Intel In-Field Scan "IFS" Poised For Linux 5.19 To Help Spot Faulty Silicon
Intel In-Field Scan is to be supported with upcoming CPUs (Xeon Saphire Rapids) and allows for initiating circuit-level tests on a CPU core for detecting hardware problems not caught by parity or ECC checks.
The Intel IFS Linux driver allows for initiating these silicon failure tests on a per-core basis and for reporting of any silicon failures from those tests. Intel In-Field Scan is intended for hyperscalers, cloud providers, and OEMs for testing and finding any "in-field" failures of Intel CPUs prior to deployment or routine testing as the silicon ages with time. Intel IFS is complementary to the like of ECC memory error reporting or machine check exceptions (MCEs).
Intel IFS is designed to help uncover silicon issues not otherwise easily identifiable from other error reporting infrastructure.
The Intel IFS Linux driver is the infrastructure for this while the actual tests themselves are loaded as a binary for a given CPU similar to Intel processor microcode. Those tests are stored within secure memory and authenticated prior to use.
Interfacing with the Intel IFS Linux driver is done within /sys/devices/system/cpu/ifs/.
Making this driver likely set for its premiere in Linux 5.19 is that as of yesterday the Intel IFS patches have queued into the platform-drivers-x86 "review-hans" branch that is for the patches being reviewed by subsystem maintainer Hans de Goede of Red Hat. Now that the patches are there, assuming no problems turn up with this new Intel driver code which shouldn't be the case, in the next few days will then be spun into pdx86's "next" branch ahead of the Linux 5.19 merge window opening later this month.
More details on the Intel In-Field Scan functionality designed for Sapphire Rapids and future Intel server CPUs can be found via this documentation commit.