AM2432: DDR ECC fault injection testing issues

Hello Jimmy,

Apologies for the delay here.

Jimmy said:

1. Performing error injection requires setting the ECC-IN bit, but TRM states that this bit must remain static before using DDR, which contradicts the need to configure the ECC-IN bit during the injection process. How should I periodically execute this diagnostic measure?

I am checking on this with the HW expert. But I think, it should be ok even if you change the ECC-IN bit just for the error injection. This is the standard procedure followed by all the customers. 

Jimmy said:

2. If an error is only injected during initialization and not corrected in the interrupt program, and triggered during periodic execution, we find that the interrupt can be entered normally when the error is triggered for the first time, but it will continue to enter the interrupt program after being triggered once. Is this phenomenon normal? How to trigger a one-time interrupt program without correction?

Can you explain a bit more. I did not understand the question properly.

Regards,

Nihar Potturu. 

Hi Potturu

Thanks for your reply

Regarding the second question, we have made such an attempt
Step 1: Inject ECC errors (not triggered) into DDR specific addresses during the initialization phase.
Step 2: Accessing DDR during the cycle run phase triggers this error.
Step 3: Do not correct errors in the interrupt callback function.

The phenomenon after our testing is as follows:
After the second step is triggered, the callback function is entered. After the callback function exits, it will immediately re-enter the callback function and cannot execute the normal program.

Jimmy

Hi Jimmy,

Jimmy said:

The phenomenon after our testing is as follows:
After the second step is triggered, the callback function is entered. After the callback function exits, it will immediately re-enter the callback function and cannot execute the normal program.

Is this being observed for double bit errors? 

Regards,

Nihar Potturu. 

Hi Potturu

This phenomenon occurs in both double bit and single bit scenarios。

Jimmy

Hello Jimmy,

Jimmy said:

After the second step is triggered, the callback function is entered. After the callback function exits, it will immediately re-enter the callback function and cannot execute the normal program.

Since, double bit errors are not correctable, the program getting stuck is expected. However, single bit errors are auto corrected, so you should not be seeing this issue if you are clearing the ECC error inside the callback function. Are you calling DDR_ClearECCError API inside the callback?(Attaching screenshot below)

Regards,

Nihar Potturu. 

Hi Jimmy,

  Feedback from internal teams (e.g. DDR-T7) is performed at launch and does not need to be performed on a regular basis.

Regards,

Linjun