Product Fault Management,
Continued
Each memory CRD entry represents one unique DRAM
within the memory subsystem. A unique set, bank, and
syndrome are stored in footprint to construct a unique ID for
the DRAM.
Rather than logging an error for each occurrence of a single
symbol correctable ECC memory error, the VMS error
handler maintains the CRD buffer; it creates a memory CRD
entry for new footprints and updates an existing memory
CRD entry for errors that occur within the range specified
by the ID in FOOTPRINT. This reduces the amount of data
logged overall without losing important information—errors
are logged per unique failure mode rather than on a per error
basis.
Each memory CRD entry consists of a FOOTPRINT,
STATUS, CRD CNT, PAGE MAPOUT CNT, FIRST EVENT,
LAST EVENT, LOWEST ADDRESS, and HIGHEST
ADDRESS.
FIRST EVENT, LAST EVENT, LOWEST ADDRESS and
HIGHEST ADDRESS are updated to show the range of time
and addresses of errors which have occurred for a DRAM.
CRD CNT is the total count per footprint. PAGE MAPOUT
CNT is the number of pages that have been marked bad for a
particular DRAM.
STATUS contains a record of the failure mode status of
a particular DRAM over time. This in turn determines
whether the CRD buffer is logged. For the first occurrence of
an error within a particular DRAM, the memory location is
scrubbed (corrected read data is read, then written back to
the memory location) and CRD CNT is set to 1. Since most
memory single-bit errors are transient due to alpha particles,
logging of the CRD buffer is not done immediately for the
first occurrence of an error within a DRAM. The CRD buffer,
however, is logged at the time of system shutdown (operator
or crash induced), or when a more severe memory subsystem
error occurs.
Continued on next page
5–67
