Re: Anomalies while handling p_nstopchild count

[Robert Elz <kre%munnari.OZ.AU@localhost>]

    Date:        Tue, 13 Oct 2015 02:27:47 +0000
    From:        Taylor R Campbell <campbell+netbsd-tech-kern%mumble.net@localhost>
    Message-ID:  <20151013022635.578F5607B8%jupiter.mumble.net@localhost>

  | In this case, and generally[*], you need a membar_consumer matching
  | the membar_producer in order for the latter to have any useful effect.

OK, understood (your example code had two lines in the wrong order,
I believe, for the effect you intended, but that doesn't matter.)

  | (b) There's a missing membar_consumer in find_stopped_child and in
  | proc_reparent (but maybe it doesn't matter on x86, if all current
  | implementations will not in practice reorder those loads).

It probably doesn't really matter anyway - the only effect would be to
defer find_stopped_child() noticing the stopped child in question until
its next iteration.  Not good, but not fatal.

  | (c) The write to p->p_waited in exit1 looks fishy: it holds p->p_lock,
  | but not proc_lock.

It was fishy.   That's the one Paul just fixed.  It holds proc_lock
now (as of a couple of hours ago.)

kre