3c59x compile problems + Question regarding 3C905C overruns in multi-switched environment

[Donald Becker]

On Sun, 18 Aug 2002, Jim Matthews wrote:

> I am trying to compile Donald Becker's latest 3c59x driver into the
> 2.4.18 kernel.
...
>  I get 0 overruns in a single switch
> environment with identical machine hardware and configuration.  These
> overruns are definitely affecting performance, some MPI jobs run as much
> as 3 times faster on the machines w/o overruns.  

You should probably move this to the vortex at scyld.com mailing list,
since it's more of a driver problem than a MPI problem.

> I have made the following changes to the
> drivers/net/Makefile to include support for pci-scan:
> 
> 20c20
> <                       net_init.o mii.o pci-scan.o
...
> 66c66
> < obj-$(CONFIG_VORTEX) += 3c59x.o pci-scan.o mii.o

You don't need "mii.o" support for the updated driver set.  (The mii.o
file contains a few bogus support routines -- functions that need not
and should not be done in the kernel.)

> When I compile and link I get lots of undefined references (see below).
> What do I need to be doing in order to resolve these symbol errors?

They look like module version problems.
What was the compile line for the 3c59x.c driver?

> Another question (probably a question for Donald Becker).  The reason
> why I am trying to compile a new driver is because I have been seeing
> lots of (1%) overruns on Dell precision 340s using the 3C905C card in a
> multi-switched environment.

Rx overruns?  That should only happen if the PCI bus is overloaded.  The
3c905C should be at least as efficient as the 82559, so there must be a
bus setting that differs between the two.  What does the /proc/pci entry
report?  (Or you can use 'lspci -v'.)


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