Re: [CFS Bandwidth Control v4 3/7] sched: throttle cfs_rq entities which exceed their local quota

[Bharata B Rao <bharata@linux.vnet.ibm.com>]

Hi Peter,

I will only answer a couple of your questions and let Paul clarify the rest...

On Wed, Feb 23, 2011 at 02:32:13PM +0100, Peter Zijlstra wrote:
> On Tue, 2011-02-15 at 19:18 -0800, Paul Turner wrote:
> 
> 
> > @@ -1363,6 +1407,9 @@ enqueue_task_fair(struct rq *rq, struct 
> >  			break;
> >  		cfs_rq = cfs_rq_of(se);
> >  		enqueue_entity(cfs_rq, se, flags);
> > +		/* don't continue to enqueue if our parent is throttled */
> > +		if (cfs_rq_throttled(cfs_rq))
> > +			break;
> >  		flags = ENQUEUE_WAKEUP;
> >  	}
> >  
> > @@ -1390,8 +1437,11 @@ static void dequeue_task_fair(struct rq 
> >  		cfs_rq = cfs_rq_of(se);
> >  		dequeue_entity(cfs_rq, se, flags);
> >  
> > -		/* Don't dequeue parent if it has other entities besides us */
> > -		if (cfs_rq->load.weight)
> > +		/*
> > +		 * Don't dequeue parent if it has other entities besides us,
> > +		 * or if it is throttled
> > +		 */
> > +		if (cfs_rq->load.weight || cfs_rq_throttled(cfs_rq))
> >  			break;
> >  		flags |= DEQUEUE_SLEEP;
> >  	}
> 
> How could we even be running if our parent was throttled?

The task isn't running actually. One of its parents up in the heirarchy has
been throttled and been already dequeued. Now this task sits on its immediate
parent's runqueue which isn't throttled but not really running also since
the hierarchy is throttled. In this situation, load balancer can try to pull
this task. When that happens, load balancer tries to dequeue it and this
check will ensure that we don't attempt to dequeue a group entity in our
hierarchy which has already been dequeued.

> > @@ -1438,10 +1524,16 @@ static void account_cfs_rq_quota(struct 
> >  
> >  	cfs_rq->quota_used += delta_exec;
> >  
> > -	if (cfs_rq->quota_used < cfs_rq->quota_assigned)
> > +	if (cfs_rq_throttled(cfs_rq) ||
> > +		cfs_rq->quota_used < cfs_rq->quota_assigned)
> >  		return;
> 
> So we are throttled but running anyway, I suppose this comes from the PI
> ceiling muck?

When a cfs_rq is throttled, its representative se (and all its parent
se's) get dequeued and the task is marked for resched. But the task entity is
still on its throttled parent's cfs_rq (=> task->se.on_rq = 1). Next during
put_prev_task_fair(), we enqueue the task back on its throttled parent's
cfs_rq at which time we end up calling update_curr() on throttled cfs_rq.
This check would help us bail out from that situation.

Regards,
Bharata.