TMS320F28377S: NMI interrupt while using VCU to compute CRC which reboots the CPU

Samy ELK

Prodigy 45 points

Part Number: TMS320F28377S

Hello,

We are facing a rare reboot of the DSP.

After investigations, it's related to a NMI when a flash uncorrectable error occurs on the address 0x701F8.

This address is in the OTP memory area, TI adds some errors to be able to test the ECC.

In our application, we don't run that test which leads to a NMI. In our .map file, we don't have that address neither.

I don't understand why this ECC test is executed on the 0x701F8 OTP address.

Could you please help us understanding this abnormal behavior and how to fix it.

Thank you in advance.

Kind regards,

Samy

over 3 years ago

0 Hareesh Janakiraman over 3 years ago

TI__Guru* 95315 points

Sami,

Please refer to this post: https://e2e.ti.com/support/microcontrollers/c2000-microcontrollers-group/c2000/f/c2000-microcontrollers-forum/514745/compute-crc-with-vcu-for-ti-opt-memory-causes-mcu-to-crash

0 Samy ELK over 3 years ago in reply to Hareesh Janakiraman

Prodigy 45 points

Hi Hareesh,

Thank you for your feedback.

I already checked that post before.

In our application, we don't check the MCU flash memory neither the OTP area.

The application works fine after boot-up, and sometimes the CPU reboots after a couple of seconds.

We are using the VCU CRC for a custom serial communication on a buffer in RAM memory where the size is fixed.

The crash does happen very rarely but we were able to catch the NMI interrupt, based on the stack, the CRC computation was ongoing just before.

Do you know if the ECC checks automatically the OTP area? if no, how this check can be executed?

Regards,

Samy

0 Vamsi Gudivada over 3 years ago in reply to Samy ELK

TI__Guru* 87155 points

Hi Samy,

SECDED (ECC) module does not read any memory range automatically - only a CPU read or fetch can bring the flash/ECC data to the SECDED logic.

Do you have debugger memory window open to the TI-OTP memory range? If yes, could you try closing it and see if the issue still occurs?

Thanks and regards,

Vamsi

0 Samy ELK over 3 years ago in reply to Vamsi Gudivada

Prodigy 45 points

Hi Vamsi,

Thank you for your feedback.

We don't have the memory window open at all.

Does the VCU CRC uses some data from the TI-OTP?

Thank you,

Regards,

Samy

0 Vamsi Gudivada over 3 years ago in reply to Samy ELK

TI__Guru* 87155 points

Hi Samy,

I don't think VCU CRC reads those locations of the TI-OTP automatically. Let me check with the team.

Thanks and regards,
Vamsi

0 Omer Amir over 3 years ago in reply to Samy ELK

TI__Mastermind 30185 points

Hello Samy,

This has been assigned to an expert for VCU CRC, but they will be out-of-office until Friday. Please expect a reply sometime after that.

Best regards,

Omer Amir

0 Sira Rao80 over 2 years ago in reply to Omer Amir

TI__Mastermind 28245 points

Samy,

So the summary is that

- application works fine for some time

- rarely, an NMI occurs that resets the CPU. During this time, a VCU CRC compute is occurring.

A few questions

- Every time the NMI occurs, is the VCU CRC compute occurring?

- Does the VCU CRC compute occur once or multiple times? Does it occur in an ISR or in the background?

- What data is used for the VCU CRC compute? RAM?

- If you don't do this VCU CRC compute, does this issue occur?

Thanks,

Sira

0 Sira Rao80 over 2 years ago in reply to Sira Rao80

TI__Mastermind 28245 points

An additional point of information - I checked the Silicon Errata for this device, and found something related to VCU and interrupts. Doesn't seem like this is related to your issue though.

https://www.ti.com/lit/er/sprz412m/sprz412m.pdf?ts=1694715283453&ref_url=https%253A%252F%252Fwww.google.com%252F

0 Samy ELK over 2 years ago in reply to Sira Rao80

Prodigy 45 points

Dear all,

We identified the root cause of the ECC flash uncorrectable error on TI-OTP that triggers the NMI.

It's a bug in a FW low-level driver, we have a table of pointers (read only operations) where the index was sometimes out of range.

It reads the next address of the table which contains a 32bits counter and not a pointer, when the value of the counter equals 0x701FB or closer, the NMI occurs when dereferencing this value.

We corrected this issue in our driver and everything works fine.

Thank you all for your time and your help.

Best regards,

Samy

C2000™︎ microcontrollers

C2000 microcontrollers forum

TMS320F28377S: NMI interrupt while using VCU to compute CRC which reboots the CPU