Re: Via KT133 pci corruption: stock 2.4.18pre2 oopses as well

[Andrew Morton]

Ville Herva wrote:
> 
> On Wed, Jan 09, 2002 at 01:00:53PM -0800, you [Andrew Morton] claimed:
> > Ville Herva wrote:
> > >
> > > >>EIP; c0131ce0 <sync_page_buffers+10/b0>   <=====
> >
> > Looks like a corrupted `next' pointer in the page's buffer_head
> > ring.  Your report is identical to Todd Eigenschink's repeatable
> > oops.  http://www.uwsg.iu.edu/hypermail/linux/kernel/0112.3/0689.html
> <snip>
> > I am able to trigger this in around ten minutes on 2.4.13 and
> > later kernels.  However 2.4.13-pre6 ran the test for nine hours
> > and did not fail.
> 
> Out of curiosity: what kind of load do you use to trigger it?

Massive VM load and ext3.  I've found the buffer-list destroyed
bug.  It's incorrect buffer locking in ext3.  It used to work,
sleazily, but blockdev-in-pagecache pulled its pants down.

> > I've put the 2.4.13-pre6 -> 2.4.13 diff at http://www.zip.com.au/~akpm/1.gz
> 
> Seems your diff didn't include some bits (Maintainers changes and something
> else.)
> 
> Anyhow, I compiled 2.4.13pre6 and it collapsed in just a few minutes. My
> best guess is that network card pci dma is somehow fubar, and it writes
> stuff to where it shouldn't.

OK.  Looks like they're different things - you have hardware problems,
I have brain problems.

-