International Business Machines Corporation

Jay W Carman

Anshuman Khandual

Jyotindra Patel

Marc Avery Gollub

An application can specify reliability values via a communication path between the application and the registers. Application reliability could increase if the application itself could specify the timeout and retry values. For instance, some errors might be prevented if the timeout value is lengthened by a short amount. A longer timeout value would result in slower performance because the memory component could not be accessed during the timeout period. However, resolving errors in memory devices would prevent unrecoverable error indicators from being returned to the application, which would in turn limit application and system crashes. Creating a communication path between the application and the hardware registers would allow the application to modify the reliability of memory operations.

Application reliability and fault tolerant chip configurations

In some example embodiments, a method includes performing a memory scrub of a memory across a scrub cycle of multiple scrub cycles. The method includes identifying correctable errors of symbols in the memory that are a result of accesses from a section of the memory in response to the memory scrub. The method also includes performing an analysis across the multiple scrub cycles, wherein the performing of the analysis comprises determining whether at least two symbols across the multiple scrub cycles have at least one correctable error. The method includes responsive to determining that at least two symbols across the multiple scrub cycles have at least one correctable error, executing at least one repair of the memory that includes the section of memory.

Preemptive memory repair based on multi-symbol, multi-scrub cycle analysis

An apparatus includes a processor, a memory, and an error module operable on the processor. The error module is configured to perform a memory scrub of the memory across a scrub cycle of multiple scrub cycles. The error module is configured to identify correctable errors of symbols in the memory that are a result of accesses from a section of the memory in response to the memory scrub. The error module is configured to perform an analysis across the multiple scrub cycles, wherein the analysis comprises a determination whether at least two symbols across the multiple scrub cycles have at least one correctable error. The error module is configured to responsive to a determination that at least two symbols across the multiple scrub cycles have at least one correctable error, execute at least one repair of the memory that includes the section of memory.

Mechanisms are provided in which firmware verifies he entire system's memory scrub coverage through some additional memory controller (MC) registers/attentions and builds up a processor runtime diagnostic (PRD) scrub coverage table during every scrub cycle. Firmware may go through the scrub coverage table rank-by-rank on a periodic basis to determine whether any ranks had not been covered by hardware scrubbing. Firmware may initiate a targeted scrub and diagnostic for all of the ranks that did not have adequate scrub coverage. If for some reason the system still has some memory ranks that have not been covered by the initial hardware scrub and the targeted scrub, then the firmware may perform some course of action for fault isolation.

Growing community of inventors

Austin, TX, United States of America

Jay W Carman