On Mar 16,  8:38am, Andreas Johansson wrote:
> I have seen this with many different kernels since the first Xamiga24.
> I have also tried with standard gcc-options and with -m68040 added.
> It seems pretty improbable that the compiler is creating bad code.

  The code generated by the compiler I am using (the stock gcc in the NetBSD
distribution) does generate correct code.

> Perhaps something with the '040 writeback cache?  Is it enabled in the
> register area of the RetinaZ3?

  Any I/O address space is supposed to be no cache and serialized.  Caching
I/O registers is extremely unlikely to work.

Michael

-- 
Michael L. Hitch			INTERNET:  osymh@montana.edu
Computer Consultant
Office of Systems and Computing Services
Montana State University	Bozeman, MT	USA