Re: [Freemint-discuss] Memory Protection on Falcon/CT60

[David G. <dga...@gm...>]

2018-05-06 4:38 GMT+02:00 Mark Duckworth <mdu...@at...>:
> 1.17 is not perfectly stable but it is orders of magnitude more stable,
> to the point that you can get some use out of it.  For instance i can
> run conholio on a busy terminal for days with no problem.  I do believe
> now though that the later kernels are just better at showing a hardware
> issue that is present in 1.17 as well.
>
> With the ethernat pulled 1.19 4c3 seems perfect.  I'll need to put an
> ethernec in to really beat it up and verify in all scenarios but I
> believe the issue is some kind of bus loading/interrupt issue on the 060
> side.  I have as far as I know all the most recent ABE/SDR and CTPCI
> images and the ones that are valid for my setup.
>
> My CTPCI was corrupting IDE transfers too.  I am curious if it still
> does it with the ethernat removed now that things seem more stable.

Your problems seems very similar with what I had with my old CT60.
I noticed that something was wrong while compiling intensive stuff
like the kernel, I got random compiler errors, memory protection was
off probably.
Things got worse when I got the Supervidel and I plugged it together
with the EtherNat. After some tests I realized that the only stable
configuration was having only one card plugged with the CPU speed set
at 66 Mhz, any increase of the speed made the system unstable again,
CPU was a rev.6.
Then during a Sommarhack party in Sweden together with the "Nature
brothers" we did many tests, we tried exchanging components in our
systems, the Falcon, memory modules, cards even the CPU, at the end
all pointed to my CT60, we made sure we have the same firmware, but
for some reason my CT60 didn't like overclocking and having more than
one card attached.
The theory suggested then to explain my CT60 behavior was the
malfunction of one of the CPLD chips, I don't remember if it was the
ABE or the SDR. I remember we talked about that there were different
versions of the same chip, one with lower specification for a property
that I don't remember but the unit of measure was in nanoseconds,
there were versions with 20 ns for this property and others with 25
ns, or some similar values, we speculated on some CT60 being assembled
with one of the lower version of the chip although marked as a better
one.