On Wed 07-01-15 10:36:58, Vlastimil Babka wrote:
> On 01/06/2015 10:44 PM, Michal Hocko wrote:
> > On Tue 06-01-15 22:10:55, Vlastimil Babka wrote:
> >> On 01/06/2015 03:30 PM, Michal Hocko wrote:
> > [...]
> >> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> >> > index 1bb65e6f48dd..1682d766cb8e 100644
> >> > --- a/mm/page_alloc.c
> >> > +++ b/mm/page_alloc.c
> >> > @@ -2175,10 +2175,11 @@ zonelist_scan:
> >> > }
> >> >
> >> > try_this_zone:
> >> > - page = buffered_rmqueue(preferred_zone, zone, order,
> >> > + do {
> >> > + page = buffered_rmqueue(preferred_zone, zone, order,
> >> > gfp_mask, migratetype);
> >> > - if (page)
> >> > - break;
> >> > + } while (page && prep_new_page(page, order, gfp_mask,
> >> > + alloc_flags));
> >>
> >> Hm but here we wouldn't return page on success.
> >
> > Right.
> >
> >> I wonder if you overlooked the return, hence your "not breaking out of
> >> the loop" remark?
> >
> > This was merely to show the intention. Sorry for not being clear enough.
>
> OK, but I don't see other way than to follow this do-while with another
>
> if (page)
> return page;
>
> So I think it would be more complicated than now. We wouldn't even be able to
> remove the 'try_this_zone' label, since it's used for goto from elsewhere as well.
Getting rid of the label wasn't the intention. I just found the
allocation retry easier to follow this way. I have no objection if you
keep the code as is.
> Now that I'm thinking of it, maybe we should have a "goto zonelist_scan" there
> instead. We discard a bad page and might come below the watermarks. But the
> chances of this mattering are tiny I guess.
If anything this would be worth a separate patch.
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>