i think Michael Hitch and i have fixed the illusive "you must clear
memory before booting" pmap bug. the key is that hp300's clear their
memory (or at least mike hibler thinks they do).

i knew it had to be a 68040 specific pmap_bootstrap bug, because of
various symptoms i had seen. Michael finally noticed that Sysptmap
wasn't being completely initialized on the 68040.

after making the following change, i was able to back out the
workarounds that i required (they were: 1. locore.s cleared memory
from _end to the 1M boundary; 2. i had to let *NULL work -- now i can
PG_NV it).

things appear to be working better.  i hope this is the right
fix. i'll continue stress testing...

*** pmap_bootstrap.c.save	Sat Sep 23 13:08:37 1995
--- pmap_bootstrap.c	Sat Sep 23 03:54:09 1995
***************
*** 246,251 ****
--- 246,258 ----
  			*pte++ = protopte;
  			protopte += NBPG;
  		}
+ 		/*
+ 		 * Invalidate all but the last remaining entries in both.
+ 		 */
+ 		epte = &((u_int *)kptmpa)[NPTEPG-1];
+ 		while (pte < epte) {
+ 			*pte++ = PG_NV;
+ 		}
  		pte = &((u_int *)kptmpa)[NPTEPG-1];
  		*pte = lkptpa | PG_RW | PG_CI | PG_V;
  	} else {