From: Christoph Lameter There is the need to use the objects per slab in the first part of __slab_alloc() which is still pretty hot. Copy the number of objects per slab into the kmem_cache_cpu structure. That way we can get the value from a cache line that we already need to touch. This brings the kmem_cache_cpu structure up to 4 even words. There is no increase in the size of kmem_cache_cpu since the size of object is rounded to the next word. Signed-off-by: Christoph Lameter Reviewed-by: Pekka Enberg Signed-off-by: Andrew Morton --- include/linux/slub_def.h | 1 + mm/slub.c | 3 ++- 2 files changed, 3 insertions(+), 1 deletion(-) diff -puN include/linux/slub_def.h~slub-avoid-referencing-kmem_cache-structure-in-__slab_alloc include/linux/slub_def.h --- a/include/linux/slub_def.h~slub-avoid-referencing-kmem_cache-structure-in-__slab_alloc +++ a/include/linux/slub_def.h @@ -17,6 +17,7 @@ struct kmem_cache_cpu { int node; unsigned int offset; unsigned int objsize; + unsigned int objects; }; struct kmem_cache_node { diff -puN mm/slub.c~slub-avoid-referencing-kmem_cache-structure-in-__slab_alloc mm/slub.c --- a/mm/slub.c~slub-avoid-referencing-kmem_cache-structure-in-__slab_alloc +++ a/mm/slub.c @@ -1512,7 +1512,7 @@ load_freelist: object = c->page->freelist; c->freelist = object[c->offset]; - c->page->inuse = s->objects; + c->page->inuse = c->objects; c->page->freelist = c->page->end; c->node = page_to_nid(c->page); unlock_out: @@ -1878,6 +1878,7 @@ static void init_kmem_cache_cpu(struct k c->node = 0; c->offset = s->offset / sizeof(void *); c->objsize = s->objsize; + c->objects = s->objects; } static void init_kmem_cache_node(struct kmem_cache_node *n) _