Re: [PATCH 6/6] sunrpc: introduce possibility that requested number of threads is different from actual

[Jeff Layton <jlayton@kernel.org>]

On Wed, 2024-10-23 at 13:37 +1100, NeilBrown wrote:
> New fields sp_nractual and sv_nractual track how many actual threads are
> running.  sp_nrhtreads and sv_nrthreads will be the number that were
> explicitly request.  Currently nractually == nrthreads.
> 
> sv_nractual is used for sizing UDP incoming socket space - in the rare
> case that UDP is used.  This is because each thread might need to keep a
> request in the skbs.
> 
> Signed-off-by: NeilBrown <neilb@suse.de>
> ---
>  include/linux/sunrpc/svc.h |  6 ++++--
>  net/sunrpc/svc.c           | 22 +++++++++++++++-------
>  net/sunrpc/svcsock.c       |  2 +-
>  3 files changed, 20 insertions(+), 10 deletions(-)
> 
> diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h
> index 9d288a673705..3f2c90061b4a 100644
> --- a/include/linux/sunrpc/svc.h
> +++ b/include/linux/sunrpc/svc.h
> @@ -36,7 +36,8 @@
>  struct svc_pool {
>  	unsigned int		sp_id;		/* pool id; also node id on NUMA */
>  	struct lwq		sp_xprts;	/* pending transports */
> -	unsigned int		sp_nrthreads;	/* # of threads in pool */
> +	unsigned int		sp_nrthreads;	/* # of threads requested for pool */
> +	unsigned int		sp_nractual;	/* # of threads running */
>  	struct list_head	sp_all_threads;	/* all server threads */
>  	struct llist_head	sp_idle_threads; /* idle server threads */
>  
> @@ -71,7 +72,8 @@ struct svc_serv {
>  	struct svc_stat *	sv_stats;	/* RPC statistics */
>  	spinlock_t		sv_lock;
>  	unsigned int		sv_nprogs;	/* Number of sv_programs */
> -	unsigned int		sv_nrthreads;	/* # of server threads */
> +	unsigned int		sv_nrthreads;	/* # of server threads requested*/
> +	unsigned int		sv_nractual;	/* # of running threads */
>  	unsigned int		sv_max_payload;	/* datagram payload size */
>  	unsigned int		sv_max_mesg;	/* max_payload + 1 page for overheads */
>  	unsigned int		sv_xdrsize;	/* XDR buffer size */
> diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
> index bd4f02b34f44..d332f9d3d875 100644
> --- a/net/sunrpc/svc.c
> +++ b/net/sunrpc/svc.c
> @@ -781,8 +781,12 @@ svc_pool_victim(struct svc_serv *serv, struct svc_pool *target_pool,
>  	}
>  
>  	if (pool && pool->sp_nrthreads) {
> -		set_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
> -		set_bit(SP_NEED_VICTIM, &pool->sp_flags);
> +		if (pool->sp_nrthreads <= pool->sp_nractual) {
> +			set_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
> +			set_bit(SP_NEED_VICTIM, &pool->sp_flags);
> +			pool->sp_nractual -= 1;
> +			serv->sv_nractual -= 1;
> +		}

This is a little strange. It decrements the thread counts before the
threads actually exit. So, these counts are really just aspirational at
this point. Not a bug, just a note to myself while I am going over
this.

>  		return pool;
>  	}
>  	return NULL;
> @@ -803,6 +807,12 @@ svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
>  		chosen_pool = svc_pool_next(serv, pool, &state);
>  		node = svc_pool_map_get_node(chosen_pool->sp_id);
>  
> +		serv->sv_nrthreads += 1;
> +		chosen_pool->sp_nrthreads += 1;
> +
> +		if (chosen_pool->sp_nrthreads <= chosen_pool->sp_nractual)
> +			continue;
> +
>  		rqstp = svc_prepare_thread(serv, chosen_pool, node);
>  		if (!rqstp)
>  			return -ENOMEM;
> @@ -812,8 +822,8 @@ svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
>  			svc_exit_thread(rqstp);
>  			return PTR_ERR(task);
>  		}
> -		serv->sv_nrthreads += 1;
> -		chosen_pool->sp_nrthreads += 1;
> +		serv->sv_nractual += 1;
> +		chosen_pool->sp_nractual += 1;
>  
>  		rqstp->rq_task = task;
>  		if (serv->sv_nrpools > 1)
> @@ -850,6 +860,7 @@ svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs)
>  			    TASK_IDLE);
>  		nrservs++;
>  	} while (nrservs < 0);
> +	svc_sock_update_bufs(serv);

Nice little change here. No need to do this on every thread exit --
just do it once when they're all done. This change probably should be
split into a separate patch.

>  	return 0;
>  }
>  
> @@ -955,13 +966,10 @@ void svc_rqst_release_pages(struct svc_rqst *rqstp)
>  void
>  svc_exit_thread(struct svc_rqst *rqstp)
>  {
> -	struct svc_serv	*serv = rqstp->rq_server;
>  	struct svc_pool	*pool = rqstp->rq_pool;
>  
>  	list_del_rcu(&rqstp->rq_all);
>  
> -	svc_sock_update_bufs(serv);
> -
>  	svc_rqst_free(rqstp);
>  
>  	clear_and_wake_up_bit(SP_VICTIM_REMAINS, &pool->sp_flags);
> diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
> index 825ec5357691..191dbc648bd0 100644
> --- a/net/sunrpc/svcsock.c
> +++ b/net/sunrpc/svcsock.c
> @@ -588,7 +588,7 @@ static int svc_udp_recvfrom(struct svc_rqst *rqstp)
>  	     * provides an upper bound on the number of threads
>  	     * which will access the socket.
>  	     */
> -	    svc_sock_setbufsize(svsk, serv->sv_nrthreads + 3);
> +	    svc_sock_setbufsize(svsk, serv->sv_nractual + 3);
>  
>  	clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags);
>  	err = kernel_recvmsg(svsk->sk_sock, &msg, NULL,

Other than maybe splitting that one change into a separate patch, this
looks good to me.

Reviewed-by: Jeff Layton <jlayton@kernel.org>