User`s guide

System Troubleshooting and Diagnostics
5.2 Product Fault Management and Symptom-Directed Diagnosis
• If the threshold has been exceeded for a particular type of cache error,
mark a flag that will signify that this resource is to be disabled (the cache
will be disabled in most, but not all, cases).
• Update the SYSTAT software register with results of error/fault handling.
• For memory uncorrectable Error Correction Code (ECC) errors:
If machine check, mark page bad and attempt to replace page.
Fill in MEMCON software register with memory configuration and
error status for use in FRU isolation.
• For memory single-bit correctable ECC errors:
Fill in Corrected Read Data (CRD) entry FOOTPRINT with set, bank,
and syndrome information for use in FRU isolation.
Update the CRD entry for time, address range, and count; fill the
MEMCON software register with memory configuration information.
Scrub memory location for first occurrence of error within a particular
footprint. If second or more occurrence within a footprint, mark page
bad in hopes that page will be replaced later. Disable soft error logging
for 10 minutes if threshold is exceeded.
Signify that CRD buffer be logged for the following events: system
shutdown (operator shutdown or crash), hard single-cell address within
footprint, multiple addresses within footprint, memory uncorrectable
ECC error, or CRD buffer full.
• For ownership memory correctable ECC error, scrub location.
• Log error.
• Crash process or system, dependent upon PSL (Current Mode) with a fatal
bugcheck for the following situations:
Retry is not possible.
Memory page could not be replaced for uncorrectable ECC memory
error.
Uncorrectable tag store ECC errors present in writeback cache.
Uncorrectable data store ECC errors present in writeback cache for
locations marked as OWNED.
Most INT60 errors.
Threshold is exceeded (except for cache errors).
System Troubleshooting and Diagnostics 5–5