UK PDP11 wanted

Noel Chiappa jnc at mercury.lcs.mit.edu
Tue Apr 14 16:16:18 CDT 2015


    > From: Johnny Billquist

    >> http://www.classiccmp.org/pipermail/cctalk/2015-February/003756.html

    > Cool.

Sure; glad you liked it.


    >> But the QBUS part of _any_ QBUS backplane is not 'directional'
    >> ...
    >>    http://www.classiccmp.org/pipermail/cctalk/2015-February/003757.html

    > there is still a question on whether the CPU just pass all signals
    > through that you need.

Perhaps I'm confused, but the only QBUS signals which are not 'broadcast'
(i.e. sent down a wired-OR bi-directional bus transmission line to all cards)
are the grant lines. The CPU doesn't pass them through, it originates them -
but memory cards ignore those lines anyway?

    > With PMI memory, it obviously works somewhat different, since the CPU
    > sources signals in just one direction (hence the reason the PMI memory
    > have to be before the CPU)

Actually, the PMI signals are all (I'm pretty sure, but have not checked) just
like most other QBUS signals - 'broadcast' bi-directionally down a
transmission line. The 'PMI memory before the CPU' thing in 11/83's is simply
an artifact of the way 'standard' Q/CD backplanes are wired on the CD side
(used for PMI) - see that second post for details.

In the 11/84 backplane, the PMI memory lines _are_ bussed to all the slots,
just like the QBUS lines, which is why 11/84 PMI memory cards go after the
CPU. (So you could in fact have an empty slot between the CPU and the first
PMI memory card in an 11/84, and it should still work.)

    > [the CPU] also interacts/intercepts with all other memory accesses to
    > the PMI memory.

My impression is that the CPU does not get involved in DMA access to PMI
memory.

In an 11/83, I gather the DMA devices just do ordinary QBUS memory cycles to
it (see the MSV11-J manual, pg. 1-1, middle). In an 11/84, with UNIBUS
adapter, IIRC the UNIBUS adapter does PMI cycles direct to the memory (but
don't quote me on that, I haven't checked the 11/84 documentation, although
the MSV11-J manual indicates it does - see pg. 1-1, bottom).

    > Which is yet another reason why the PMI memory sits before the CPU. If
    > you put PMI memory after the CPU, it will instead work as normal Qbus
    > memory.

I'm pretty sure that whether PMI memory works as PMI memory or QBUS memory
depends only on what it hears on the busses, and nothing else. If it sees a
PMI cycle for it, it does a PMI cycle; if it sees a QBUS cycle for it, it does
a QBUS cycle.*1 So in an 11/84, where it's on the bus after the CPU, it does
PMI cycles with the CPU because the CPU does a PMI cycle.

(1: Note that the MSV11-JB and -JC variants don't do QBUS properly, and are
PMI only, and so can't be used in an 11/83. I seem to recall hearing it's
just QBUS block mode transfers that they don't do properly?)


Which all does raise an interesting question: if an M8190 is in a system with
a _mix_ of PMI and regular QBUS memory, do accesses to both memories work
fine? (E.g. if it's in a Q/CD backplane with a PMI board before it, and a
regular QBUS memory card after it.)

Something to try, I guess - I have a working Q22/CD box, and plenty of cards
- should just plug them in and try it!

	Noel


More information about the cctalk mailing list