Re: NULL pointer dereference in pick_next_task_fair

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Nov 07, 2019 at 07:27:53PM +0000, Quentin Perret wrote:
> On Thursday 07 Nov 2019 at 19:43:56 (+0100), Peter Zijlstra wrote:
> > But you mean something like:
> > 
> > 	for (class = prev->sched_class; class; class = class->next) {
> > 		if (class->balance(rq, rf))
> > 			break;
> > 	}
> > 
> > 	put_prev_task(rq, prev);
> > 
> > 	for_each_class(class) {
> > 		p = class->pick_next_task(rq);
> > 		if (p)
> > 			return p;
> > 	}
> > 
> > 	BUG();
> > 
> > like?
> 
> Right, something like that, though what I had was basically doing the
> pull from within the pick_next_task_*() functions directly, like we were
> doing before. I'm now seeing how easy it is to get this wrong, and that
> even good-looking code in this area can be broken in very subtle ways,
> so I didn't feel comfortable refactoring again so close to rc7. If you
> feel more confident, I'm more than happy to test a patch implemeting the
> above :)

Thing is, if we revert (and we might have to), we'll have to revert more
than just the one patch due to that other (__pick_migrate_task) borkage
that got reported today.