Re: 2.6.4-rc1-mm2: 3 dumps at __make_request, system freeze

[Andrew Morton <akpm@osdl.org>]

Rui Saraiva <rmps@joel.ist.utl.pt> wrote:
>
> On Thu, 4 Mar 2004, Andrew Morton wrote:
> 
> > Rui Saraiva <rmps@joel.ist.utl.pt> wrote:
> > >
> > > Yesterday I got these 3 dumps (on dmesg) while compiling (on ext3 fs) the
> > > kernel and some other userland utilities.
> >
> > Could you please add this?
> >
> > --- 25/drivers/block/ll_rw_blk.c~blk-unplug-when-max-request-queued-fix	Wed Mar  3 16:03:01 2004
> > +++ 25-akpm/drivers/block/ll_rw_blk.c	Wed Mar  3 16:03:32 2004
> 
> [CUT]
> 
> I'm still experiencing some problems with that patch applied. I was again
> compiling the kernel (no tvtime this time) and got this:
> 
> Unable to handle kernel paging request at virtual address c1810f70 printing eip:
> c02783ae
> *pde = 00006063
> *pte = 01810000
> Oops: 0000 [#1]
> PREEMPT DEBUG_PAGEALLOC
> CPU:    0
> EIP:    0060:[<c02783ae>]    Not tainted VLI
> EFLAGS: 00010086
> EIP is at __make_request+0x3ae/0x6a0

OK, I screwed up.  If we ended up finding a merge in the elevator, local
variable `req' in __make_request() can end up pointing at the now-freed-up
request.

Thanks.  I suggest that you disable CONFIG_DEBUG_PAGEALLOC for now - the
bug is otherwise harmless.