It's possible, but I'd be surprised if it were anything but software.
Many DG Nova programmers umm discovered a feature of locations (I think)
40 through 47, what we locally called the auto-excrement. The factory I
believe called them auto-increment and auto-decrement.
An indirect memory reference in this range Xcremented the value there
after reference. It was intended as an index-register-like feature, and
was a terrible place to put a return address stack.
tomj
Cores can die (intrinsic flaw, damage, overheat, etc) but it's
awwwwwwfully suspicious that all the bad bits are in the same word...
cores are manufactured in bit planes and wired into words, they have no
other relationship.
Core drivers could fail, but again all in the same words? Usually a
whole column will drop out and you'll see the Nth bit bad in some
largeish address range.
On Fri, 2004-06-25 at 15:07, Paul Koning wrote:
>>>> "Wai-Sun" == Wai-Sun Chia <waisun.chia(a)hp.com> writes:
Wai-Sun> I have MM11-DP (16kW parity core) in my /04 which seems to
Wai-Sun> have a little problem.
Wai-Sun> Several locations which I've checked seems to be losing
Wai-Sun> content (core is supposed to be non-volatile). I've been
Wai-Sun> debugging custom bootloaders for the past week and it has
Wai-Sun> since gone past annoying. It's like 3 locations out of 50
Wai-Sun> that are always reverting back to 000000 after a reboot.
Wai-Sun> Perhaps it's the driver logic to these 3 particular cores
Wai-Sun> that are not functioning properly? Or is it the cores
Wai-Sun> itself?
Or a bug in your program is clearing those three words?
It seems mighty peculiar for a hardware problem to consistently turn
three words to be all zero. (Then again, what was the original
contents? Losing one bit is possible -- losing many bits is
unlikely.)
Given how core works, if you can write it, then read it several times,
and it produces the right answers, it is REALLY unlikely that the bits
would just go away by themselves after that.
I can imagine some "bit rot" mechanisms, though it's a stretch -- but
those wouldn't work in nice whole-word patterns.
paul