Backport AMD Venice MCA new bank type support to veLinux kernel 6.6

[mohanasv2]

Backport AMD Venice MCA new bank type support to veLinux kernel 6.6

Patch series

1. x86/mce/amd, EDAC/mce_amd: Move long names to decoder module
2. x86/MCE/AMD: Add new MA_LLC, USR_DP, and USR_CP bank types
3. x86/mce, EDAC/mce_amd: Reorder SMCA bank type enums
4. x86/mce, EDAC/mce_amd: Update CS bank type naming
5. x86/mce, EDAC/mce_amd: Add new SMCA bank types

Overall summary
This 5-patch series refactors and extends AMD SMCA bank type handling in the x86 MCE and EDAC subsystems. It moves long bank names into the decoder module, reorganizes SMCA bank type enums, and updates existing CS bank naming for better consistency with AMD documentation. The series also introduces support for new bank types such as MA_LLC, USR_DP, and USR_CP, along with additional SMCA bank definitions. Overall, the changes improve code maintainability, align the kernel with newer AMD hardware specifications, and enhance error decoding support for recent AMD processors.

Functional tests:
Kernel Configuration: The kernel was built with the following options enabled:
CONFIG_X86_MCE=y
CONFIG_X86_MCE_AMD=y
CONFIG_X86_MCE_INJECT=m
CONFIG_EDAC=y
CONFIG_EDAC_DECODE_MCE=m
CONFIG_EDAC_AMD64=m
CONFIG_AMD_ATL=m
CONFIG_AMD_NB=y

1. MCA / SMCA sanity

   • Verified CPU model and MCA bank count via /proc/cpuinfo and /sys/devices/system/machinecheck/machinecheck0/bank*.
   • Confirmed that the hardware exposes SMCA bank types that the unpatched OLK-6.6 kernel does not decode.

2. New SMCA bank type decoding

   • Used mce-inject to inject correctable errors into the newly added SMCA banks.
   • On the unpatched kernel, the relevant (HWID, McaType) pairs are missing from the SMCA table, so decode_smca_error() prints no bank name.
   • With the backported patches, the same injections produce decoded bank names and unit descriptions matching upstream for the new SMCA banks (including the die-to-die interconnect and new data cache controller units).

3. CS bank rename (user-visible ABI change)

   • Checked the sysfs MCA bank names:
     Before backport: coherent_slave_*
     After backport: coherent_station_*
   • Used mce-inject on the CS bank and confirmed that the decoded error text changes only in the unit name:
     Before: "Coherent Slave Ext. Error Code: ..."
     After: "Coherent Station Ext. Error Code: ..." while the underlying error information remains unchanged.

4. Regression checks for existing bank types

   • Injected errors into existing SMCA banks that are not renamed (e.g. UMC, LS).
   • Confirmed that the decoded dmesg output and EDAC logging are identical before and after the backport, despite the Enum reordering.

5. Long-name table validation

   • Verified that the long-name strings for the newly added SMCA bank types (die-to-die interconnect and DACC front-end/back-end units) are present in the built edac_mce_amd module and vmlinux image and used by the decoder.

6. Kernel stability

   • Booted kernels before and after the backport on the test platform.
   • Ran MCA injection tests and basic workload smoke tests.
   • Confirmed dmesg contains no BUG, Oops, panic, or WARNING entries related to MCE/EDAC on either kernel; only the usual Spectre v2 advisory about unprivileged eBPF appears on both builds.