M4 / M5 Servers: Corrected Memory Errors in Linux but Not BMC Logs

Problem and scope

A customer saw memory-related MCE records in Linux but no corresponding memory errors in BMC or black-box logs, raising the question of DIMM replacement. The source covers M4/M5 servers running Linux, including RHEL, CentOS, SLES and Ubuntu.

Why the records can differ

The vendor describes two corrected-ECC reporting paths on these systems: BIOS collects information and reports it to BMC SEL after a firmware error threshold triggers an SMI; the operating system can receive corrected-machine-check notifications through CMCI. Consequently, an OS record can appear before the firmware/BMC threshold is reached.

ECC (error-correcting code) can correct certain memory errors. A corrected-error record indicates that the reported error was corrected; the record itself does not establish that a DIMM must immediately be replaced.

Interpretation of the source advice

The source describes BIOS/BMC handling CPU, memory and PCIe faults, applying memory-error thresholds and filtering, and logging the affected memory location and corrected-error type when the firmware threshold is met. It contrasts that with the OS reporting path and recommends disregarding OS-only corrected-memory MCE entries for the described case.

That recommendation is limited to the corrected-error case. It must not be read as permission to ignore every MCE, uncorrectable errors or a recurring hardware fault. Review the error type, frequency, DIMM location, firmware/BMC records and applicable vendor replacement criteria.

The source’s statements that every corrected error generates an unfiltered OS record and that corrected errors can never affect performance or stability are not universal guarantees. Linux supports firmware CMCI thresholds and corrected-error handling policies; actual behavior depends on the CPU, firmware and kernel configuration.