this appears to do the trick, both when waiting for a particular pid
and when waiting for any child.  when I also apply some enhancements
to the pool code (which I'll post about shortly), the fork+exit benchmark
shows linear scaling up to 8000 processes (which is as many as we can
have on i386 with the current GDT code).

-Chuck


On Thu, Nov 06, 2003 at 10:27:48AM +0000, David Laight wrote:
> > My current plan is to count the number of children in SSTOP/SZOMB
> > and to requeue them to the top of the p_children list.
> > find_stopped_child() can then be optimised the case when there are 
> > no such children.
> 
> Patch at http://www.btinternet.com/~david.laight/netbsd/stopchild.diff
> it's about 18k, mainly due to fixing some related MP locking issues.
> 
> It should apply to current (as now!), anyone able to test the
> benchmark?  I've only tested the kernel on a very slow system....
> 
> 	David
> 
> -- 
> David Laight: david@l8s.co.uk