On Fri, 10 Jul 2009, Pekka Enberg wrote:
> On Thu, 2009-07-09 at 16:26 -0700, David Rientjes wrote:
> > Without DEBUG_SIZE_FLAGS, the only way to determine what flags have
> > increased the size is in calculate_sizes() and then disable them by
> > default if slub_debug=O is specified. calculate_sizes() is used by
> > the `store', `poison', and `red_zone' callbacks, so the admin still has
> > the ability to enable these options even though slub_debug=O was used.
> >
> > So we can either mask off the size-increasing debug bits when the cache is
> > created in kmem_cache_flags() like I did, or we can move the logic to
> > calculate_sizes() with an added formal to determine whether this is from
> > kmem_cache_open() or one of the attribute callbacks.
> >
> > I think my solution is the cleanest and provides a single entity,
> > DEBUG_SIZE_FLAGS, which specifies the flags that slub_debug=O clears if
> > the minimum order increases.
>
> Yup, agreed. I applied the patch, thanks everyone!
There is a simpler solution. Call calculate sizes again if the resulting
sizes increased the order. Something like this.
Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c 2009-07-10 13:45:02.000000000 -0500
+++ linux-2.6/mm/slub.c 2009-07-10 13:46:07.000000000 -0500
@@ -2454,6 +2454,10 @@ static int kmem_cache_open(struct kmem_c
if (!calculate_sizes(s, -1))
goto error;
+ if (get_order(s->size) != get_order(s->objsize) && flag is set) {
+ switch off debug flags.
+ calculate_sizes(s, -1);
+ }
/*
* The larger the object size is, the more pages we want on the
partial
* list to avoid pounding the page allocator excessively.
--
To unsubscribe from this list: send the line "unsubscribe kernel-testers" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
