mm/slab_common: use SLAB_NO_MERGE instead of negative refcount

When CONFIG_MEMCG_KMEM is enabled, we disable cache merging for
KMALLOC_NORMAL caches so they don't end up merged with a cache that uses
ad-hoc __GFP_ACCOUNT when allocating. This was implemented by setting
the refcount to -1, but now we have a proper SLAB_NO_MERGE flag, so use
that instead.

Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: David Rientjes <rientjes@google.com>

Author: tehcaster

mm/slab_common.c

@@ -878,17 +878,17 @@ new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
 		flags |= SLAB_CACHE_DMA;
 	}
 
-	kmalloc_caches[type][idx] = create_kmalloc_cache(
-					kmalloc_info[idx].name[type],
-					kmalloc_info[idx].size, flags, 0,
-					kmalloc_info[idx].size);
-
 	/*
 	 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
 	 * KMALLOC_NORMAL caches.
 	 */
 	if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
-		kmalloc_caches[type][idx]->refcount = -1;
+		flags |= SLAB_NO_MERGE;
+
+	kmalloc_caches[type][idx] = create_kmalloc_cache(
+					kmalloc_info[idx].name[type],
+					kmalloc_info[idx].size, flags, 0,
+					kmalloc_info[idx].size);
 }
 
 /*