Manuel Bouyer <bouyer@antioche.lip6.fr> writes:

> > So, if (p_stat == SRUN && (p_flag & P_INMEM) != 0) then it's on run
> > queue, isn't it?
> 
> I'm not sure it's always true.

I believe that's the way done in other place.  Actually,
suspendsched() itself already implicitly relying on this.  If not,
some process with p_stat == SSTOP may left on run queue.  But
fortunately, penalty is just a redundant test.

> I looks safer to traverse the run queues.  Anyway, as we're at
> shutdown time the time spent in suspendsched() isn't really a big
> issue.

It's not safe since not all runnable processes are on run queue and
swapper is still working, isn't it?

enami.

void
suspendsched()
{
	struct proc *p;
	int s;

	/*
	 * Convert all non-P_SYSTEM SSLEEP or SRUN processes to SSTOP.
	 */
	proclist_lock_read();
	SCHED_LOCK(s);
	for (p = LIST_FIRST(&allproc); p != NULL; p = LIST_NEXT(p, p_list)) {
		if ((p->p_flag & P_SYSTEM) != 0)
			continue;
		switch (p->p_stat) {
		case SRUN:
			if ((p->p_flag & P_INMEM) != 0)
				remrunqueue(p);
			/* FALLTHROUGH */
		case SSLEEP:
			p->p_stat = SSTOP;
			break;
		case SONPROC:
			/*
			 * XXX SMP: we need to deal with processes on
			 * others CPU !
			 */
			break;
		default:
			break;
		}
	}
	SCHED_UNLOCK(s);
	proclist_unlock_read();
}