Re: [RFC 4/4] mm, page_alloc: fix premature OOM when racing with cpuset mems update

[Vlastimil Babka]

On 01/18/2017 08:12 AM, Hillf Danton wrote:
> On Wednesday, January 18, 2017 6:16 AM Vlastimil Babka wrote:
> > @@ -3802,13 +3811,8 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
> >  	 * Also recalculate the starting point for the zonelist iterator or
> >  	 * we could end up iterating over non-eligible zones endlessly.
> >  	 */
> Is the newly added comment still needed?

You're right that it's no longer true. I think we can even remove most of the 
zoneref trickery and non-NULL checks in the fastpath (as a cleanup patch on 
top), as the loop in get_page_from_freelist() should handle it just fine. IIRC 
Mel even did this in the microopt series, but I pointed out that NULL 
preferred_zoneref pointer would be dangerous in get_page_from_freelist(). We 
didn't realize that we check the wrong pointer (i.e. patch 1/4 here).

Vlastimil

> > -	if (unlikely(ac.nodemask != nodemask)) {
> > -no_zone:
> > +	if (unlikely(ac.nodemask != nodemask))
> >  		ac.nodemask = nodemask;
> > -		ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
> > -						ac.high_zoneidx, ac.nodemask);
> > -		/* If we have NULL preferred zone, slowpath wll handle that */
> > -	}
> >
> >  	page = __alloc_pages_slowpath(alloc_mask, order, &ac);
> >
> > --
> > 2.11.0