On Tue, Nov 05, 2024 at 11:48:11AM -1000, Tejun Heo wrote:

[...]

>  static bool consume_dispatch_q(struct rq *rq, struct scx_dispatch_q *dsq)
>  {
>  	struct task_struct *p;
>  retry:
>  	/*
> +	 * This retry loop can repeatedly race against scx_ops_bypass()
> +	 * dequeueing tasks from @dsq trying to put the system into the bypass
> +	 * mode. On some multi-socket machines (e.g. 2x Intel 8480c), this can
> +	 * live-lock the machine into soft lockups. Give a breather.
> +	 */
> +	scx_ops_breather(rq);

Should we move this to after the list_empty() check? Or before the goto retry
below so we can avoid having to do the atomic read on the typical hotpath?

> +
> +	/*
>  	 * The caller can't expect to successfully consume a task if the task's
>  	 * addition to @dsq isn't guaranteed to be visible somehow. Test
>  	 * @dsq->list without locking and skip if it seems empty.
> @@ -4550,10 +4587,11 @@ bool task_should_scx(struct task_struct
>   */
>  static void scx_ops_bypass(bool bypass)
>  {
> +	static DEFINE_RAW_SPINLOCK(bypass_lock);
>  	int cpu;
>  	unsigned long flags;
>  
> -	raw_spin_lock_irqsave(&__scx_ops_bypass_lock, flags);
> +	raw_spin_lock_irqsave(&bypass_lock, flags);
>  	if (bypass) {
>  		scx_ops_bypass_depth++;
>  		WARN_ON_ONCE(scx_ops_bypass_depth <= 0);
> @@ -4566,6 +4604,8 @@ static void scx_ops_bypass(bool bypass)
>  			goto unlock;
>  	}
>  
> +	atomic_inc(&scx_ops_breather_depth);
> +
>  	/*
>  	 * No task property is changing. We just need to make sure all currently
>  	 * queued tasks are re-queued according to the new scx_rq_bypassing()
> @@ -4621,8 +4661,10 @@ static void scx_ops_bypass(bool bypass)
>  		/* resched to restore ticks and idle state */
>  		resched_cpu(cpu);
>  	}
> +
> +	atomic_dec(&scx_ops_breather_depth);
>  unlock:
> -	raw_spin_unlock_irqrestore(&__scx_ops_bypass_lock, flags);
> +	raw_spin_unlock_irqrestore(&bypass_lock, flags);
>  }
>  
>  static void free_exit_info(struct scx_exit_info *ei)
> @@ -6275,6 +6317,13 @@ static bool scx_dispatch_from_dsq(struct
>  		raw_spin_rq_lock(src_rq);
>  	}
>  
> +	/*
> +	 * If the BPF scheduler keeps calling this function repeatedly, it can
> +	 * cause similar live-lock conditions as consume_dispatch_q(). Insert a
> +	 * breather if necessary.
> +	 */
> +	scx_ops_breather(src_rq);
> +
>  	locked_rq = src_rq;
>  	raw_spin_lock(&src_dsq->lock);
>  
>