Re: 2.6.25.6 kernel BUG at mm/highmem.c:319!

[Hugh Dickins <hugh@veritas.com>]

On Thu, 26 Jun 2008, Arjan van de Ven wrote:
> On Thu, 26 Jun 2008 10:59:42 -0400
> Dave Jones <davej@redhat.com> wrote:
> 
> > Anyone know how this can happen ?
> > We just got a report from a user who is hitting the 
> > 
> >         BUG_ON(!PageHighMem(page));
> > 
> > in set_page_address().
> 
> we discussed this 2 or 3 weeks ago as part of the kerneloops reports...
> everyone is puzzled... it's the number 20 report right now.
> 
> http://www.kerneloops.org/search.php?search=set_page_address
> 
> it seems that fuse is a very popular module when this hits...

I'm as puzzled now as back then.  And fuse isn't in Dave's list.
But I did notice after reporting back on it, that i915 appears in all
of the reports I looked at; and Fedora does have some drm/i915
patches, which do involve kmap.

I know precious little about drm, but didn't spot anything in those
patches which was likely to cause this: they might be a good place
for Dave to start looking though.

If this user is suffering from it repeatedly, might s/he be prepared
to run a kernel with more debug in it?  Though at this moment,
I don't know what it is that I'd like to be looking for anyway -
it's always easier to come up with debug patches when you've a
hypothesis as to what's going on, which I don't have.

Hugh