Re: [PATCH v10 2/2] sched/rt: Add support for SD_PREFER_SIBLING on find_lowest_rq()

[Steven Rostedt]

On Fri, 25 Aug 2017 10:31:00 +0900
Byungchul Park <byungchul.park@xxxxxxx> wrote:

> On Thu, Aug 24, 2017 at 09:55:14AM -0400, Steven Rostedt wrote:
> > On Thu, 24 Aug 2017 10:06:03 +0900
> > Byungchul Park <byungchul.park@xxxxxxx> wrote:
> >   
> > > Hello Steven,
> > > 
> > > Can I keep your reviewed-by with comments modified?  
> > 
> > Yes, but I have one nit.
> > 
> >   
> > > @@ -1668,9 +1691,35 @@ static int find_lowest_rq(struct task_struct *task)
> > >  				return this_cpu;
> > >  			}
> > >  
> > > -			best_cpu = cpumask_first_and(lowest_mask,
> > > -						     sched_domain_span(sd));
> > > +			/*
> > > +			 * If a cpu being in lowest_mask & current sd &
> > > +			 * ~prefer sd is valid, that becomes our choice.  
> > 
> > I can't parse the above sentence.  
> 
> I might have simplified the comment too much.
> 
> ~prefer means we want to exclude the domain span having SD_PREFER_SIBLING,
> since they should be considered as fallback candidates.
> 
> W/o ~prefer, a cpu in the domain might be chosen when we consider higher
> domain than the domain having SD_PREFER_SIBLING, as the best cpu, not
> fallback.

That actually wasn't the hard part. It was the entire sentence that is
very confusing. What is the definition of "is valid" in this context?
Do you just mean "< nr_cpu_ids"? If that's the case you are also being
redundant. Because if the "if" statement is true, then it will be
valid. What about just saying something like:


/*
 * If a cpu exists that is in the lowest_mask and the current sd span,
 * but not in the prefer sd span, then that becomes our choice.
 */

-- Steve