One more before going to bed.

So far I implicated 64bit kernel would be designed in N64.

However, N32 kernel makes sense as well if the entire
system is assumed N32 only.  This makes possible to have
compact kernel and compact application while allowing
64bit integer and 64bit FP math.  The lowend and highend
2GB address spaces of 64bit range are used (Sign extension
is done handsomely for every 32bit PTR).  This would spoil
fancy cache controlling address ranges available in 64bit mode,
but it's not a lossage since our pmap.c in use does not run it
anyway.   Note that NetBSD/mips will not pursue this way,
I guess.  (no particular reason to prevent it, though, as long as
N64 kernel is going realized in such the way I mentioned...)

Toru Nishimura/ALKYL Technology