> Oh - the child times were completely wrong before, they were only
> added in if the parent asked for the childs resource use in exit().
> 
> Since I know how this code works, I'll sort out a fix shortly.

Yes - it wasn't my fault, the bug has been there for (probably) ever.
My test kernel tree won't show it because I changed p->p_rtime from
struct timeval to a uint64_t (of usecs).

This fixes it:

Index: kern_resource.c
===================================================================
RCS file: /cvsroot/src/sys/kern/kern_resource.c,v
retrieving revision 1.69
diff -u -p -r1.69 kern_resource.c
--- kern_resource.c	2003/03/05 11:44:01	1.69
+++ kern_resource.c	2003/03/14 19:14:20
@@ -421,13 +422,18 @@ calcru(p, up, sp, ip)
 			microtime(&tv);
 			sec += tv.tv_sec - spc->spc_runtime.tv_sec;
 			usec += tv.tv_usec - spc->spc_runtime.tv_usec;
+			if ((long)usec < 0) {
+				usec += 1000000;
+				sec--;
+			}
 		}
 	}
 
 	tot = st + ut + it;
 	if (tot == 0) {
 		/* No ticks, so can't use to share time out, split 50-50 */
-		if (usec > 1000000 && (sec & 1)) {
+		if (usec >= 1000000 && (sec & 1)) {
 			usec -= 1000000;
 			sec++;
 		} else {

The second bug is mine :-( but extremely unlikely)

	David

-- 
David Laight: david@l8s.co.uk