Barry Scott <barry.scott@xxxxxxxxxxxx> wrote:I'm willing to help get this fixed. I'm happy working inside kernels and drivers
Under FC4's build of 2.6.13 the broadcom 5752 works well. But when the
I use the FC4 build of 2.6.16 it no long works.
The hardware is an HP dc7600U, P4 2.8GHz CPU, 512MB memory
SATA disk.
mii-tool correctly reports the state of the eth0. Removing and inserting the
cable is reported as expected. But DHCP or static IP configuration does not
work under 2.6.16.
In dmesg output on 2.6.16 I see this message:
ADDRCONF(NETDEV_UP): eth0: link is not ready
I have recompiled the 2.6.13 version of tg3.c for 2.6.16 and that does not fix
the problem.
Looking at /proc/interrupts I see that a lot of difference between .13 and .16 kernels.
Is this related to the problem?
Attached are the output of dmesg and /proc/interrupts on 2.6.13 and 2.6.16 kernels
as well as lspci output.
It appears that the 2.6.13 kernel did not bring up the machine's io-APICs,
but 2.6.16 did. However you are receiving eth0 interrupts on 2.6.16 so
perhaps that's not relevant.
Don't know, sorry - tg3 works OK for most people. You could try booting
with the `noapic' kernel paremeter, perhaps.
Note that googling for "noapic" gets 212,000 hits - we've _really_ screwed
something up in there. Maybe one day some developer will lay hands on one
of these machines and will fix something.
If noapic doesn't work (and I suspect it won't) then a next step would be
to compile a kernel.org kernel and start enabling debug options. It's
hard, when we don't know which kernel subsystem broke.