[Simh] MicroVAX 3900 simulator fails BIST sometimes?

[Mark Pizzolato]

On Saturday, June 23, 2018 at 12:41 PM, Robert Armstrong wrote:
> > Mark Pizzolato <Mark at infocomm.com> wrote:
> >Heisenberg effects mean that the problem is never seen
> 
>   Yep, the uncertainty principle applies to software as well...
> 
>   I'm just curious - do we (well, I don't, but somebody) have source listings for
> the KA655 EPROM?  Supposedly one of the other hex numbers in the "?53 2 0A
> FF 00 0000" message identifies the exact subtest that's failing.  It might be
> interesting to know that.

The listings we have are for a close, but not precise, version that the ROM 
is built from.  Without regard to precise listings, the listings don't actually 
have a mapping of the test numbers (and sub test numbers) to specific 
tests.   The assignment of tests is magically built by small macros that briefly
define PSECTs that each tests contributes a little content towards.  There 
must be some nice tool that can decode the resulting collection of test 
numbers and map them back to particular tests.  Maybe a special 
interpretation of data in the link map might help get there, but how it is 
done isn't particularly obvious.

ROM debugging is a very tedious process.  You poke around in the
ROM binary and disassemble  a section.  Then you locate particular instruction 
Sequences in the listing.  Or you do the opposite (start from the listing and
look for instruction patterns in the ROM).   You then put simh breakpoints in 
some ROM locations that you want to stop at.  The presence of breakpoints
adds to Heisenberg variances since simh breakpoints are implemented by 
checking instruction fetch addresses rather than modifying the target code to 
insert actual breakpoint instructions (which are architecturally meaningful 
within the VAX simulation).  For most things Heisenberg doesn't come into 
play, but timing calibration stuff is the exception.

Anyway, this problem is benign enough, and the lack of solid debug data 
has it sitting there unsolved.

- Mark