0: Abnormal interrupt, status 00002002.
> Aug 20 15:11:50 220a kernel: eth0: Abnormal interrupt, status 00000002.

You have receive errors, likely CRC errors.
These are Bad.

Check the cables.

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Beowulf-II Cluster Distribution
Annapolis MD 21403