Q-Bus Memory Diagnostics and Repair

[Jerry Weiss]

On Sep 9, 2016, at 6:11 PM, Noel Chiappa <jnc at mercury.lcs.mit.edu> wrote:
> 
>> If your CPU is an 11/73 (which can directly 'access' [hate that verbism
>> :-] all of memory from ODT, unlike the 11/23 which is restricted to the
>> bottom 256KB), try playing around with a failing location, and its
>> alternative, directly, and see if a store of random data into one can be
>> read back directly from the other
> 
> Note: The 11/'73' CPU powers up with the cache enabled, even for ODT!
> 
> So if you write xxx into some location, if you then read it back, you will get
> the correct data even if the memory location is busted - the CPU is getting
> the (correct) data from the cache. To have your 'memory' reads and writes
> actually go to the memory, you need to turn off the cache:
> 
>  17777746/ 02000 
> 
> Note that starting the machine does an INIT, which will again enable the
> cache.
> 

That’s a very good piece of information, I hadn’t considered that.  I have 11/73. 
I’ve checked the memory with ODT and can confirm the stuck bit.

I found that EK-MSV1Q-UG-002_MSV11Q_5-85.pdf describes how the CVMSAA diagnostic works.
"Section 1: Address Tests
These tests verify the uniqueness of every memory address.Test 1 writes and reads the
value of each memory word address into that memory location. After all memory has 
been written, all locations are checked again. “

I interpret this as mean it will deposit 1000 into @1000, and that will be read back as 1400
given the bad bit.  So if you have stuck bit, the test for uniqueness gives a false positive.

A comic ray probably took out a logic gate, as there is a pattern of stuck 1 bits in the chip.
As soon as I pick up a spare chip, I start to get invasive.  A careful clipped VCC is easily
restored if its the wrong chip.  

Jerry