How to inject ECC errors for DDR5 RAM in MemTest86?

[maxsu]

Hi,

If my understanding is correct, the description of ECC injection in MemTest86 User Manual is that the memory errors will be generated by the system to simulate the real situation.
Is it correct? Shall I have the injection by hardware like your ECC Tester during running MemTest86 with ECC Injection enabled?

My platform si Alder Lake. I had enable ECC support in BIOS and ECC injection in MemTest86.

[David (PassMark)]

Yes, ECC injection simulates an ECC error to test the ECC reporting / monitoring mechanisms work as expected.
Not all systems support this however. And some support it, but have it hidden / disabled in BIOS as some vendors think the ability to simulate errors a security risk (we don't agree with this however).

We also have this hardware based ECC tester however for DDR4. This creates real RAM errors to verify that the system can detect and correct them.

More information about ECC can be found here.

[maxsu]

Hi David,

Thanks for your reply!

According to your Technical Information, it's locked or disabled by BIOS.
I enabled ECC options I got.
I also attached BIOS screenshots.
The problem we encountered is it only reports [ECC Inject] events only.
We CANNOT find [ECC Erros Detected] events.
Please refer to the logs attached.

Is anything I missed for ECC injection?

[David (PassMark)]

There are some unusual options that I haven't seen before in that BIOS.

For the log ECC injection and detection seem to be enabled.

We'll take a closer look. It might be a few days before we can get back to you however.

[keith]

Hi David,

Thanks for your reply!

According to your Technical Information, it's locked or disabled by BIOS.
I enabled ECC options I got.
I also attached BIOS screenshots.
The problem we encountered is it only reports [ECC Inject] events only.
We CANNOT find [ECC Erros Detected] events.
Please refer to the logs attached.

Is anything I missed for ECC injection?

Based on the logs, there is a possibility that the ECC injection has been locked in BIOS.
I've sent a DM with a debug build to collect additional information.

[maxsu]

Hi Keith,

I finished the test with the debug build. Please refer to the attachment.

[keith]

Thanks for the logs. There is a strong indication that the ECC injection has been disabled on this platform, as any attempt to trigger the injection in MemTest86 is ignored by the chipset.
The ECC injection options in the BIOS are intriguing though, as it implies that the ECC injection functionality is available.
I would consult the vendor for more details on how these BIOS options are used.

[maxsu]

Hi Keith,

We have another platform with ECC injection test. It reports a yellow PASS message. It also reports ECC errors were found and connected.
I'd checked the manaul. It means corrected ECC errors were detected. How to identify the corrected errors is real errors or a simulation in ECC injection test?

[David (PassMark)]

Yes, the Yellow pass message means that there were no errors that were not corrected.

The easy way to tell if you have real ECC corrections (as compared to corrections of the deliberately injected errors) is to turn off the ECC injection option.

[maxsu]

I ran MemTest86 ECC injection enabled and disabled with innodisk and Transcend DDR4 ECC SO-DIMMs.
For ECC injection disabled in MemTest86, the results of are the same. I have green PASS results.
For ECC injection enabled in MemTest86, I have different results. I have a green PASS result for Transcend SO-DIMM and a yellow PASS result for innodisk SO-DIMM.
Based on the results of innodisk SO-DIMM, the errors seem to be injected by MemTest86 and they are corrected and reported on the screen.
For Transcend SO-DIMM, the errors are injected and corrected? Why have different results from different SO-DIMMs?

[keith]

Thanks for the logs.

When ECC injection was enabled, the logs seem to indicate that the ECC injection functionality was disabled and locked by the BIOS. So it was surprising to see that ECC errors were being detected for the innodisk SO-DIMMS.

Also, it was strange to see that only ECC errors were detected for the innodisk SO-DIMMS during the first 3 tests. When ECC injection is enabled, it should perform the injection at the beginning of every test. So if the ECC errors were the result of ECC injection, I would expect to see more errors being detected in subsequent tests.

For the Transcend SO-DIMMs, there were no ECC errors detected so ECC injection didn't appear to work. This is why the result was a green "PASS".

I can see that the CPU/motherboard for these tests are the same, but is it on the same physical system? (ie. did you run the test on different machines or simply just swap the SO-DIMMS for the test)

Have you tested on other SO-DIMM models other than innodisk/Transcend? It would be curious to see if there is any other model other than the innodisk modules that triggers the ECC errors.

[maxsu]

Hi Keith,

Thanks for your reply!

I used the same machine and BIOS. I used the same SO-DIMM slot and just swapped SO-DIMM for the test.

I ran 5 more DDR4 SO-DIMMs with ECC injection enabled in MemTest86. I found the same result (Yellow PASS) at Transcend TS4GSH74V6E-I 32GB DDR4 2666 ECC SO-DIMM as innodisk SO-DIMM. Others reported the green PASS.

Please refer to the attachemnt for logs.

[keith]

Thanks for the logs.

It appears that only the 32GB modules resulted in ECC errors (ie. yellow "PASS"). For these modules, were there ECC errors detected (ie. green "PASS") if you disable ECC injection?

I have also sent a DM with a new build for testing.

[maxsu]

Hi Keith,

I ran the test build, 11.3.1013, with MemTest86 ECC injection disabled at 32GB module. The result is a yellwo PASS.
Without ECC injection, it still can correct the errors. We could say the correction function is workable.
Please refer to the logs of Transcend 32GB DIMM.

I'm still working on other memory modules for the logs. When it's done, I will attach the logs soon.

[maxsu]

Hi Keith,

I finished the tests with the rest of DDR4 memory modules. The results from the rest are green PASS.
Please refer to the attachment for logs.