This looks great. Is there a suggestion of load type that I can
generate that should trigger the problem?  Do you have a specific test
case and/or scenario that I could try?  I don't have a way to
reproduce this otherwise.  If I did, I could build with this patch and
try it.  If I could eliminate this problem on my server, this would
be very good.

I looked in CVS, and it does not look like this code in 3.1/4.0/current yet.

-dgl-

>Hello,
>
>I patched from
>
>http://mail-index.netbsd.org/port-powerpc/2006/06/27/0001.html
>
>and it seems to work well.
>
>Takaaki Kobayashi