Product Fault Management,
Continued
Look for the following:
SCRUBBED
If SCRUBBED is the only bit set in the STATUS register,
memory modules should not generally be replaced.
The kernel performs memory scrubbing of DRAM memory
cells that may flip due to transient alpha particles. Scrubbing
reads the corrected data and writes it back to the memory
location.
HARD SINGLE ADDRESS
If the second occurrence of an error within a footprint is
at the same address (LOWEST ADDRESS = HIGHEST
ADDRESS then HARD SINGLE ADDRESS is set in STATUS
along with SCRUBBED. Scrubbing is not tried after the first
occurrence of any error within a particular footprint. The
page is marked bad by VMS.
Unlike uncorrectable ECC errors, the error handling code
cannot indicate if the page has been replaced. To make a
determination, use the DCL command SHOW MEMORY. If
the page mapout threshold has not been reached ("PAGE
MAPOUT THRESHOLD EXCEEDED" is not set in SYSTAT
packet header register), the system should be restarted
at a convenient time to allow the power-up self-test and
ROM-based diagnostics to map out these pages. This can
be done by entering TEST 0 at the console prompt, running
an extended script TEST A9, or by powering down then
powering up the system. In all cases, the diagnostic code
marks the page bad for hard single address errors, as well as
any uncorrectable ECC error by default.
If there are many locations affected by hard single-cell errors,
on the order of one or more pages for each megabyte of
system memory, the memory module should be replaced.
Use the console command SHOW MEMORY to indicate the
number of bad pages for each module. For example, if the
system contains 64 MB of main memory and there are 64 or
more bad pages, the affected memory should be replaced.
Continued on next page
5–77