Bug#525296: other stuff I tried
First, I updated to 2.6.26-2-686 (from Lenny), but this has the same
problem for me. No command I know will set that NIC to full duplex
under that kernel.
I tried building a custom 2.6.29.1 kernel next, but it panics when
it can't mount /dev/md0 as the root file system. I'm not in the mood
to try to figure out how all this initrd stuff works. Too much work.
Next, I rebooted into etch's 2.6.18-6-686 kernel (this is an etch->lenny
box, with the old kernel still there). This works fine for me, as far
as I can tell so far. The NIC goes into full duplex mode due to the
"up sleep 5; mii-tool -F 100baseTx-FD eth0" line I have in
/etc/network/interfaces, and all the things that should be running
(NFS, Samba) appear to be running.
Then, I tried building upstream tg3 drivers against the 2.6.26-2-686
kernel headers. I was able to build tg3-3.92n and tg3-3.98e modules
successfully, but I was not able to build 3.77e or 3.81e, nor was I
able to build the tg3.{c,h} sources yanked out of a kernel.org
2.6.18.8 source tarball -- all of those failed with messages like
"error: implicit declaration of function 'netif_poll_disable'" (and
many more).
I rebooted into 2.6.26-2-686 again, unloaded the tg3 driver module,
and loaded the tg3-3.92n and tg3-3.98e modules in turn, and attempted
to set the NIC to full duplex using mii-tool and ethtool. All of
these attempts failed. The NIC was left in half duplex mode after
bouncing up and down for a few seconds. (Except in the case of
3.98e + ethtool, where the command aborted with an error without even
touching the NIC.)
For now, I am running on 2.6.18-6-686 since this kernel's tg3 driver
supports what I need.
Reply to: