TMS320F28377D: ECC test NMI triggers processor reset before the NMI watchdog counter expires

Gobind Singh

Intellectual 720 points

Part Number: TMS320F28377D

Hi,

I am trying to test the ECC logic and the code is something like below:

/* Double-bit error in Data */

FECC_CTRL.bit.ECC_TEST_EN = 0;

FADDR_TEST.all = <Address>;

FDATAH_TEST = <SomeData>

FDATAL_TEST = <SomeData with double bit error>

FECC_TEST.all = <ECC>;

FECC_CTRL.bit.ECC_SELECT = 0;

FECC_CTRL.bit.ECC_TEST_EN = 1;

FECC_CTRL.bit.DO_ECC_CALC = 1;

NMIFLGCLR.all = 0x0E7F; /* Clear NMI flags */

if(FECC_STATUS.bit.UNC_ERR != 1)

{

/* ECC logic did not work, error */

}

The above code works fine without generating processor reset. But if I move the instruction "NMIFLGCLR.all = 0x0E7F;" after the if condition, the processor keeps getting reset.

I want to first check if "NMIINT" flag has been set and then I want to clear the flag, not just clear the flag. But as soon as I do any small operation (even a simple assignment) after "DO_ECC_CALC = 1", the processor gets reset.

When I check NMIWDCNT at the time of reset, it is only either "0xAA" or 0xAB" which is not even close to NMIWDPRD that is set to 0xFFFF,

I am able to test the single bit errors in data and in ECC without any problem. The only problem is with Uncorrectable errors (double bit error in data/ECC or single bit error in address).

I am using CCS8.1.0.00011.

Thank you.

over 4 years ago

0 Vamsi Gudivada over 4 years ago

TI__Guru* 84185 points

Gobind,

Are you executing this code from RAM or Flash? Please confirm.

Thanks and regards,
Vamsi

0 Gobind Singh over 4 years ago in reply to Vamsi Gudivada

Intellectual 720 points

Hi Vamsi,

We are executing this code from RAM.

Thank you.

0 Vamsi Gudivada over 4 years ago in reply to Gobind Singh

TI__Guru* 84185 points

Gobind,

I asked our system control expert to help you further on this.

Thanks and regards,

Vamsi

0 Vamsi Gudivada over 4 years ago in reply to Vamsi Gudivada

TI__Guru* 84185 points

Gobind,

Do you have a NMI ISR defined in your application? If yes, NMI interrupt should take your application to the ISR indicating the NMIINT. You can clear the flags in the ISR.

On your original question, I asked our system control expert to help you.

Thanks and regards,
Vamsi

0 Vivek Singh over 4 years ago in reply to Vamsi Gudivada

TI__Guru** 109055 points

Hi,

When I check NMIWDCNT at the time of reset, it is only either "0xAA" or 0xAB" which is not even close to NMIWDPRD that is set to 0xFFFF,

That should not happen. How are you reading the NMIWDCNT at time of reset?

Also the correct method for this is to have the proper NMI handler and in handler, you should clear the flag. What code you have in NMI handler ?

Regards,

Vivek Singh

0 Gobind Singh over 4 years ago in reply to Vivek Singh

Intellectual 720 points

When I run it free, it resets and halts after it receives the NMI and I look at the Registers window to see the counter's value.

There is no NMI handler written. I wanted to make sure if I am getting the interrupt at the right time as mentioned in the Technical reference manual.

Thank you.

0 Vivek Singh over 4 years ago in reply to Gobind Singh

TI__Guru** 109055 points

Gobind,

There is no issue with NMI. After reset the counter will be cleared and then restart if there is another NMI. Are you manually halting the CPU or have a breakpoint? It'll be good to write the NMI handler and then check the counter value inside the handler.

Regards,

Vivek Singh

0 Gobind Singh over 4 years ago in reply to Vivek Singh

Intellectual 720 points

I am not manually halting the CPU. It receives NMI and halts automatically.

I can try with the NMI handler but I am just wondering why the below code 1) works and 2) does not. In 2), there is just one if condition before I clear the NMI flags and I am sure the NMI watchdog counter will not expire just by executing one if statement.

FECC_CTRL.bit.DO_ECC_CALC = 1;

NMIFLGCLR.all = 0x0E7F; /* Clear NMI flags */

if(FECC_STATUS.bit.UNC_ERR != 1)

{

/* ECC logic did not work, error */

}

FECC_CTRL.bit.DO_ECC_CALC = 1;

if(FECC_STATUS.bit.UNC_ERR != 1)

{

/* ECC logic did not work, error */

}

NMIFLGCLR.all = 0x0E7F; /* Clear NMI flags */

0 Vivek Singh over 4 years ago in reply to Gobind Singh

TI__Guru** 109055 points

Sorry for late reply.

In case 1, NMI flag may have got cleared before NMI gets generated hence you don't see issue. In such case even one instruction can make the difference. You need to follow correct procedure to handle the NMI.

Regards,

Vivek Singh

0 Gobind Singh over 4 years ago in reply to Vivek Singh

Intellectual 720 points

That is right. But the question is how can the whole of NMI WD counter starting from 0xFFFF expires in just one instruction?

The counter starts at:

FECC_CTRL.bit.DO_ECC_CALC = 1;

And I clear the flag just after executing a single if condition.

0 Vivek Singh over 4 years ago in reply to Gobind Singh

TI__Guru** 109055 points

It is not expiring in one instruction. One instruction may cause NMI to not trigger at all if get cleared hence you don't see reset. As mentioned multiple time, please have NMI handler and map is correctly.

Regards,

Vivek Singh

0 Gobind Singh over 4 years ago in reply to Vivek Singh

Intellectual 720 points

Hmm I think I got what you are saying now. I will check it with NMI handler and update if that works fine.

Thank you very much.

C2000™︎ microcontrollers

C2000 microcontrollers forum

TMS320F28377D: ECC test NMI triggers processor reset before the NMI watchdog counter expires