Re: [RFC PATCH 1/4] workqueue: support pausing ordered workqueues

[Tejun Heo]

Hello,

This looks great to me in general.

On Wed, May 10, 2023 at 06:04:25PM +0200, Johannes Berg wrote:
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index b8b541caed48..418d99ff8325 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -3863,10 +3863,16 @@ static void pwq_adjust_max_active(struct pool_workqueue *pwq)
>  	struct workqueue_struct *wq = pwq->wq;
>  	bool freezable = wq->flags & WQ_FREEZABLE;
>  	unsigned long flags;
> +	int new_max_active;
>  
> -	/* for @wq->saved_max_active */
> +	/* for @wq->saved_max_active and @wq->flags */
>  	lockdep_assert_held(&wq->mutex);
>  
> +	if (wq->flags & __WQ_PAUSED)
> +		new_max_active = 0;
> +	else
> +		new_max_active = wq->saved_max_active;

Nothing is using new_max_active and I think we can probably combine this
with the freezing test.

> +void __workqueue_pause_resume(struct workqueue_struct *wq, bool pause)
> +{
> +	struct pool_workqueue *pwq;
> +
> +	mutex_lock(&wq->mutex);
> +	if (pause)
> +		wq->flags |= __WQ_PAUSED;
> +	else
> +		wq->flags &= ~__WQ_PAUSED;
> +
> +	for_each_pwq(pwq, wq)
> +		pwq_adjust_max_active(pwq);
> +	mutex_unlock(&wq->mutex);
> +
> +	if (pause)
> +		flush_workqueue(wq);
> +}
> +EXPORT_SYMBOL_GPL(__workqueue_pause_resume);

I'd just make pause and resume separate functions. The sharing ratio doesn't
seem that high.

Thanks.

-- 
tejun