Re: [PATCH 2/2] sunrpc: new tracepoints around svc thread wakeups

[Simon Horman <horms@kernel.org>]

On Tue, May 27, 2025 at 08:12:48PM -0400, Jeff Layton wrote:
> Convert the svc_wake_up tracepoint into svc_pool_thread_event class.
> Have it also record the pool id, and add new tracepoints for when the
> thread is already running and for when there are no idle threads.
> 
> Signed-off-by: Jeff Layton <jlayton@kernel.org>

...

> diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
> index e7f9c295d13c03bf28a5eeec839fd85e24f5525f..de80d3683350dc86bee3413719797dcf7a4562e8 100644
> --- a/net/sunrpc/svc.c
> +++ b/net/sunrpc/svc.c
> @@ -749,13 +749,16 @@ void svc_pool_wake_idle_thread(struct svc_pool *pool)
>  		WRITE_ONCE(rqstp->rq_qtime, ktime_get());
>  		if (!task_is_running(rqstp->rq_task)) {
>  			wake_up_process(rqstp->rq_task);
> -			trace_svc_wake_up(rqstp->rq_task->pid);
> +			trace_svc_pool_thread_wake(pool, rqstp->rq_task->pid);
>  			percpu_counter_inc(&pool->sp_threads_woken);
> +		} else {
> +			trace_svc_pool_thread_running(pool, rqstp->rq_task->pid);
>  		}
>  		rcu_read_unlock();
>  		return;
>  	}
>  	rcu_read_unlock();

Hi Jeff,

Above, rqstp is conditionally initialised if ln
(pool->sp_idle_threads.first) is not NULL.
But below it is dereferenced unconditionally.
This does not seem consistent.

Reported by clang-20.1.4 builds and Smatch.

> +	trace_svc_pool_thread_noidle(pool, rqstp->rq_task->pid);
>  
>  }
>  EXPORT_SYMBOL_GPL(svc_pool_wake_idle_thread);
> 
> -- 
> 2.49.0
>