Core volatility

From: Tony Duell <ard_at_p850ug1.demon.co.uk>
Date: Fri Jun 25 17:14:22 2004

>
> I have MM11-DP (16kW parity core) in my /04 which seems to have a little
> problem.
>
> Several locations which I've checked seems to be losing content (core is
> supposed to be non-volatile). I've been debugging custom bootloaders

Core is non-volatile, but readout is destructive (you read a core by
writing a 0 to it and seeing if there was a change in flux through the
core). One PDP11s (and on just about every other mini I've worked on) the
core memory hardware restores the old contents after a read (Actually, on
a PDP11 it's more compiicated than that. There's a particular type of
read cycle -- a DATIP cycle -- where the processor ias actually going to
do a read-modify-write to that location in core, and the next bus cycle
is going to be a write to the same location. The DATIP cycle tells the
core contol electronics not to waste time doing the restore since it's
going to be overwritten anyway)

> for the past week and it has since gone past annoying. It's like 3
> locations out of 50 that are always reverting back to 000000 after a reboot.

Between reboots can you read that location as many times as you like? In
other words is the restoration circuit working correctly? (I would guess
it was, because it's common to all locations in the memory, but maybe
marginal drive currents can cause problems).

> Perhaps it's the driver logic to these 3 particular cores that are not
> functioning properly? Or is it the cores itself?

I can't think of any reason why particular locations should be getting
cleared by a reboot. Addressing is done by an XY matrix kind of scheme,
so a problem with the address drivers would normally affect an entire row
or column (And anyway, why would it only show on a reboot). Are you sure
no software is overwriting these locations?

-tony
Received on Fri Jun 25 2004 - 17:14:22 BST