Andreas Dilger <adilger@xxxxxxxxxxxxx> writes:
> On Sep 07, 2003 16:21 -0600, Eric W. Biederman wrote:
> > Below is one good oops we have captured. I would have to check but
> > I believe we have updated the tg3 driver in this instance to the
> > one that comes with 2.4.23-pre3.
> > The very puzzling part is that in the crashes I don't see the tg3
> > driver at all just the network stack. All module addresses according
> > to /proc/ksyms started with at 0xf8, and the tg3 driver was built as
> > a module.
> > I have been having trouble understanding why skb_clone would be called
> > to transmit a packet. Any ideas?
> Do you have the stack overflow checking enabled?
> That has been a source
> of problems for us. It was especially difficult to reproduce, because it
> only happened during a double interrupt.
So a quick update on what I am seeing.
I have now tried with a myriad of driver and kernel versions watching very
carefully for a pattern.
What I have observed is memory corruption in what looks like it may
be a confined area of memory. The ECC SDRAM is being closely
monitored and I am not getting as much as a correctable error much
less and uncorrectable error that would show up so memory is ruled
out. When I am connected to a particular Extreme Networks gigabit
switch. (The switch has some problems and it is hypothesized the
switch is injecting problematic packets into the network).
Bad packets should not be a problem but it looks like they are
triggering the problem whatever it is.