[tulip] 21140 and Interrupt Mitigation?

[Donald Becker]

On Fri, 11 May 2001, Roisman, Dani wrote:

> I have old netgear FA310TX cards that use DEC 21140 chipsets:
> 
> eth1: Digital DS21140A Tulip rev 34 at 0xb800, 00:40:05:41:84:73, IRQ 10.
> eth1:  EEPROM default media type Autosense.
> eth1:  Index #0 - Media MII (#11) described by a 21140 MII PHY (1) block.
> eth1:  MII transceiver #0 config 1000 status 782d advertising 01e1.
> tulip.c:v0.92t 1/15/2001  Written by Donald Becker <becker@scyld.com>
> 
> I recently upgraded to v0.92t, and unsuccessfully tried to force interrupt
> mitigation:

The 21140 doesn't have interrupt mitigation hardware.
That feature was added in the later 21143 chip.

> I am, however, seeing these log entries:
> 
> May 11 19:25:01 soewfv2 kernel: eth1: Too much work during an interrupt,
> csr5=0xfc630040. 

Some thing is strange on your machine.
This is usually another device driver using taking too much time with
interrupts blocked.  Check that your IDE hard disks have been set with
/sbin/hdparm -u 1 /dev/hda

> May 11 19:25:10 soewfv2 kernel: eth1: Restarted Rx at 1821234174 /
> 1821234174. 

This confirms the above problem -- the interrupt handler is being
blocked so much that the card is running out of receive buffers.
You can avoid the first message by increasing max_interrupt_work, but
you'll still be dropping packets.  Interrupt mitigation will help a
little bit here, but the real fix is to identify the problem driver and
fix it.


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