Re: NetBSD/vax kernel seems stable now

[Matt Thomas <matt%3am-software.com@localhost>]

On Feb 23, 2008, at 3:40 PM, Johnny Billquist wrote:

> 4000/500?
> I started a build of the world yesterday on a 4000/90. Let's see  
> who finishes first... :-)
>
> Anyway, nice to see someone else back on track as well.
> I think Matt's done an excellent job. My machine have been chugging  
> away for more than 24 hours now, and looks very solid.
> I'm at libutil right now...
>
> By the way, Matt. Have you done anything about the atomic ops for  
> VAX? And what should we do? After all, a CAS operation is hardly  
> optimal for the VAX (a shame really, when we have a nice set of  
> instructions which are guaranteed to be atomic). But do we even  
> have any working MP machines possible, or could we almost NOP the  
> atomic stuff?

Alas, ADAWI doesn't really help much.

if you look at the way CAS is emulated, it's not really that bad.

basically, it does block interrupts, locks one of 2048 mutexes, does
the cmp and possible store, unlocks the mutex, drops ipl.  If it's
not a multiprocessor kernel then no mutexes are taken.  All that is
done in 8 or 6 instructions (MP or UP).

I've love a better algorithm but I haven't been able to think of one.

Looking at the existing code, I doubt MP works anymore.  It'd be nice
to get that working again.