> : Probably some heuristic. When you add up all RAM reported by the BIOS you
> : get less than is really installed. Except on a Tyne which reports 256MB
> : to much ...
> Hmmm, if the BIOS is gone when the kernel is started, how do I get
> that information out of the BIOS? Don't I need that to find out where
> the tags start? I notice the PICA and Jazz based machines check a
> register that the R4400PC doesn't seem to have :-(.
A reasonable heuristic is to query the ARC BIOS for the RAM, filter out
obvious wrong answers (like on the Tyne), the round up the end address of
the last chunk returned to the next multiple of two.
Or, if your machine doesn't throw exceptions for accessing non-existing
memory, do test by writing date to the memory.
These two methods will of course fail for machines with multiple, non
continuously mapped chunks of memory - the ARC standard and portable
bootloader standard allow them - but I didn't encounter one of them yet.
Well, except that 640k-1024k hole in your machine.
Completly unrelated - the new libc 960501 is significantly better than
the old. I've just built a usable bash and do a native recompile of
the lib to get rid of some crosscompiltion artefacts.