Re: BUG: scheduling while atomic: cron/668/0x10c9a0c0 (was: Re: mm, page_alloc: avoid looking up the first zone in a zonelist twice)

[Mel Gorman]

On Tue, May 31, 2016 at 11:28:05AM +0200, Geert Uytterhoeven wrote:
> Hi Mel,
> 
> On Mon, May 30, 2016 at 8:56 PM, Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
> > Thanks. Please try the following instead
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index bb320cde4d6d..557549c81083 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -3024,6 +3024,7 @@ get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
> >                 apply_fair = false;
> >                 fair_skipped = false;
> >                 reset_alloc_batches(ac->preferred_zoneref->zone);
> > +               z = ac->preferred_zoneref;
> >                 goto zonelist_scan;
> >         }
> 
> Thanks a lot, that seems to fix the issue!.
> 
> Tested-by: Geert Uytterhoeven <geert@xxxxxxxxxxxxxx>
> 
> JFTR, without the fix, sometimes I get a different, but equally obscure, crash
> than the one I posted before:
> 

I'm afraid I don't recognise it. Given the nature of the previous bug
though, I have a vague suspicion that someone is not handling a page
allocation failure properly and goes boom later.

-- 
Mel Gorman
SUSE Labs