AMD Refactors MCE Driver Code, Prepares For Future While Finally Adding DF3/Rome Support

AMD has published a set of patches refactoring their MCE kernel driver, making various machine check architecture (MCA) address translation updates in preparing for "future systems" while at the same time finally introducing Data Fabric 3 support for EPYC 7002 "Rome" processors and newer.

Published on Friday were a set of 25 patches to improve the AMD MCA address translation code within their machine-check exception driver. The patch cover letter notes that "the reference code was recently refactored in preparation for updates for future systems."

The vast majority of the patch work is refactoring of the existing code. The patch series doesn't introduce any of the new support for those "future systems" yet, but it does finally add address translation support for Data Fabric version 3 servers. Data Fabric version 3 is for EPYC 7002 "Rome" and later. That support is arguably long overdue with EPYC 7003 series having already debuted last quarter and Rome being out since 2019, but in any case it's now here in patch form and hopefully this future system support will come out timely and land presumably before next year's EPYC Genoa launch. With Data Fabric version 3 there are new memory interleaving modes now supported and a number of bit fields have been altered compared to prior generations, which need to be compensated for in the driver.

These AMD Linux patches are out for review on the kernel mailing list. It's too late for seeing in 5.13 but could be tidied up in time for 5.14 later this summer.