25 Sep
2014
25 Sep
'14
4:25 p.m.
this is the advice I was looking for. we already found a bad zero
address. hacking into the console is not a bad idea. I will look into
the schematic for that. right now we wrote a very small program
outputting all ascii characters to a serial vt220. that works and is
only 5 instructions:
0003 7001 increment ac +1
0004 6446 output ac to serial and set flag
0005 6441 skip if flag is 1
0006 5005 goto 5
0007 5003 goto 3
it really is just a start, but something formatting the address and
contents of addresses in octeds and some formatting of the output string
is feasable. fortunately this machine has 24Kword core, so there is
still some empty space. (or is it gone because of a faulty rewrite... we
have to test that)
simon
Although, if by restore machinery you mean the state-machine/sequencing-logic that does
the (re)-write, it can still be more subtle than that.
In my case it was just one bad address-wire driver transistor, such that the failure was
for only one direction of current flow - the 'write' or 'set' direction.
It failed as you suggest - could read once but failed to rewrite - but did not affect the
entire core array, just a sub-set of 64-word blocks.
--
Met vriendelijke Groet,
Simon Claessen
drukknop.nl
Due to a hardware fault in the core memory, I (had)
had only one opportunity to get a full dump, which I had missed due to checking memory
from the front panel and not knowing about the fault, so I didn't get the entire
contents. In my case hindsight is great, but it's something to keep in mind if
recovering the contents is of some concern.
That?s a good point. If the restore machinery is not working, you can only read once,
after that you?ll see zeroes. A quick check would be to pick some location and read it
several times. If that works, the restore logic is at least somewhat functional. Still, safest would be to hook to the console. A
BeagleBone Black has enough I/O lines to do that, I think, so it sounds like just some
level shifters and a modest amount of software.
paul