Next project: 11/24. Does it need memory?--Working-ish!

Noel Chiappa jnc at mercury.lcs.mit.edu
Wed Oct 28 18:48:49 CDT 2020


    > From: Chris Zach

    > The last board is a new style 11/24 and it doesn't do anything but it
    > does have three switch packs. So I just need to find the manual for it

Appendix D in the 003 rev of the /24 TM has the details of the -YA..


    > In the meantime the UNIBUS problem also appears to be fixed: The
    > problem was that sure enough: One of the memory slot SPC's (4) did not
    > have the DMA jumper.

Checking grant line continuity is part of system setup for me; the fact than
on many backplanes, the DMA grant jumpering is wirewrap on the back of the
backplane is a PITA, but back when it wasn't that big a deal. (Heck,
configuration jumpering on boards was all soldered wires BITD!)

The need for grant continuity is actually a characteristic of the _CPU_, not
the bus. I was astonished the first time I played with an /04 (or naybe it
was a /34, I forget), because on them, I'm pretty sure (I can't find it in
the documentation, but I'm pretty sure I remember it) that on power-on, the
CPU checks for grant continuity, and won't work without it. (They all require
the M9302 teminator or equivalent, which has active circuitry to turn around
an unclaimed grant onto the BSACK line.)

When the CPU powers up, it sends grants down the grant lines and expects to
see them back on SACK (from the M9302), and it won't operate if this doesn't
happen.  That floored me when I ran across it, because on the older UNIBUS
CPUs I'd worked with BITD (/40's, etc) it was the done thing to be able to
leave a gap in the grant _downtream_ of the last card that could do
interrupts; no possible operational problem.

I'm not sure why DEC added that 'feature'; probably somebody thought it would
make systems more robust, but I bet all it did was generate a bunch of Field
Circus calls. (Kind of like the LSI-11 'feature' where ODT won't start unless
there's working memory at 0 - another wonderful little Easter Egg.)

Ironically, although the QBUS uses _exactly_ the same kind of bus grant lines,
no QBUS processor seems to check their continuity in this way; I habitually
run my QBUS machines without all the slots filled, and F11 and J11 processors
all run fine with open grant lines (after the last card that uses
interrupts). I guess DEC found out the hard way that that check wasn't
useful!

But this, as I said, is a characteristic of the CPU, not the bus.


    > Anyone even know why Q-Bus doesn't need termination

QBUS documentation has long ediscussions of maximum cable lengths between
backplane sections, but little about cases where termination on both ends is
not needed. One exception is in the " pdp11 bus hanbook", pg. 128:

  If nore than 20 AC loads are included [on a single bacplane], the other end
  of the bus must be terminated with 120 ohms.

The clear implication is that with <20 AC loads, termination at the second
end is not needed. No explanation is given as to why, but I had the same
thoughts as Patrick Finnegan - that it's down to the very short bus lengths.
(With a short bus, the reflection from the un-terminated end will be
very close in time - the line length for the lines on a single backplane
is spec'd at a max od about 14"; so about 1 nsec.  That's on the
order of magnitude of the bus signal rise times.)

If you think about it, every time you plug in a card, you've added a
short branch to the bus on every bus line (since they aren't normally
terminated at the transmitter/reveiver chips). But very short
branches, so their reflections are even closer in time.

	Noel


More information about the cctech mailing list