Re: NetBSD/vax vm for github testing

>> As for VAXen, [... i]n the one case I know enough about to be
>> confident of - the MicroVAX-II - I'm fairly sure it's not true of
>> real iron; different processor boards have independent clock
>> oscillators, which can never be _exactly_ the same speed, [...].

> MicroVAX-II was never used as Multiprocessor systems.

Not often, but hardly never.  I worked on a robotics project that used
a dual-CPU uVAX2 as the host (I wrote the kernel which ran on the
auxiliary CPU, and the support for it in the OS on the arbiter - the
arbiter ran BSD, the auxiliary ran the control loop).  If 35-year-old
memory is to be believed, I heard that VAXELN was a comparatively
popular use case for multiprocessor uV2s at the time.

> In general, the Qbus wasn't designed for multiprocessor systems, and
> while it is possible, it [does] require some things to be done
> differently on the other boards than the main CPU, which is the bus
> master.  Also, the MicroVAX-II have the memory on a private bus, so
> other CPUs would not be able to access it, except via DMA and
> mappings managed by the master, so it would be really hard to to MP
> uVAX-II.

Yes, it is AMP, not SMP.  It also ran into suit-imposed restrictions in
many cases, because DEC made it really hard to get a KA630 without a
MicroVAX around it (the version I heard was that they wanted to make it
hard to build third-party machines by buying CPUs and building the
rest); the processor boards which were easy to get on their own were
KA620s, which are almost identical, just tweaked enough that most OSes
wouldn't run on them (process pagetables are in physical space, not
system space).

KA6x0s also had a hardware bug.  When CPU X pokes CPU Y's
interprocessor doorbell, a small fraction (in the tens-to-hundreds of
ppm, IIRC) of the time the interrupt will get lost.  There was an etch
run that crossed the whole board; between the inherent capacitance and
the driving impedance, the timing was marginal - what I heard was that
it never should have worked at all, and wouldn't have but for
overdesign.  There was an ECO that reduced a resistor, for the few
people for whom an occasional dropped doorbell interrupt was a problem.

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		[email protected]
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B