Re: [PATCH] net: skbuff: allocate the fclone in the current NUMA node

[Alexander Lobakin <aleksander.lobakin@intel.com>]

From: Huang Shijie <shijie@os.amperecomputing.com>
Date: Tue, 20 Feb 2024 10:18:04 +0800

> The current code passes NUMA_NO_NODE to __alloc_skb(), we found
> it may creates fclone SKB in remote NUMA node.
> 
> So use numa_node_id() to limit the allocation to current NUMA node.
> 
> Signed-off-by: Huang Shijie <shijie@os.amperecomputing.com>
> ---
>  include/linux/skbuff.h | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> index 2dde34c29203..ebc42b2604ad 100644
> --- a/include/linux/skbuff.h
> +++ b/include/linux/skbuff.h
> @@ -1343,7 +1343,7 @@ static inline bool skb_fclone_busy(const struct sock *sk,
>  static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
>  					       gfp_t priority)
>  {
> -	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
> +	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, numa_node_id());

Because it tries to defragment the memory and pick an optimal node.

__alloc_skb() and skb clones aren't anyway something very hotpathish, do
you have any particular perf numbers and/or usecases where %NUMA_NO_NODE
really hurts?

>  }
>  
>  struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);

Thanks,
Olek