[2751] in Release_7.7_team
complication from GX150s running Athena 8.4
daemon@ATHENA.MIT.EDU (andrew m. boardman)
Wed May 23 15:40:56 2001
Date: Wed, 23 May 2001 15:40:52 -0400
Message-Id: <200105231940.PAA19313@pothole.mit.edu>
From: "andrew m. boardman" <amb@MIT.EDU>
To: release-team@mit.edu
The 8.4 release (Red Hat 6.2) has two relevant ethernet drivers included
for our hardware, 3c95x and 3c90x. The former works with only the old
hardware, and the latter works with new and old hardware *if* it's forced
to 10Mb/s operation. There is no known easy way to tell the different
types of hardware apart in software.
The options I see now are:
- install the GX150s from a separate miniroot which forces the interface
to 10Mb/s. Advantages are that it lets new non-GX150 installs run
faster (maybe). Disadvantages are maintaining a separate boot server
and complicating the issue for anyone (Brian et al) doing installs.
- change the miniroot on the default boot server and thus force all new
installs to 10Mb/s. Simple to install, but anyone on a fast net,
installing old hardware, who doesn't know about the hack, loses
bandwidth.
- write a script to install one driver, see if it works, and then switch
if it doesn't. This seems a bit gross, but would have the cleanest
result.
Thoughts? I'm for the second. Deadline to have something installed is a
week from now when the 8.4 GX150s start arriving. (The boot image that
implements either of the first two options is done.)
BTW, as a side effect of making this work, I'd be switching new installs
from conf.modules to modules.conf during the install, since the latter
works with some but not all things while the latter works with most
things.
Also, since 9.0 does not have a 3c90x driver, kudzu will automatically
stomp on the 3c90x entry and replace it with 3x59x at boot time. I think
it will also nuke conf.modules entirely and replace it with modules.conf
(this seems to be what has happened in past, and is definitely the right
thing), but I'm not totally sure yet.
