Re: [PATCH v3 0/5] sync a se with its cfs_rq when att(det)aching it

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Aug 20, 2015 at 07:46:09PM +0900, Byungchul Park wrote:
> On Thu, Aug 20, 2015 at 05:38:41PM +0900, Byungchul Park wrote:
> > On Thu, Aug 20, 2015 at 03:17:21AM +0200, Peter Zijlstra wrote:
> > > 
> > > I did something like this on top.. please have a look at the XXX and
> > > integrate.
> > 
> > i am not sure, what do you intend for me to do.
> > 
> > do you mean that i am supposed to integrate this cleanup patch you gave me
> > including the XXX comment?

No, the intent was for you to think about the point marked XXX, which
you've done below.

> > > +	 *
> > > +	 * XXX this appears wrong!! check history, 
> > > +	 * we appear to always set queued and RUNNING under the same lock instance
> > > +	 * might be from before TASK_WAKING ?
> > >  	 */
> > 
> > is it impossible to happen to check if vruntime is normalized, when doing
> > something like e.g. active load balance where queued != TASK_ON_RQ_QUEUED
> > and p->state == TASK_RUNNING?
> 
> furthermore, in any migration by load balance, it seems to be possible..
> 
> > 
> > i think it can happen..

OK, then we need to change the comment to reflect the actual reason the
test is needed. Because I think the currently described scenario is
incorrect.