Re: Winnt question

Eric Smith (
4 Nov 1996 21:27:26 -0000

> WNT runs using different PALcode callbacks, both privileged and unprivileged.
> They are incompatible with Unix (Digital or Linux) just as OpenVMS's PALcode
> is incompatible.

Yes, my point was that perhaps a new version of PALcode could have a feature
to intercept all PALcode calls from a specific process (priviledged or not),
suspend that process, and pass info on that process to another process that
would handle it.

I haven't studied the architecture manual enough to be certain that this is
possible, but casual reading doesn't indicate any obvious reason why it
wouldn't be. It would have the disadvantage that it would reduce the
performance of all PALcode traps due to the need to test a mode bit.

It would really be a shame if there was no way to do this on an Alpha, which
seems otherwise to be the most elegant processor arcitecture yet. Motorola
processors have been capable of providing a full virtual machine since the
introduction of the MC68010 in 1981, and that's a CISC architecture! Designing
a RISC architecture that can support a virtual machine should be trivial by
comparison. Minicomputers and mainframes offered this capability even earlier.

> The real killer though is that WNT runs in a different addressing
> mode (superpage and NOT KSEG) over which it overlays it's braindead memory
> management scheme.

I don't think that should actually be too difficult a problem. WNT would
*think* it was running privileged, and it would *think* that it was mucking
with page tables, but in reality it would be running at user level in a Linux
process. The memory mamangement for WNT would be entirely simulated based on
the host OS/PALcode's paging system (perhaps using mmap() calls).


To unsubscribe: send e-mail to with
'unsubscribe' as the subject.  Do not send it to

Feedback | Store | News | Support | Product Errata | About Us | Linux Info | Search | JumpWords
No Frames | Show Frames

Copyright © 1995-1997 Red Hat Software. Legal notices