On Sat, Jul 08, 2000 at 06:57:53PM -0400, John Klos wrote:

> I remember that some people discussed the fact that the 1.4.2 kernel is
> significantly slower than the 1.4.1 kernel; the discussion was based upon
> comparisons on 68040 processors. Ignatios even said this was a 68040 cache
> issue.

It's not an "68040 cache issue" as in "somebody made a mistake making 68040
slow."

It is an "68040 cache issue" as in "the 68040 code used to be fast, but wrong
for a certain condition, and was fixed, making it, unfortunately, slower, but
there wasn't any choice doing otherwise".

> Since I am running a production server, I do not have the luxury of
> testing different kernels. However, I may be upgrading my server soon,
> which means that if I'm going restart the computer, that is the time to do
> it.
> 
> Does anyone know if the performance hit is as significant on 68060s as on
> 68040s? Or does it only affect 68040s?

This shouldn't affect your decision to upgrade. It fixes a bug, the
optimization introduced "back then" was incorrect. You wouldn't want a
compiler that produces 10% faster code that creates 10% wrong results,
right?

But you're lucky, it only affects 68040 cpus. Entry from CHANGES-1.4.2 is
added below for your convenience.

Regards,
	-is

sys/arch/amiga/amiga/sys_machdep.c              1.26-1.27
sys/arch/atari/atari/sys_machdep.c              1.21-1.22
sys/arch/hp300/hp300/sys_machdep.c              1.27-1.28
sys/arch/mvme68k/mvme68k/sys_machdep.c          1.19-1.20
sys/arch/next68k/next68k/sys_machdep.c          1.9-1.10
sys/arch/x68k/x68k/sys_machdep.c                1.18-1.19
sys/arch/mac68k/include/pmap.h                  1.27
sys/arch/mac68k/mac68k/sys_machdep.c            1.14-1.15

  Fix i-cache/d-cache synchronization for M68040 cpus (see M68040
  UM p.4-10).  Makes regress/sys/kern/sigtramp work even on M68040.
  Fix a typo which caused excessive cache flushes.