On Mon, Nov 09, 2015 at 09:25:22PM +0100, Jesper Dangaard Brouer wrote: > On Mon, 9 Nov 2015 22:13:35 +0300 > Vladimir Davydov <vdavydov@xxxxxxxxxxxxx> wrote: > > > On Mon, Nov 09, 2015 at 07:17:31PM +0100, Jesper Dangaard Brouer wrote: > > ... > > > @@ -2556,7 +2563,7 @@ redo: > > > if (unlikely(gfpflags & __GFP_ZERO) && object) > > > memset(object, 0, s->object_size); > > > > > > - slab_post_alloc_hook(s, gfpflags, object); > > > + slab_post_alloc_hook(s, gfpflags, 1, object); > > > > I think it must be &object > > The object is already a void ** type. Let's forget about types for a second. object contains an address to the newly allocated object, while slab_post_alloc_hook expects an array of addresses to objects. Simple test. Suppose an allocation failed. Then object equals 0. Passing 0 to slab_post_alloc_hook as @p and 1 as @size will result in NULL ptr dereference. > > > BTW why is object defined as void **? I suspect we can safely drop one > > star. > > Maybe Christoph can explain this? > > > > > > > > return object; > > > } > > ... > > > @@ -2953,11 +2958,15 @@ bool kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, > > > memset(p[j], 0, s->object_size); > > > } > > > > > > + /* memcg and kmem_cache debug support */ > > > + slab_post_alloc_hook(s, flags, size, p); > > > + > > > return true; > > > > > > error: > > > __kmem_cache_free_bulk(s, i, p); > > > local_irq_enable(); > > > + memcg_kmem_put_cache(s); > > > > I wouldn't tear memcg_kmem_put_cache from slab_post_alloc_hook. If we > > add something else to slab_post_alloc_hook (e.g. we might want to call > > tracing functions from there), we'll have to modify this error path > > either, which is easy to miss. > > > > What about calling > > > > slab_post_alloc_hook(s, flags, 0, NULL); > > > > here? > > Maybe the correct behavior here, to adhere to all the debugging options, > is to call: > > error: > local_irq_enable(); > slab_post_alloc_hook(s, flags, i, p); > __kmem_cache_free_bulk(s, i, p); > return false; Yeah, I think you're right, because __kmem_cache_free_bulk calls slab_free_hook, which is supposed to undo slab_post_alloc_hook, so we must call the latter for allocated objects. Thanks, Vladimir > > > > > return false; > > > } > > > EXPORT_SYMBOL(kmem_cache_alloc_bulk); > > > -- 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>