On 13 May 1997, Aaron Brown wrote:

> Last time I looked into this, it looked pretty difficult. The Ultras
> are pretty radically different from the old Sparcs; they have the
> 64-bit address space, a different processor memory model (PSO, I think),
> a fourth distinct MMU, a new memory bus architecture, and different 
> devices. So a sun4u would be much more difficult than the Sun4m port,
> especially if we wanted to keep one NetBSD/sparc instead of NetBSD/sparc-32
> and NetBSD/sparc-64.

They're not quite that different.  While there is a new memory model
(RMO), Ultra I and Ultra II CPUs support PSO and TSO as well.  The address
space is 64-bit and the MMU is radically different, not to mention the
ASIs, ASRs, and a whole bunch of new instructions, which makes a unified
32-bit/64-bit kernel a bad idea.  However, the SBus architecture is
similar and the Ultra 1 models 140 and 170 (not to be mistaken for 140E,
170E and 200E) use most of the same devices as sun4m machines.

A 32-bit sun4u port is probably only a little more work than the sun3x
port.  A 64-bit port, such as what the Linux guys are claiming, has
compiler and toolchain issues and is therefore much more difficult.

=========================================================================
Eduardo Horvath				eeh@btr.com
"Cliffs are for climbing.  That's why God invented grappling hooks."
					- Benton Frasier