On Thu, 10 Apr 2014, Philip Guenther wrote:
> I'll float a diff later when I have a chance to test things, but I think
> the solution for now will be to change the condition on the
> while()tsleep() loop that the main thread blocks on and to move the
> wakeup().
Here's the diff I'm poking at, though I haven't had a chance to do
systrace and other tests yet.
Philip
Index: kern/kern_exit.c
===================================================================
RCS file: /cvs/src/sys/kern/kern_exit.c,v
retrieving revision 1.138
diff -u -p -r1.138 kern_exit.c
--- kern/kern_exit.c 30 Mar 2014 21:54:48 -0000 1.138
+++ kern/kern_exit.c 11 Apr 2014 06:42:28 -0000
@@ -158,12 +158,10 @@ exit1(struct proc *p, int rv, int flags)
TAILQ_REMOVE(&pr->ps_threads, p, p_thr_link);
if ((p->p_flag & P_THREAD) == 0) {
/* main thread gotta wait because it has the pid, et al */
- while (!TAILQ_EMPTY(&pr->ps_threads))
+ while (pr->ps_refcnt > 1)
tsleep(&pr->ps_threads, PUSER, "thrdeath", 0);
if (pr->ps_flags & PS_PROFIL)
stopprofclock(pr);
- } else if (TAILQ_EMPTY(&pr->ps_threads)) {
- wakeup(&pr->ps_threads);
}
rup = pr->ps_ru;
@@ -356,7 +354,8 @@ exit1(struct proc *p, int rv, int flags)
/* just a thread? detach it from its process */
if (p->p_flag & P_THREAD) {
/* scheduler_wait_hook(pr->ps_mainproc, p); XXX */
- --pr->ps_refcnt;
+ if (--pr->ps_refcnt == 1)
+ wakeup(&pr->ps_threads);
KASSERT(pr->ps_refcnt > 0);
}
