Red Hat Bugzilla – Bug 98499
tg3 with bcm5703x doesn't negotiate correctly
Last modified: 2008-10-06 12:25:22 EDT
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2.1) Gecko/20030225
Description of problem:
background: we disable n-way autonegotiation on our FE switch ports, and we
force the interface to 100FD from the host. When we first boot up with the
Broadcom 5701 interfaces, the interface links at 100HD until we use ethtool to
force it to 100FD. This is fine, since we get link even with the duplex
mismatch, so we can login to the box and fix the problem.
However, with the Broadcom 5703X chipset (on the motherboard on the HP ML370G3),
the interface tries to come up at 10HD. Since the switchport is 100FD, we don't
get any link at all. We then have to login to local console and force 100FD
manually before we can apply our server management tools to the machine.
It would be cool if the BCM5703X would act like the 5701 and come up at 100HD at
post-install modules.conf options don't get taken.
Version-Release number of selected component (if applicable):
negotiates to 10HD when it should find 100FD (driver behaves differently on 5701
Steps to Reproduce:
1. configure tg3 to come up as 100FD in modules.conf
Actual Results: bcm5703x comes up as 10HD
bcm5701 comes up as 100HD
Expected Results: both come up at 100FD
post-install lines in modules.conf don't get taken.
We're really suffering from this, too.
The tg3 module is unreliable like Starfighters (if you still know
See Service requests #267715 and #265063 for this...
Just two days ago I had another box not getting the primary interface
up, equal what I tried....
They even tend to not even give a status on driver initialization:
tg3.c:v1.2e4 (Feb 20, 2003)
eth0: Tigon3 [partno(NA) rev 1002 PHY(5703)] (PCIX:100MHz:64-bit)
10/100/1000BaseT Ethernet 00:0b:cd:b1:45:cf
eth1: Tigon3 [partno(NA) rev 1002 PHY(5703)] (PCIX:100MHz:64-bit)
10/100/1000BaseT Ethernet 00:0b:cd:b1:45:df
tg3: eth1: Link is up at 100 Mbps, half duplex.
tg3: eth1: Flow control is off for TX and off for RX.
Sometimes the driver even tells you everything's fine, but you get no
Sometimes the switches don't see a MAC address although you have a
running ssh session on the interface...
This is crap especially because all our (accepted) vendors (HP & IBM)
provide these as onboard chipsets now...
Who decided to write a driver without any options (like
"speed_duplex" for the e100(0) or "line_speed" etc. for bcm5700?
BTW: I think this is kind of a duplicate of #111250
Nick (Gunnar) Bluth
DrKW Linux team
*** This bug has been marked as a duplicate of 111250 ***
111250 is opened against RHEL3. Reopening this ticket in order to
track tg3 issues separatly for AS2.1.
Almost forgat this ;-)
It showed up in the meantime that Linux boxes with tg3 cards are the
only ones that negotiate correctly with Cisco switches set to
autonegotiation... (which no U**X host here ever managed to).
It's normally ok to just leave the switchport set to "Auto".
In some cases we do a
ethtool -s ethX speed 100 duplex full autoneg on
which works fine.
included in U6 beta
User firstname.lastname@example.org's account has been closed