Re: kernel BUG at lib/radix-tree.c:473!

[Nick Piggin <nickpiggin@yahoo.com.au>]

On Monday 18 August 2008 02:19, Ian Campbell wrote:
> On Fri, 2008-08-15 at 01:00 +0100, Hugh Dickins wrote:
> > On Fri, 15 Aug 2008, Johannes Weiner wrote:
> > > Jeremy Fitzhardinge <jeremy@goop.org> writes:
> > > > Hugh Dickins wrote:
> > > >> An additional useful input would be: what happens if you replace
> > > >> that /dev/fb0 by a symlink /dev/fb0 pointing to an fb0 device node
> > > >> in one of your disk filesystems?  I rather expect that to cause the
> > > >> same trouble, which would argue that the driver is wrong and shmem
> > > >> right.
> > > >
> > > > I don't follow. Do you mean make /dev/fb0 a plain file on a
> > > > filesystem?  Or make it a disk device node?  Something else?
> > >
> > > Creating a device node on a different filesystem to see if the driver
> > > only worked with the safe shmem set_page_dirty and now breaks due to
> > > exposure to the generic version.  Or if the driver works with the
> > > generic version through other mappings and the shmem code screws it up
> > > somewhere else.
> >
> > Yes, that's it.  I think it was ext2 I referred to, when I worried
> > about this when making the change to tmpfs; and my reading of it
> > was that ext2 left a device node's a_ops unset, as I was changing
> > tmpfs to do.  (Looking at it again, ext2 doesn't even specify its
> > .set_page_dirty, so even if it had assigned an a_ops, it wouldn't
> > have avoided the default behaviour.)  But I'd like to hear what
> > actually happens in practice, rather than relying on my reading.
>
> Creating /tmp/fb0 on an ext3 filesystem gives the same behaviour.
>
> FWIW the patch below apparently makes it work for me, but I'm not going
> to pretend I follow what's going on, why or what else it breaks ;-)

I think Iwould prefer fs_defio to use its own set_page_dirty
function. __set_page_dirty_no_writeback is supposed to be used
on pagecache, by filesystems.