Re: [PATCH v2] io_uring/lock: Keep spinlock release before wake_up()

[Jens Axboe]

On 9/23/26 10:16 AM, Xiaochuan Li wrote:
> @@ -574,9 +575,8 @@ static struct io_wq_work *io_get_next_work(struct io_wq_acct *acct,
>  		unstalled = io_wait_on_hash(wq, stall_hash);
>  		raw_spin_lock(&acct->lock);
>  		if (unstalled) {
> +			*need_wake = true;
>  			clear_bit(IO_ACCT_STALLED_BIT, &acct->flags);
> -			if (wq_has_sleeper(&wq->hash->wait))
> -				wake_up(&wq->hash->wait);
>  		}
>  	}

*need_wake = wq_has_sleeper(&wq->hash->wait);

? And probably retain the sme ordering with the bit clear.

Apart from that, I think it looks fine. The lock shuffling is somewhat
annoying and inefficient, an unfortunate side effect of the raw vs
normal spinlocks.

-- 
Jens Axboe