Subject: Re: Status of sun4c, sun4m and MP ports
To: None <dej@eecg.toronto.edu>
From: der Mouse <mouse@Collatz.McRCIM.McGill.EDU>
List: port-sparc
Date: 09/30/1994 14:31:22
[Disclaimer: I am not core, or anything remotely approaching it.  I'm
just someone running NetBSD/sparc and watching the mailing lists.  None
of this should be taken as official.]

> For this reason, I am seriously considering cutting over to NetBSD.
> The last time I looked, sun4c was supported.

It is, assuming you mean NetBSD-current or (not-yet-out, as far as I
know) NetBSD-1.0.  0.9 did/does not support the SPARC at all, as far as
I can tell.  The support's not great yet; the major shortcomings in my
opinion are that the sd driver doesn't support changing the disk labels
(gotta boot SunOS to repartition) and that DDB (the in-kernel debugger)
doesn't support disassembly or breakpoints.

> I have heard rumors that sun4m is also supported.

I saw a message fly by from someone who was working on such a port and
had it partially working, enough to boot diskless.  Unless the changes
verge on trivial I suspect it's unlikely 1.0 will support it; I get the
feeling the tree is getting pretty cold, nearly frozen for the release.

> A few questions:

> - How stable are the sun4c/sun4m ports?

More stable than SunOS, in my experience - but to be fair, my
experience with NetBSD involves one toy machine, and with SunOS, a user
database of hundreds, with upwards of twenty on at once (we regularly
run out of ptys on one machine, all the way through ttyrf).  To me, the
biggest win is not how stable the system is as delivered, but that when
SunOS breaks, all we can do is listen to "fixed in a future major
release" and just don't-do-that in the meantime, while when NetBSD
breaks, I can dive in and quite likely fix it.  (And when I can't,
well, I once worked on a project that had bought access to help from
Sun tech people.  This proved useless; they usually knew less than we
already did.  The one problem I've had with NetBSD that was beyond my
capabilities was fixed after about five email exchanges with someone
who really knew what was going on.)

>   Will they run on the newer hardware such as the Sparc5?

I don't know.  Want to loan me one so I can try it and find out? :-)

> - (Putting asbestos suit on) How difficult would it be to support the
>   dual-processor SparcServer 670/MP?

Oof.  I've done some highly-asymmetric multiprocessor work on
MicroVAX-IIs.  Supporting multiple processors is difficult only if you
want them to be fairly tightly coupled and close to symmetric.  If you
were willing for the two processors to appear as two distinct hosts,
with a very fast network connection between them, I feel sure that
would border on trivial - say, a week or two of work, most of it spent
figuring out the machine.  If all memory is dual-ported between the
processors and you're willing to put up with all syscalls being
forwarded to the "main" processor, that too would probably be fairly
easy; you'll get most of the win of true SMP if your applications are
cpu hogs, and nowhere near if they're syscall-bound.  The real hair
comes in when you want things to be symmetric.  In my brief dips into
the NetBSD kernel I've not seen anything that looks like locking
against other processors, though I haven't been looking for any such.

All this assumes a decent level of documentation.  Given that getting
usable docs out of Sun is like drawing blood from a turnip, this would
probably mean between a week and a few months uncompiling and reading
the Sun kernel and back-deducing the hardware interfaces.

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu