Sunday, August 31, 2014

Progress on memory but not done yet - also printer close to working

I had an abbreviated work day on Sunday but did get a few key tasks accomplished. I didn't take any pictures today, making this a text only posting.


Having definitively identified the bad card causing the parity errors in the first 8K of memory, and temporarily booting the problem up to high memory by swapping cards between compartments, I then had to determine enough about the errors in high memory that I could know the bad part and work out a fix. This involved trying various loads and stores and many addresses in the upper 8K to find what were the fixed conditions that produced the problem.

There is one problem for addfresses of the form xx11 00xx xxxx xxxx due to the bad card I swapped into place - it means that no read occurs for any words in the 1K of memory beginning at that address. However, I was also shooting the problem of the 1 and 5 bits that always come back as being on, even though they were stored as zeroes.

I determined that this occurred in any address in the upper 8K, which is peculiar because there are virtually completely independent cards and wiring paths for each 4K half of the memory. I would have to believe that two cards, side by side, both failed in exactly the same way, if I were to trace this to a failing SLT card.

I checked power supplies, looked at physical conditions on the backplane and did lots of deep thinking to try to spot anything that could cause this problem for both. I am still stumped, although it could be a failure in a ribbon cable connecting the two storage compartments together.

I tried to scope the problem but didn't really get a good reading on the issue from that. I decided to swap cards around - the inhibit/sense cards for bits 1 & 5, at A4 and B4, work on the upper and lower half of memory respectively. I swapped them with their peers a row above, in A3 and B3, which control the 7&9 bits. The problem did not move with the cards, which supports my suspicion that the problem is not related to those cards. The sense and inhibit lines in the core stack itself are independent for the two 4K halves.

The signal to inhibit bit 1 and 5 come in via a cable from the other (low 8K) compartment, and are passed to a pair of cards which send the signal to their respective half of memory. If it were a card problem, it would mean a pair of cards with the exact same error.

Furthermore, when I powered up later and tried some code, I found that the 1&5 bit contagion had spread to the lower 8K of memory. That certainly sounds like cabling or a short on that part of the backplane. I didn't touch any cards in the low 8K compartment.


The 1132 does emit characters, respond to all the XIO commands and even prints in the right columns when I ask it to. The only issue is that somehow I am printing the wrong characters. this could be a timing issue with the encoder wheel in the printer, a defect in the cycle stealing circuitry, or a defect in my hand entered program. I would bet on the last as the cause, but I will need to get the memory working again before I can fix and rerun the program.

No comments:

Post a Comment