STM32H7 hardfault from random point in code

[dinhlx]

Hi all.

I develop controller for PMSM motor and got a problem as below:

Some time MCU jump to hardfault handler. In debug, I see that it cause from random point (different from time to time), some time it cause from very basic function (such as sinf(), cosf(), delay..), as below

disassembly at <signal handler called>() at 0xffffffe9 as below:

"ffffffe9:   Failed to execute MI command:

          -data-disassemble -s 4294967273 -e 4294967413 -- 3

          Error message from debugger back end:

          Cannot access memory at address 0xffffffe8"

I also think it may cause from hardware because inverter may generate lot of noise. But whenmeasure noise in power, clock of MCU, it is clean: as below

When hardfault about to occur, the debug connection on cube monitor corrupt many time

So, where I can check for this issue?

Thanks

[FBL · Accepted solution]

Hello @dinhlx,

Could you check the fault analyzer window?

[SofLit]

Hello,

Since you're running a motor, with random points of hardfault, are you sure you're not facing any EMC issues?

[dinhlx]

Thank F.Belaid for reply.

I will check and revert to you when have result.

[dinhlx]

thank SofLit. I also worry about EMI/EMC, so I measure voltage to check and result as mention above. Voltage very clean. with this voltage, can I conclude that isn't EMC issue?

[SofLit]

Hello,

I'm not an expert of EMC/EMI topics but from what you have described and the randomness behavior, it looks like an EMI/EMC or HW issue. If you disconnect the motor, does the HF happen again?

For your power supply it can be some spikes not seen by a simple run.. Use scope triggers to detect them..

Check your grounds.. your connections your shielding, isolation etc..

You can also refer to this application note.

[Tesla DeLorean]

For a custom board make sure the VCAP pins and capacitors are correct and placed close to the MCU. Check the voltage.

Double check FLASH configuration, especially wait-states.

Faults probably coming from interrupt / callback code. Make sure you're not overwriting the stack via auto/local variables or arrays out-of-bounds.

The 0xFFFFFFE9 is a magic call-gate address used by the MCU to unstack context related to interrupts/handlers. Not going to see code there.