Re: [PATCH] workqueue: remove the unneeded cpu_relax() in __queue_work()

[Tejun Heo <tj@kernel.org>]

On Thu, May 22, 2014 at 04:44:16PM +0800, Lai Jiangshan wrote:
> When pwq->refcnt == 0, the retrying is guaranteed to make forward-progress.
> The comment above the code explains it well:
> 
> 	/*
> 	 * pwq is determined and locked.  For unbound pools, we could have
> 	 * raced with pwq release and it could already be dead.  If its
> 	 * refcnt is zero, repeat pwq selection.  Note that pwqs never die
> 	 * without another pwq replacing it in the numa_pwq_tbl or while
> 	 * work items are executing on it, so the retrying is guaranteed to
> 	 * make forward-progress.
> 	 */
> 
> It means the cpu_relax() here is useless and sometimes misleading,
> it should retry directly and make some progress rather than waste time.

cpu_relax() doesn't have much to do with guaranteeing forward
progress.  It's about giving a breather during busy wait so that the
waiting cpu doesn't busy loop claiming the same cache lines over and
over ultimately delaying the event being waited on.  If you're doing a
busy wait, you better use cpu_relax().

Thanks.

-- 
tejun