Linux EDAC Memory Events Absent from BMC: Source M5/M6 Advisory

Reported symptom and scope

The source discusses M5/M6 servers running Linux, including RHEL, CentOS, SLES and Ubuntu, where an EDAC memory event appears in OS logs but no corresponding event is present in BMC SEL. Its example is:

Feb  4 21:00:06 localhost kernel: EDAC MC6: 1 CE memory read error on CPU_SrcID#3_MC#0_Chan#1_DIMM#0 (channel:1 slot:0 page:0x9a97c19 offset:0xac0 grain:32 syndrome:0x0 –  err_code:0101:0091 socket:3 imc:0 rank:1 bg:1 ba:1 row:6cb5 col:2f8)

EDAC means Error Detection And Correction; MC identifies a memory controller, CE a corrected error and UE an uncorrected error. A corrected event is not automatically a false event just because the BMC has no matching entry.

Source workaround: conditionally blacklist the relevant modules

The vendor source recommends preferring its BIOS/BMC reporting path on the affected configurations and disabling EDAC unless the customer requires it. This suppresses one OS reporting path; it does not repair a faulty DIMM. Confirm applicability with the platform/OS support procedure and retain diagnostic evidence before altering monitoring.

1. List the loaded EDAC modules:

lsmod | grep -i edac

2. The source edits the blacklist file:

vi /etc/modprobe.d/blacklist.conf

For its example modules, append:

blacklist skx_edac
blacklist edac_core

Use the modules actually present; do not blacklist unrelated hardware drivers by copying these names. 3. After the planned restart, repeat the list command to verify the intended module state. Depending on the distribution/initramfs and driver configuration, adding blacklist lines alone may not cover every loading path; follow the supported OS procedure.

Source explanation and qualification

The source attributes the missing BMC entry to platform error registers whose contents may be cleared after reading, with EDAC reading before BMC, and to firmware thresholds/filtering before SEL logging. It describes SMI-based collection and CMCI/MCE reporting for corrected memory events. These are attributed explanations for its platform case, not a universal specification of all CPU registers, BIOSes, BMCs or Linux RAS paths.

The original concludes that OS memory MCE records can be ignored. Do not generalize that recommendation to all memory errors: repeated corrected events, uncorrected/fatal events and symptoms still require diagnosis. Compare time, location, frequency, firmware and OS evidence. A quiet SEL or unloaded EDAC module is not proof that memory is healthy or an error was fixed.

English references: Linux kernel EDAC documentation; Linux kernel RAS documentation.