I made several changes over the weekend to how `soft' interrupts are
handled in the kernel.  I wasn't expecting much of a change except for
perhaps slightly more reliable serial port.  Informal testing, however,
indicates that far fewer clock ticks are lost being lost; after compiling
for 9 hours, my clock lost only 20 minutes as opposed to the 90 minutes I
would have expected (yes, I lose about 10 minutes per hour when doing
heavy compiles).

The results may not be as drastic as I've indicated, due to a number of
factors, but these changes certainly increase reliability and the
efficiency of `soft' interrupts.  If anything breaks, it's because we were
depending on a bug which this uncovers, but I expect no serious trouble as
I've been running one version or another of these changes since last week.

Note also that these new kernels are the first to be widely distributed
that were compiled with optimization at -O2; there should be a small
increase in overall performance.  Of course, all of the
machine-independent changes to the kernel have been included.

The GENERIC-21 and GENERICSBC-21 kernels have replaced their respective
-20 versions at:

	ftp://ftp.netbsd.org/pub/NetBSD/arch/mac68k/kernels/

These kernels have no other changes applied (such as the Quadra 700
internal video patch, or the patch to make the kernel work on a Quadra
AV).  People with spare time to test these kernels should email me any
reports of (non-)working status.

Thanks,

--scott