On Wed, 20 Dec 2000, Richard Earnshaw wrote:

> bjh21@netbsd.org said:
> > (FWIW, it'd be theoretically possible to have a combined arm26/arm32
> > kernel
> 
> Not really, since the pmap code would be almost entirely different;

That can be handled (function pointers are easy), as can the different
exception handling (you'd need two versions of zero page).

> not to 
> mention the fact that in 26-bit mode you have to be much more careful 
> about using "movs pc,..." in the right places, and not using the msr or 
> mrs instructions.  We'd practically end up with two of everything.

Once we get GCC to emit mode-neutral APCS-32 code, all the C code would be
dealt with.  There are a few parts of locore which would need to switch on
CPU type, but that's known technology.  All the tedious bits of the kernel
(VM, filesystems, networking) would share quite happily.  It's not a
sensible thing to do, but it indicates that "theoretically possible" isn't
the best guide.

> BTW, I wasn't saying the rule was hard and fast, just that it was a guide 
> -- the current situation is clearly wrong: we should move towards a single 
> GENERIC, or to separate ports -- the wrong move is not to move.

I agree.  Now all we need is for someone (possibly plural) to stand up and
volunteer to do one or the other.

-- 
Ben Harris                                                   <bjh21@netbsd.org>
Portmaster, NetBSD/arm26               <URL:http://www.netbsd.org/Ports/arm26/>