On Thu, 2013-01-17 at 21:28 +0000, Christoph Lameter wrote:
> On Thu, 17 Jan 2013, Steven Rostedt wrote:
>
> > > --- a/mm/slub.c
> > > +++ b/mm/slub.c
> > > @@ -2399,7 +2399,7 @@ redo:
> > >
> > > object = c->freelist;
> > > page = c->page;
> > > - if (unlikely(!object || !node_match(page, node)))
> > > + if (unlikely(!object || !page || !node_match(page, node)))
> >
> > I'm still trying to see if c->freelist != NULL and c->page == NULL isn't
> > a bug. The cmpxchg_doubles are a little confusing. If it's not expected
> > that page is NULL but freelist isn't than we need to figure out why it
> > happened.
>
> hmmm.. We may want to change the sequence of updates to c->page and
> c->freelist. Update c->freelist to be NULL first so that we always enter
> the slow path for these cases where we can do more expensive
> synchronization.
>
> Index: linux/mm/slub.c
> ===================================================================
> --- linux.orig/mm/slub.c 2013-01-15 10:42:08.490183607 -0600
> +++ linux/mm/slub.c 2013-01-17 15:27:48.973051155 -0600
> @@ -1993,8 +1993,8 @@ static inline void flush_slab(struct kme
> deactivate_slab(s, c->page, c->freelist);
>
> c->tid = next_tid(c->tid);
> - c->page = NULL;
> c->freelist = NULL;
> + c->page = NULL;
> }
>
> /*
> @@ -2227,8 +2227,8 @@ redo:
> if (unlikely(!node_match(page, node))) {
> stat(s, ALLOC_NODE_MISMATCH);
> deactivate_slab(s, page, c->freelist);
> - c->page = NULL;
> c->freelist = NULL;
> + c->page = NULL;
> goto new_slab;
> }
>
> @@ -2239,8 +2239,8 @@ redo:
> */
> if (unlikely(!pfmemalloc_match(page, gfpflags))) {
> deactivate_slab(s, page, c->freelist);
> - c->page = NULL;
> c->freelist = NULL;
> + c->page = NULL;
> goto new_slab;
> }
>
Without appropriate barriers, this change is a shoot in the dark.
--
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>