Re: [PATCH] mm/percpu: clean up callers of pcpu_obj_full_size()

[Harry Yoo]

On Mon, Oct 19, 2026 at 09:28:02AM -0700, Joshua Hahn wrote:
> The size of a pcpu allocation can be interpreted in two ways:
> 1. The size of the allocation attributable to a single CPU.
> 2. The total size of the allocation across all CPUs, including the
>    allocation overhead (for obj_cgroup membership).
> 
> In most accounting paths for percpu memory, we need the second
> definition (total size). Make this more obvious by coalescing the
> callers of pcpu_obj_full_size().
> 
> No functional changes intended.
> 
> Signed-off-by: Joshua Hahn <joshua.hahnjy@xxxxxxxxx>
> ---
>  mm/percpu.c | 39 +++++++++++++++++++--------------------
>  1 file changed, 19 insertions(+), 20 deletions(-)
> 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index d7883507dc151..76848e096caee 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -1623,7 +1623,7 @@ static bool pcpu_memcg_pre_alloc_hook(size_t size, gfp_t gfp,
>  	if (!objcg || obj_cgroup_is_root(objcg))
>  		return true;
>  
> -	if (obj_cgroup_charge(objcg, gfp, pcpu_obj_full_size(size)))
> +	if (obj_cgroup_charge(objcg, gfp, size))
>  		return false;
>  
>  	*objcgp = objcg;
> @@ -1642,11 +1642,10 @@ static void pcpu_memcg_post_alloc_hook(struct obj_cgroup *objcg,
>  		chunk->obj_exts[off >> PCPU_MIN_ALLOC_SHIFT].cgroup = objcg;
>  
>  		rcu_read_lock();
> -		mod_memcg_state(obj_cgroup_memcg(objcg), MEMCG_PERCPU_B,
> -				pcpu_obj_full_size(size));
> +		mod_memcg_state(obj_cgroup_memcg(objcg), MEMCG_PERCPU_B, size);
>  		rcu_read_unlock();
>  	} else {
> -		obj_cgroup_uncharge(objcg, pcpu_obj_full_size(size));
> +		obj_cgroup_uncharge(objcg, size);
>  	}
>  }
>  
> @@ -1662,11 +1661,10 @@ static void pcpu_memcg_free_hook(struct pcpu_chunk *chunk, int off, size_t size)

Perhaps it's better to rename 'size' parameter to 'full_size' in those
memcg hooks to make it clear it's expecting total size?

Otherwise LGTM!
Thanks for cleaning this up.

And it might be beyond the scope of this patch, but I think
trace_percpu_alloc_percpu_fail() should take full_size instead of size.

Wdyt?

-- 
Cheers,
Harry / Hyeonggon