System and method for memory fault resiliency in a server using multi-channel dynamic random access memory
View Patent ↗An on-package multi-channel dynamic random access memory stores data associated with write requests. An off-package memory stores a copy of the data associated with the write requests. A resiliency driver detects a write request, stores data for the write request to the on-package multi-channel dynamic random access memory as a primary image, stores a backup of the data for the write request to the off-package memory as a secondary image, detects a read request, and reads data for the read request from the on-package multi-channel dynamic random access memory as the primary image.
1. An information handling system comprising:
an on-package multi-channel dynamic random access memory to store data associated with write requests;
an off-package memory to store a copy of the data associated with the write requests; and
a processor core of a central processing unit to communicate with the on-package multi-channel dynamic random access memory and with the off-package memory, the processor core to execute a resiliency driver, the resiliency driver to detect a write request, to store data for the write request to the on-package multi-channel dynamic random access memory as a primary image, to store a backup of the data for the write request to the off-package memory as a secondary image, to detect a read request, and to read data for the read request from the on-package multi-channel dynamic random access memory as the primary image;
wherein the resiliency driver to determine whether a persistent memory read error has occurred on the on-package multi-channel dynamic random access memory while the data for the read request is being read from the on-package multi-channel dynamic random access memory, if the persistent memory read error has occurred, the resiliency driver to read the data for the read request from the off-package memory.
2. The information handling system of claim 1 , the resiliency driver to determine whether the data read from the off-package memory is correct, if the data read from the off-package memory is correct, the resiliency driver to perform a mirror scrub to write the correct data from the off-package memory to the on-package multi-channel dynamic random access memory.
3. The information handling system of claim 2 , wherein the correct data from the off-package memory is overwritten on corrupted data of the on-package multi-channel dynamic random access memory during the mirror scrub.
4. The information handling system of claim 2 , the resiliency driver to determine whether the mirror scrub is successful, if the mirror scrub is not successful, the resiliency driver to direct future read and write request only to the off-package memory.
5. The information handling system of claim 1 , wherein the off-package memory is a dual in-line memory module in communication with the processor core.
6. The information handling system of claim 1 , wherein the on-package multi-channel dynamic random access memory is an on-chip memory for the central processing unit.
7. A method comprising:
executing, by a processor of a central processing unit, a resiliency driver to perform mirroring between an on-package multi-channel dynamic random access memory and an off-package memory;
detecting, by the resiliency driver, a write request;
storing, by the resiliency driver, data for the write request to the on-package multi-channel dynamic random access memory as a primary image;
storing, by the resiliency driver, a backup of the data for the write request to the off-package memory as a secondary image;
detecting, by the resiliency driver, a read request;
reading, by the resiliency driver, data for the read request from the on-package multi-channel dynamic random access memory as the primary image;
determining, by the resiliency driver, whether a persistent memory read error has occurred on the on-package multi-channel dynamic random access memory while the data for the read request is being read from the on-package multi-channel dynamic random access memory; and
if the persistent memory read error has occurred, reading the data for the read request from the off-package memory.
8. The method of claim 7 , further comprising:
determining, by the resiliency driver, whether the data read from the off-package memory is correct; and
if the data read from the off-package memory is correct, performing a mirror scrub to write the correct data from the off-package memory to the on-package multi-channel dynamic random access memory.
9. The method of claim 8 , wherein the correct data from the off-package memory is overwritten on corrupted data of the on-package multi-channel dynamic random access memory during the mirror scrub.
10. The method of claim 8 , further comprising:
determining, by the resiliency driver, whether the mirror scrub is successful; and
if the mirror scrub is not successful, the resiliency driver to direct future read and write request only to the off-package memory.
11. The method of claim 7 , wherein the off-package memory is a dual in-line memory module in communication with the processor core.
12. The method of claim 7 , wherein the on-package multi-channel dynamic random access memory is an on-chip memory for the central processing unit.
13. A method comprising:
executing a resiliency driver to perform mirroring between an on-package multi-channel dynamic random access memory and an off-package memory;
detecting, by the resiliency driver, a write request including a tag identification;
determining, by the resiliency driver, whether the write request is to be mirrored based on the tag identification; and
in response to a determination that the write request is to be mirrored:
storing, by the resiliency driver, data for the write request to the on-package multi-channel dynamic random access memory as a primary image; and
storing, by the resiliency driver, a backup of the data for the write request to the off-package memory as a secondary image;
detecting, by the resiliency driver, a read request;
reading data for the read request from the on-package multi-channel dynamic random access memory as the primary image;
determining, by the resiliency driver, whether a persistent memory read error has occurred on the on-package multi-channel dynamic random access memory while the data for the read request is being read from the on-package multi-channel dynamic random access memory: and
if the persistent memory read error has occurred, reading the data for the read request from the off-package memory.
14. The method of claim 13 , further comprising:
determining, by the resiliency driver, whether the data read from the off-package memory is correct; and
if the data read from the off-package memory is correct, performing a mirror scrub to write the correct data from the off-package memory to the on-package multi-channel dynamic random access memory.
15. The method of claim 14 , further comprising:
determining, by the resiliency driver, whether the mirror scrub is successful; and
if the mirror scrub is not successful, the resiliency driver to direct future read and write request only to the off-package memory.
16. The method of claim 13 , wherein the off-package memory is a dual in-line memory module in communication with the processor core.
17. The method of claim 13 , wherein the on-package multi-channel dynamic random access memory is an on-chip memory for the central processing unit.