Re: [PATCH] sched_ext: Avoid relocking DSQ during remote consumption

[Tejun Heo]

Applied to sched_ext/for-7.4 with an unnecessary line break removed.

Possible follow-ups:

- The !dsq branch comment in scx_dispatch_dequeue() only mentions direct
  dispatch as a source of holding_cpu. Remote consume now reaches it too.
  Can you update the comment? The dsq branch comment also names
  dispatch_to_local_dsq() as the racer when it's
  unlink_dsq_and_switch_rq_lock().

- The remote local DSQ branch of move_task_between_dsqs() has the same
  relock. Calling dispatch_dequeue_locked() before dropping src_dsq->lock
  would avoid it. scx_bpf_dsq_move() isn't widely used, but it'd still be
  nice.

Thanks.

--
tejun