Re: [slub] WARNING: CPU: 0 PID: 0 at mm/slub.c:1511 __kmem_cache_create()

[Dave Hansen <dave.hansen@intel.com>]

[-- Attachment #1: Type: text/plain, Size: 447 bytes --]

> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/commit/?id=c65c1877bd6826ce0d9713d76e30a7bed8e49f38

I think the assert is just bogus at least in the early case.
early_kmem_cache_node_alloc() says:
 * No kmalloc_node yet so do it by hand. We know that this is the first
 * slab on the node for this slabcache. There are no concurrent accesses
 * possible.

Should we do something like the attached patch?  (very lightly tested)

[-- Attachment #2: slub-lockdep-workaround.patch --]
[-- Type: text/x-patch, Size: 653 bytes --]



---

 b/mm/slub.c |    6 ++++++
 1 file changed, 6 insertions(+)

diff -puN mm/slub.c~slub-lockdep-workaround mm/slub.c
--- a/mm/slub.c~slub-lockdep-workaround	2014-01-14 09:19:22.418942641 -0800
+++ b/mm/slub.c	2014-01-14 09:29:55.441297460 -0800
@@ -2890,7 +2890,13 @@ static void early_kmem_cache_node_alloc(
 	init_kmem_cache_node(n);
 	inc_slabs_node(kmem_cache_node, node, page->objects);
 
+	/*
+	 * the lock is for lockdep's sake, not for any actual
+	 * race protection
+	 */
+	spin_lock(&n->list_lock);
 	add_partial(n, page, DEACTIVATE_TO_HEAD);
+	spin_unlock(&n->list_lock);
 }
 
 static void free_kmem_cache_nodes(struct kmem_cache *s)
_