>Jonathan...

>IT WAS NEVER CHANGED!  This is the way it's been since, like, day one for
>NEtBSD.

Jason,

4.4BSD really did support other APIs than the one used by the hp300 pmap.

So, okay, you're saying that at some point the way the hp300 way was
decreed ``correct'' for NetBSD and other semanitcs were deemed
incorrect.  Which would be fine. No problem. The only gotcha is that
the decree was a bit quiet.

As I said to Castor *and* to you, I'm happy to do the right thing.
Castor says that the changes for UMV (which _are_ correct for UVM;
Michael and I ran them by Chuck) broke the old Mach-derived VM system.

Personaly I'd sooner just deprecate the old Mach VM on mips.  I take
it you' dhave no complaints if we just make the mips pmap module do
what you now say is ``right'', even if Castor says it breaks the old
VM?

I still think having the VM system depend on the pmap tracking
_reference_ bits, in order to correctly write back _dirty_ pages
(given that dirty bits *are* being correctly tracked) doesn't seem
``right'' -- because of TLB hardware which doesn't have reference bits.

I've said as much to Chuck. We'll see what happens.