Re: [patch 4/4] sched: Distangle worker accounting from rq->lock

[Tejun Heo <tj@kernel.org>]

Hello, again.

So, let's get it correct first.

On Wed, Jun 22, 2011 at 05:52:15PM -0000, Thomas Gleixner wrote:
> The worker accounting for cpu bound workers is plugged into the core
> scheduler code and the wakeup code. This is not a hard requirement and
> can be avoided by keeping track of the state in the workqueue code
> itself.
> 
> Keep track of the sleeping state in the worker itself and call the
> notifier before entering the core scheduler. There might be false
> positives when the task is woken between that call and actually
> scheduling, but that's not really different from scheduling and being
> woken immediately after switching away. There is also no harm from
> updating nr_running when the task returns from scheduling instead of
> accounting it in the wakeup code.

I think false positives on schedule() should be safe.  As said
earlier, the gap between ttwu and actually running is a bit worrisome
but it might be nothing, but please at least describe the behavior
change.

> Index: linux-2.6/kernel/workqueue.c
> ===================================================================
> --- linux-2.6.orig/kernel/workqueue.c
> +++ linux-2.6/kernel/workqueue.c
> @@ -137,6 +137,7 @@ struct worker {
>  	unsigned int		flags;		/* X: flags */
>  	int			id;		/* I: worker id */
>  	struct work_struct	rebind_work;	/* L: rebind worker to cpu */
> +	int			sleeping;	/* None */

bool?

> -struct task_struct *wq_worker_sleeping(struct task_struct *task,
> -				       unsigned int cpu)
> +void wq_worker_sleeping(struct task_struct *task)
>  {
> -	struct worker *worker = kthread_data(task), *to_wakeup = NULL;
> -	struct global_cwq *gcwq = get_gcwq(cpu);
> -	atomic_t *nr_running = get_gcwq_nr_running(cpu);
> +	struct worker *worker = kthread_data(task);
> +	struct global_cwq *gcwq;
> +	int cpu;
>  
>  	if (worker->flags & WORKER_NOT_RUNNING)
> -		return NULL;
> +		return;

This doesn't look safe.  It can race with trustee_thread() setting
WORKER_ROGUE.  Let's just grab gcwq->lock on entry to
wq_worker_sleeping() for now; then, the schedule() trickery in
trustee_thread() can go away too.  This also means we can remove the
weird sync rules from ->flags and ->idle_list and just use simple
gcwq->lock for those, which is pretty nice.

> -	/* this can only happen on the local cpu */
> -	BUG_ON(cpu != raw_smp_processor_id());
> +	if (WARN_ON_ONCE(worker->sleeping))
> +		return;

Re-entrance is prevented by the scheduler hook being called only for
non-premption schedule().  Maybe it's better to explain that in the
function comment?

Hmmm... Also, I think worker->sleeping should be cleared by
trustee_thread() when WORKER_ROGUE is set for the worker; otherwise,
it can get out of sync and the above WARN_ON_ONCE() will trigger.

Thanks.

-- 
tejun