SRM memtest

[J.C. Roberts]

I noticed something odd on an Alpha Personal Workstation 433 that I got
off of eBay. The ARC/AlphaBIOS would occasionally report 256MB rather
than the usual 384MB. This weirdness was intermittent. I have reseated
everything in the system to make sure there are no connection/connector
issues but I think it would be prudent to actually test the memory
itself.

I kicked the system into SRM Console mode and I've been trying to run
memtest to no avail. I believe *I* am the real problem since I don't
know what the heck I'm doing in SRM in spite of the fact that I've read
the SRM Console user guide.
http://ftp.digital.com/pub/Digital/info/semiconductor/literature/srmcons.pdf

The SRM version is v7.2-1  Mar 6, 2000

Running even the most simple tests seems to basically lock up the system
since the command fails to ever exit even if you let it run for a couple
hours to try completing two passes.

    >>> memtest -rb -p 2

If you background the memtest process and run show_status, it seems to
pass at least once?

>>> memtest -rb -p 2 &
>>> show_status
ID        Program   Device   Pass   Hard/Soft   Written   Read
-------- --------- -------- ------ ----------- --------- ------
00000001      idle system       0      0    0         0      0 
0000004F   memtest memory       1      0    0         0      0 


Using >>>kill_diags afterwards only locks up the system.

I've searched around for more detailed instructions on the web. I found
a cryptic post to the DebianAlpha list
http://lists.debian.org/debian-alpha/2004/11/msg00064.html

It mentions using

>>>dynamic -r

to figure out values to use with memtest switches but I still don't
understand what was meant. The whole "zone" thing is a mystery. Worse
yet, the SRM Console user guide doesn't even mention "dynamic" as a
command and the man/help pages in the SRM itself are useless.

I've reduced the system memory to 128MB (two DIMS) so I can test the
pairs and by accident I figured out which pair is bad (i.e. running
"dynamic -h" by mistake resulted in errors with one pair).

When you guys use memtest properly, how do you do it?

Thanks,
JCR