On Fri, Aug 06, 2004 at 11:20:28PM -0700, Linus Torvalds wrote:
> 
> 
> On Fri, 6 Aug 2004, Chris Shoemaker wrote:
> > 
> > I _was_ able to find the attached oops, but I don't think I have the
> > corresponding object files, so I hope the decoding it contains is
> > good enough. 
> 
> It's fine.

Well then, maybe you'd like more?  I attached two more from the same
period.  Please remember that these are 5 months old, and could
represent bugs already fixed.  I think this was stock 2.6.4.

> 
> It oopses on
> 
> 	inode->i_sb->s_op
> 
> where "i_sb" is bad and contains the pointer "0x0b7eebf8" which is 
> definitely not a valid kernel pointer.
> 
> There's a few other strange details in your oops report too. One being 
> that the inode pointer (in %ebx, apparently) doesn't show on the stack 
> where I'd expect it to show. Hmm. That might be just a different compiler 
> issue, though.

Perhaps due to CONFIG_REGPARM?  I haven't used it for quite a while, but
back in March I was a bit bolder about config options marked
experimental.  Gene, are you using REGPARM?

-chris

> 
> Anyway, this does look somewhat like the ones Gene is seeing. If I had to 
> guess, I'd guess that either the inode pointer is bad, or it's just stale 
> from an inode that has already been free'd. Most likely because of 
> prune_dcache() having had a corrupt LRU list with a stale/corrupt entry.
> 
> That would blow the prefetch theory out of the water. 
> 
> 		Linus