Author: Will Liang
Primary assignee: Will Liang
Created: 2019-02-26
The IPMI SELs only define memory Error Correction Code (ECC) errors for host memory rather than BMC.
The aim of this proposal is to record ECC events from the BMC in the IPMI System Event Log (SEL). Whenever ECC occurs, the BMC generates an event with the appropriate information and adds it to the SEL.
The IPMI specification defines memory system event log about ECC/other correctable or ECC/other uncorrectable and whether ECC/other correctable memory error logging limits are reached.[1]. The BMC ECC SEL will follow IPMI SEL format and creates BMC memory ECC event log.
OpenBMC currently support for generating SEL entries based on parsing the D-Bus event log. It does not yet support the BMC ECC SEL feature in OpenBMC project. Therefore, the memory ECC information will be registered to D-Bus and generate memory ECC SEL as well.
[1]Intelligent Platform Management Interface Specification v2.0 rev 1.1, section 41
Currently, the OpenBMC project does not support ECC event logs in D-Bus because there is no relevant ECC information in the OpenBMC D-Bus architecture. The new ECC D-Bus information will be added to the OpenBMC project and an ECC monitor service will be created to fetch the ECC count (ce_count/ue_count) from the EDAC driver. And make sure the EDAC driver must be loaded and ECC/other correctable or ECC/other uncorrectable counts need to be obtained from the EDAC driver.
ECC-enabled memory controllers can detect and correct errors in operating systems (such as certain versions of Linux, macOS, and Windows) that allow detection and correction of memory errors, which helps identify problems before they become catastrophic faulty memory module.
Many ECC memory systems use an "external" EDAC between the CPU and the memory to fix memory error. Most host integrate EDAC into the CPU's integrated memory controller.
According to Section 42.2 of the IPMI specification, Table 42 [2], these SEL sensor types will be defined as Memory and Event Data 3 field can be used to provide an event extension. Therefore, the BMC ECC event sets "Event Data 3" with the value FEh to identify the BMC ECC error.
[2] Intelligent Platform Management Interface Specification v2.0 rev 1.1, section 42.2
The main purpose of this function is to provide the BMC with the ability to record ECC error SELs.
There are two new applications for this design:
It also devised a mechanism to limit the "maximum number" of logs to avoid creating a large number of correctable ECC logs. When the maximum quantity is reached, the ECC service will stop to record the ECC log. The maximum quantity (default:100) is saved in the configuration file, and the user can modify the value if necessary.
This will always run the application and look up the ECC error count every second after service is started. On first start, it resets all correctable ECC counts and uncorrectable ECC counts in the EDAC driver.
It also provide the following path on D-Bus:
xyz.openbmc_project.Memory.ECC/xyz/openbmc_project/metrics/memory/BmcECCxyz.openbmc_project.Memory.MemoryECCThe interface with the following properties: | Property | Type | Description | | -------- | ---- | ----------- | | isLoggingLimitReached | bool | ECC logging reach limits| | ceCount| int64 | correctable ECC events | | ueCount| int64 | uncorrectable ECC events | | state| string | bmc ECC event state |
The error types for xyz::openbmc_project::Memory::Ecc::Error::ceCount and ueCount and isLoggingLimitReached will be created which generated the error type for the ECC logs.
Use the phosphor-sel-logger package to record the following logs in BMC SEL format.
ce_count from EDAC driver parameter and the count exceeds previous count.ue_count from EDAC driver parameter and the count exceeds previous count.maximum quantity.Another consideration is that there is no stopping the recording of the ECC logging mechanism. When the checks ce_count and value exceeds the previous value, it will record the ECC log. But this will encounter a lot of ECC logs, and BMC memory will also be occupied.
This application implementation only needs to make some changes when creating the event log, so it has minimal impact on the rest of the system.
Depending on the platform hardware design, this test requires an ECC driver to make fake ECC errors and then check the scenario is good.