Re: BUG in tmpfs

[Hugh Dickins <hugh@veritas.com>]

On Thu, 19 Jan 2006, Grzegorz Kulewski wrote:
> > On Thu, Jan 19, 2006 at 01:14:56AM +0100, Grzegorz Kulewski wrote:
> > > 
> > > kernel BUG at mm/shmem.c:836!
> > > Jan 17 23:31:42 kangur [4312977.009000] CPU:    0
> > > Jan 17 23:31:42 kangur [4312977.009000] EIP:    0060:[<b014a429>]
> > > Tainted: P      VLI
> > > Jan 17 23:31:42 kangur [4312977.009000] EFLAGS: 00210206 
> > > (2.6.15-rc6-git4-ck1)
> > > Jan 17 23:31:42 kangur [4312977.009000] EIP is at 
> > > shmem_writepage+0xa6/0x15a
>... 
> I don't have infrastructure for making 24/7 stress tests of tmpfs (or anything
> other). But maybe if somebody has some spare machines he could try doing it
> because I am seeing different oopses/crashes/bugs in tmpfs on different kernel
> (on x86 and x86-64) from time to time.

If you still have the logs from those, please do send me.  But nobody
else has reported any such tmpfs crashes for two or three years - though
I bet just the saying of that will help someone to remind me I'm wrong ;)

While it's easy to imagine some nasty interaction between swap prefetch
and shmem_writepage's peculiar use of swap, a quick glance at that patch
didn't ring any alarm bells for me, and it's not obvious how a problem
there would result in your particular "BUG_ON(entry->val);" - which could
as well come from random corruption (and GregKH has pointed out that your
kernel is P Tainted).

I've not been testing tmpfs on -ck, but the swap prefetch patch was in
-mm for a while, and I didn't see any such problem when testing there.
Please let me know if you see more such, or can unearth earlier logs.

Hugh