On Thu, Nov 07, 2019 at 01:51:47PM +0000, Valentin Schneider wrote:
> Using that, the fail is on:
> 
> 	if (need_pull_dl_task(rq, prev)) {
> 
> Which is most likely explained by the above call ending up doing a 
> 
>   dl_prio(prev->prio);
> 
> which doesn't play well with 
> 
>   class->pick_next_task(rq, NULL, NULL);
> 
> 
> Now, this is no longer an issue (I think) with the rest of Peter's series,
> since the above deref is gone with
> 
> 67692435c411 ("sched: Rework pick_next_task() slow-path")
> 
> It would be interesting to know whether LKP found this on a mainline kernel
> and bisected it down, or if it stumbled on this while bisecting something
> else.

That seems pausible indeed.