Re: [PATCH v3] sched_ext: Fix lock imbalance in dispatch_to_local_dsq()

[Andrea Righi <arighi@nvidia.com>]

On Sat, Jan 25, 2025 at 07:56:08AM +0100, Andrea Righi wrote:
...
> @@ -2557,6 +2567,7 @@ static void dispatch_to_local_dsq(struct rq *rq, struct scx_dispatch_q *dst_dsq,
>  {
>  	struct rq *src_rq = task_rq(p);
>  	struct rq *dst_rq = container_of(dst_dsq, struct rq, scx.local_dsq);
> +	struct rq *locked_rq = rq;

I just noticed that we have an unused variable here with !CONFIG_SMP, so
ignore this. I'll send a new version soon.

Sorry for the noise.
-Andrea

>  
>  	/*
>  	 * We're synchronized against dequeue through DISPATCHING. As @p can't
> @@ -2593,12 +2604,16 @@ static void dispatch_to_local_dsq(struct rq *rq, struct scx_dispatch_q *dst_dsq,
>  	atomic_long_set_release(&p->scx.ops_state, SCX_OPSS_NONE);
>  
>  	/* switch to @src_rq lock */
> -	if (rq != src_rq) {
> -		raw_spin_rq_unlock(rq);
> +	if (locked_rq != src_rq) {
> +		raw_spin_rq_unlock(locked_rq);
> +		locked_rq = src_rq;
>  		raw_spin_rq_lock(src_rq);
>  	}
>  
> -	/* task_rq couldn't have changed if we're still the holding cpu */
> +	/*
> +	 * If p->scx.holding_cpu still matches the current CPU, task_rq(p)
> +	 * has not changed and we can safely move the task to @dst_rq.
> +	 */
>  	if (likely(p->scx.holding_cpu == raw_smp_processor_id()) &&
>  	    !WARN_ON_ONCE(src_rq != task_rq(p))) {
>  		/*