[3120] in Release_7.7_team
Status report GX150 Net thing
daemon@ATHENA.MIT.EDU (Bill Cattey)
Wed Jan 30 17:42:43 2002
Message-ID: <swK7PUJz000190nOoR@mit.edu>
Date: Wed, 30 Jan 2002 22:42:40 +0000 ()
From: Bill Cattey <wdc@MIT.EDU>
To: release-team@MIT.EDU
I looked at some GX150's today to try and get a better handle on the
failure. I learned that power cycling the Test Cluster GX150,
gorf.mit.edu nearly always results in a system that mis-negotiates its
connection.
I used my vortex-diag tool and looked at the Manufacture Date of the
3com chipset in Gorf, Tokata, and Athenaphobia. They are ALL 6/10/2000.
Except that Athenaphobia seems not to have a problem with its
autonegotiate.
cfox's system also seems to have not trouble.
I note in passing that Tokata and Gorf were installed 9.0,
and that cfox's system and Athenaphobia were installed 8.4.
THING TO TRY: Install gorf at 8.4 and see if it still likes to screw up.
----
BTW: All 4 systems complained of
***WARNING*** No MII transceivers found!
in dmesg.
Perhaps if we can get this error message to go away, it will be a
reasonable test case for us. (It's almost too much to hope for.)
----
Gorf was manifesting network performance problems even while BOTH
mii-diag and the stats run by net-ops said that it was 100base-Tx-FD.
We NEED a definitive test case.
I've been doing an FTP of a 3MB file. It takes 2 seconds when all is
well, and FOREVER if there is a disagreement.
----
I attempted to create a kernel module building environment on my home
machine, but have failed. The 3c59x.o that I build complains of
unresolved symbols.
Somebody else, amb perhaps, needs to try and create a build environment,
or to nurse maid me through the process of building a module that, built
from source, reproduces the fault.
----
Once version 1.1.16 of 3c59x.o is successfully built and tickling the
bug on gorf, we can take 1.1.17 out of /mit/wdc/3com/new and build it
and see if
the bug is still tickled. If so, we can run in the patch that is
embodied in /mit/wdc/3com/new/patch and see what happens. It is that
file 'patch' that Andrew Morton forwarded us.
I am currently unable to make much of any more progress until the clue
fairy helps me get that 3c59x.o object file built.
-wdc