PATCH 2.2.15-pre15: tulip fix for 21041 chips

[Donald Becker]

On Wed, 22 Mar 2000, Jeff Garzik wrote:

> > Errrmm, that means that you are not correctly handling autonegotation.
> > The test: Do you negotiate 10baseT-FDX with a switch?  Does the switch know
> > it?
..
> No, don't have a switch, I'll see if I can cajole someone into testing
> that situation.

OK, then just use a cross-over cable to another machine.

Use 'mii-diag' on both ends to set and observe the transceiver settings, for
all possible settings.  The 'mii-diag -w' option exists for this reason.
Physically disconnect the cable while transmitting and receiving.

It typically takes 60 minutes for a 10-only and 90 minutes for a 10/100 card
to test a change to driver transceivers setting.  You will want it to *power
off* the machine to make certain the residual settings don't keep a driver
working -- the reset signal doesn't go to all board areas.

I sometimes don't test the power-off scenerios when just updating a driver,
but with the Tulip driver it can bite you.  The driver is expected to
software reset some transceiver types.  There are four boards types that
have miswired MII transceivers which start up in bogus modes.

The D-Link Rev2 and Rev3, when the first started using Level One
transceivers, are the most notorious.  They advertise 10baseT only when
booted up and when the GP-line-driven reset pin is tickled.  Normally the
driver should *never* retrigger autonegotiation if it has a link, since that
can take up to 2.4/3.0 seconds for media agreement (and perhaps another
minute for the Cisco to start talking to you again).  But in this case the
driver must change the advertised media types, and reset the
autonegotiation.  If you don't actively reset the link this some link
partners (especially older switches) will become very confused.

A key bug report in situations like this is if the adapter "works sometimes"
and "works in Windows".  Ask "does it work when warm booted from Windows?"

Donald Becker
Scyld Computing Corporation, becker@scyld.com

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