hi,

there's no reason that I know of why m68k platforms should have such
low limits on kernel memory.  the numbers that you show are the upper
bounds on the various platforms, see kmeminit_nkmempages() for details.

the best thing would be for you to build kernels with NKMEMPAGES_MAX
set really high and see if that causes any problems on your
well-endowed machines.  if it works, then you can talk to the
port-maintainers about raising the maximum on the platforms where
you've tested.

-Chuck


On Mon, Jun 04, 2001 at 04:58:19PM -0400, John wrote:
> Hello,
> 
> I posted this to tech-kern and port-amiga, but now I think this might be
> most appropriate in port-m68k.
> 
> What I really would like to know is whether the MMU in the m68k is so
> different from that of i386 or PowerPC that the number of NKMEMPAGES
> should be significantly different between m68k and (i386, ppc, Alpha).
> 
> Should this be considered on all m68k ports that have a very modest amount
> of NKMEMPAGES? The breakdown:
> 
> Amiga:		384 (8k)
> Atari:		384 (8k)
> hp300:		1536 to 2048 (4k)
> luna68k		768 (4k)
> mac68k:		768 (4k)
> mvme68k:	1024 to 1536 (4k)
> news68k:	1024 (4k)
> next68k:	1024 (4k)
> sun3:		384 (8k)
> x68k:		768 to 1024 (4k)
> 
> Compare this to PowerPC, Alpha, and i386 (up to 32768); are they really
> THAT different?
> 
> With memory as cheap as it is, I have most of my m68k machines maxed out -
> I have several Quadras with 136 megs, three Amigas with 144 megs, one
> Amiga with 256. I even have 104 megs in my VAX (1024 pages, 4k sized pages)
> 
> If this doesn't get changed in the sources, then could someone at least
> tell me what a good guideline might be?
> 
> 
> The original post:
> 
> I've looked into the default values for NKMEMPAGES for all of the
> architectures NetBSD supports, and I'm a little puzzled. According to the
> comments in the code, the number of NKMEMPAGES is calculated by physical
> ram size times some machine dependent code.
> 
> Looking at i386 machines, this is true: a machine with 192 megs of memory
> has 12260 vm.nkmempages; a machine with 256 megs has 16280.
> 
> But Amigas have 384 always, whether 16 or 256 megs.
> 
> Is there something fundamentally different about the MMUs in m68k and
> i386? If there isn't, then shouldn't
> NKMEMPAGES_MAX_DEFAULT  ((3 * 1024 * 1024) >> PAGE_SHIFT)
> be increased in sys/arch/amiga/include/param.h?
> 
> Finally, should I assume that my 128 meg Amiga should have 4000 NKMEMPAGES
> (i386 has 8000 for 128 meg with 4k pages; Amiga uses 8k pages).
> 
> Thanks very much,
> John Klos
> -- 
> The proof of a system's value is its existence.