Charles M. Hannum writes:
 > 
 > So I just checked in a large batch of changes to modify the
 > pmap_enter() API as previously noted.
 > 
 > This has not been tested on other platforms than arm32.  However,
 > since all other platforms currently ignore the `access_type' argument,
 > it should not cause any problems (other than potential compilation
 > errors).

Well... I guess you warned me... but this one cost me several days...
Note the uninitialized variable "pa" that is passed to pmap_enter.
(Gee... Let's enter some RANDOM mappings!  What fun!:)

Watch out for those temptations to "tweak while there"...
Lesson: when in pre-release, do the minimum and get out!
(Listening Charles?)

diff -rc date0326/sys/arch/sun3/sun3/vm_machdep.c ...
*** ../date0326/sys/arch/sun3/sun3/vm_machdep.c	Wed Mar 24 00:51:15 1999
--- sys/arch/sun3/sun3/vm_machdep.c	Tue Apr  6 23:44:52 1999
***************
*** 311,323 ****
  
  	kpmap = vm_map_pmap(kernel_map);
  	while (len > 0) {
! 		pa = pmap_extract(kpmap, fva);
! 		if (pa != 0) {
! 			/* this does the cache flush work itself */
! 			pmap_remove(kpmap, fva, fva + NBPG);
! 			pmap_enter(kpmap, tva, pa,
! 			    VM_PROT_READ|VM_PROT_WRITE, 1);
! 		}
  		fva += NBPG;
  		tva += NBPG;
  		len -= NBPG;
--- 311,320 ----
  
  	kpmap = vm_map_pmap(kernel_map);
  	while (len > 0) {
! 		/* this does the cache flush work itself */
! 		pmap_remove(kpmap, fva, fva + NBPG);
! 		pmap_enter(kpmap, tva, pa,
! 		    VM_PROT_READ|VM_PROT_WRITE, 1, VM_PROT_READ|VM_PROT_WRITE);
  		fva += NBPG;
  		tva += NBPG;
  		len -= NBPG;