The continuing saga of multicasting from locked up machines...

[Robert Schwartz]

This is a multi-part message in MIME format.
--------------BF4A0EB40435874DC37F4607
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

About 6 months ago Don and the list helped us out with a problem we've
narrowed down as a nasty problem of network multicasts emanating from
exclusively Linux machines running with the eepro100 driver when some
machines where told to shutdown, but not powered off. The resulting
patch at the time came from Donald B. and has made it into his latest
2.3 driver image:

speedo_close(struct net_device *dev) {
....
     /* Shutting down the chip nicely fails to disable flow control.
So.. */
     outl(PortPartialReset, ioaddr + SCBPort);
...
}

We successfully tested this fix and included it in our initial release
of our Corel Linux OS distro, in place of the standard kernel's eepro100
driver. Several months have passed and I thought I'd bring the list up
to date with our findings concerning this problem:

- We are still seeing these flow control multicasts on our network,
every day.
- Most of them are a result of Linux machines running other distro's
(original kernel driver), being shut down but not powered off.
- Some of them are a mix of Corel Linux and other distro's that may have
locking up and not noticed for a period of time.
- NONE of them are Window's boxes.

This last point is what has me a bit concerned. Why are we only seeing
this under Linux? Is it possible that the flow control feature, if this
is in fact the origin of the multicasts, is NOT turned ON in the
Window's driver?? Could it be that by default, Intel may not be setting
up the card the same way Don's drivers are under Linux?

Any feedback would be appreciated. I would also like to ask that if any
other distro builder is on this list that they seriously consider
upgrading their driver with the above mentioned fix. It would make our
IT dudes real happy! I believe that this fix will be making it into the
2.4 kernel (right Don??) If you need more info on what exactly we did,
please let me know. For those that don't know what the result of these
multicasts are; any other eepro100 card on the network seeing this
multicast, would start to throttle back on its transmissions, even to
other hosts. Our main network servers where (past tense, cause IT
changed all the nics!) running eepro100 cards, so this effectively was
slowing down our network.

Thanks,

-Rob


--------------BF4A0EB40435874DC37F4607
Content-Type: text/x-vcard; charset=us-ascii;
 name="roberts.vcf"
Content-Transfer-Encoding: 7bit
Content-Description: Card for Robert Schwartz
Content-Disposition: attachment;
 filename="roberts.vcf"

begin:vcard 
n:Schwartz;Robert
tel;fax:+1 613 761-9338
tel;work:+1 613 728-0826 x1499
x-mozilla-html:TRUE
url:http://linux.corel.com
org:Corel Linux OS;<A HREF="http://linux.corel.com/products/linux_os/download.htm"><IMG SRC="http://linux.corel.com/products/linux_os/graphics/1199_2726_ani_buyordownload_linux.gif"><BR><FONT SIZE=2>The free Corel® LINUX® OS Download is NOW available!</FONT></A>
adr:;;1600 Carling Ave.;Ottawa;Ontario;K1Z 8R7;Canada
version:2.1
email;internet:roberts@corel.com
title:Product Development Manager
x-mozilla-cpt:;-704
fn:Robert Schwartz
end:vcard

--------------BF4A0EB40435874DC37F4607--

-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to linux-eepro100-request@beowulf.org