Re: [RFC v4 PATCH 3/7] sched: Enforce hard limits by throttling

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

On Fri, Dec 04, 2009 at 05:09:55PM +0100, Peter Zijlstra wrote:
> On Tue, 2009-11-17 at 20:05 +0530, Bharata B Rao wrote:
> > sched: Enforce hard limits by throttling.
> > 
> > From: Bharata B Rao <bharata@linux.vnet.ibm.com>
> > 
> > Throttle the task-groups which exceed the runtime allocated to them.
> > Throttled group entities are removed from the run queue.
> 
> This patch is just vile, all those _locked variants should really go.
> 
> Nor it is entirely clear why they're there.
> 

update_curr() is the place where I check if the group has exceeded its
runtime and it needs to take cfs_rq->cfs_runtime_lock. However there are
2 places from where update_curr() gets called with cfs_rq->cfs_runtime_lock
already held. Hence _locked() version of update_curr() exists.

These two call paths are both enqueue paths (enqueue_task_fair and
enqueue during unthrottling when period timer fires). Hence _locked()
versions of entity_enqueue() exists.

I see that you don't have this sort of requirement (of holding
rt_rq->rt_runtime_lock) in rt. I will recheck on this one to see why you are
able to do this in rt and I can't in cfs. If convinced, I shall get rid
_locked versions.

Thanks Peter for taking time to review the hard limit patches.

Regards,
Bharata.