[3128] in Release_7.7_team
Re: Case 254686: GX150 autonegotiate problem
daemon@ATHENA.MIT.EDU (Bill Cattey)
Wed Feb 6 19:11:51 2002
Message-ID: <UwMQN39z0001MJkMlj@mit.edu>
Date: Thu, 7 Feb 2002 00:11:47 +0000 ()
From: Bill Cattey <wdc@MIT.EDU>
To: vortex@scyld.com, akpm@zip.com.au
CC: release-team@MIT.EDU
In-Reply-To: <cwKm7Ixz00011azKF3@mit.edu>
We've done some testing, and perhaps those in the know will be able to
account for the behavior, but it has left me more perplexed than ever.
Test 0: Build stock 1.16 3c59x.o from source and confirm the problem
still occurs. RESULTS:
Problem still occurs.
Power cycling machine results in bad initial negotiation.
mii-diag says the negotiated speed is 100Base-Tx-FD.
The full duplex light on the switch is OUT.
Network throughput is TERRIBLE.
Test1: Build new 1.17 3c59x.o from Andrew Morton. RESULTS:
mii-diag says autonegotiation is NOT POSSIBLE and
that speed is fixed at 10Base-T half duplex.
The full duplex light on the switch is OUT.
Network througput is REASONABLE.
Test2: Take JUST the mii-related changes as per Andrew Morton's diffs,
and apply them to the 1.16 source. RESULTS
Same as test 1.
----
The next test to run is to install 1.17 on a GX150 that has no problem
with autonegotiate and see if the new driver ALSO perceives it as having
no MII.
AKPM: What next steps do these test results suggest to you?
-wdc