[3161] in Release_7.7_team
Re: Case 254686: Poor GX150 performance revisited.
daemon@ATHENA.MIT.EDU (Bill Cattey)
Tue Feb 26 15:49:19 2002
Message-ID: <IwSzHBxz00018Rg3tH@mit.edu>
Date: Tue, 26 Feb 2002 20:49:17 +0000 ()
From: Bill Cattey <wdc@MIT.EDU>
To: net-help@MIT.EDU
CC: release-team@MIT.EDU, owls@MIT.EDU
Status report:
We understand what is causing this problem, and the net-help aspect
can be CLOSED.
The problem is that recent revisions of the chips moved the media
interconnect from an external chip to inside the one chip. The
documentation used to write the Linux drivers was incomplete, and
so the chips are not getting properly initialized.
Anybody with a 3Com 3c905c ethernet NIC manufactured around June 2000 is
at risk here.
If the EEPROM values are set in a certain way (as they seem to show up
with Dell GX150's) the Linux driver will not see the media interconnect,
and will randomly negotiate.
I'm working with the Linux driver community on this.
I have a tool in-hand that can change the EEPROM values to something the
driver CAN handle.
If there's anyone with friends who are 3Com engineers, I'd like to talk
to them because we need to get hold of some lore that IS embodied in
Windows drivers, but heretofore UNKNOWN in the publicly disclosed 3Com
documentation.
-wdc
