Re: [PATCH 3/7 V2] workqueue: introduce get_pwq_unlocked()

[Tejun Heo <tj@kernel.org>]

Hello,

So, it looks like we'll need get_pwq_unlocked().  Reviews below.

On Tue, May 12, 2015 at 08:32:31PM +0800, Lai Jiangshan wrote:
> attrs management code may reuse existed pwq and it has open code
> to do "lock();get_pwq();unlock()", we move this open code into
> get_pwq_unlocked().
> 
> get_pwq_unlocked() will also be used in later patches to allow
> apply_wqattrs_prepare() to resue the original default or per-node pwq.

                             reuse

>  /**
> + * get_pwq_unlocked - get_pwq() with surrounding pool lock/unlock
> + * @pwq: pool_workqueue to get (should not %NULL)
> + *
> + * get_pwq() with locking.  The caller should have at least an owned
> + * reference on @pwq to match the guarantees required by get_pwq().
> + *
> + * Return itsefl for allowing chained expressions.
> + */
> +static struct pool_workqueue *get_pwq_unlocked(struct pool_workqueue *pwq)
> +{
> +	spin_lock_irq(&pwq->pool->lock);
> +	get_pwq(pwq);
> +	spin_unlock_irq(&pwq->pool->lock);
> +
> +	return pwq;
> +}

As I mentioned before, please drop the return value.  It may be
tempting to do this to match the pattern with alloc and whatnot but
these things end badly in the long term.  Please stick to what's
necessary.  The function can't fail.  Make its return type void.

Thanks.

-- 
tejun