Re: [PATCH] mm/slub: remove duplicate initialization for early_kmem_cache_node_alloc()

[David Rientjes <rientjes@google.com>]

On Sat, 6 Apr 2024, Sangyun Kim wrote:

> The struct track for every object in a new slab is already set up by
> new_slab(),
> so remove the duplicate initialization in early_kmem_cache_node_alloc().
> 
> Signed-off-by: Sangyun Kim <sangyun.kim@snu.ac.kr>
> 
> Co-developed-by: Hyunmin Lee <hyunminlr@gmail.com>
> Signed-off-by: Hyunmin Lee <hyunminlr@gmail.com>
> 
> Co-developed-by: Jeungwoo Yoo <casionwoo@gmail.com>
> Signed-off-by: Jeungwoo Yoo <casionwoo@gmail.com>
> 
> Cc: Gwan-gyeong Mun <gwan-gyeong.mun@intel.com>
> ---
>  mm/slub.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/mm/slub.c b/mm/slub.c
> index 0dfc0c18a78b..5ffe46843b36 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -4938,7 +4938,6 @@ static void early_kmem_cache_node_alloc(int node)
>  	BUG_ON(!n);
>  #ifdef CONFIG_SLUB_DEBUG
>  	init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
> -	init_tracking(kmem_cache_node, n);
>  #endif
>  	n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
>  	slab->freelist = get_freepointer(kmem_cache_node, n);

I think this is technically safe based on the current implementation 
because, as you said, allocate_slab() takes care of this for 
SLAB_STORE_USER.

What user observable effect does this have given it would only make a 
difference when slab_state == DOWN?